Qualification: 
Ph.D, M.Tech
v_sowmya@cb.amrita.edu

Dr. Sowmya V. currently serves as Assistant Professor at Amrita Center for Computational Engineering and Networking (CEN), Coimbatore Campus.

Professional Achievements

  • Received "Women In AI Leadership Awards 2019 (Only Academician among the winners)" sponsored by Jigsaw Academy, during the one day Conference, "The Rising 2019" on Women in Analytics and AI, organized by the Analytics India on March 8, 2019 at Taj Hotel, Bangalore.
  • Awarded with Deep Learning Instructor Ambassadorship Grant by NVIDIA (December 2018).

Professional Experience

  • Programmer Analyst Trainee at Cognizant Technology Solutions, Chennai (August 2010 -June 2011).
  • Assistant Professor, Center for Excellence in Computational Engineering and Networking, Amrita Vishwa Vidyapeetham, Coimbatore, (July 2011-till date).
  • Successfully defended her Ph. D. Thesis titled, "Significance of Incorporating Chrominance Information for Scene Classification" on July 2018 under the supervision of Dr. D. Govind, Assistant Professor (SG), CEN, and Dr. K. P. Soman, Professor & Head, CEN, Amrita School of Engineering, Coimbatore.
  • Promoted as Assistant Professor (Senior Grade) on November 2015.

Research Area

  • Color Image Processing
  • Hyperspectral Image Processing
  • Pattern Classification
  • Machine Learning
  • Deep Learning
  • Bio-Medical Signal Processing
  • Bio-Medical Image Processing
  • Image Analysis using Drones

Active Reviewer in the following SCI Indexed Journals

  1. ISPRS Journal of Photogrammetry and Remote Sensing (Elsevier).
  2. Computers and Electronics in Agriculture (Elsevier).
  3. Computer and Information Sciences (Elsevier).
  4. Neural Networks (Elsevier).
  5. Remote Sensing Letters (Taylor & Francis).
  6. Signal, Image and Video Processing (Springer).
  7. Digital Signal Processing (Elsevier).
  8. International Journal of Image and Data Fusion.

Invited Talks

  • Rendered a session on "Basic theory of Deep Neural and Convolutional Network" during "Two days workshop on Machine Learning - Hands on with Matlab and Python" , organized by Center for Development Advanced Computing (C-DAC), Trivandrum during 5-6 July, 2019.
  • Rendered a session on "Fundamentals of Computer Vision using NVIDIA DIGITS " (as a part of NVIDIA DLI University Ambassador Grant) during "CSIR sponsored National Level Seminar on Deep Learning" , organized by P.A.College of Engineering and Technology, Pollachi on June 26, 2019.
  • Rendered a session on "Fundamentals of Computer Vision using NVIDIA DIGITS " (as a part of NVIDIA DLI University Ambassador Grant) during "FDP on Deep Learning for Object Detection", organized by Sona College of Technology, Salem on June 21, 2019.

@ CDAC-Trivandrum

  • Rendered a session on "Generative Adversarial Networks (GAN)", along with the hands on in python at a National Level Faculty Development Program on "Deep Learning Unfolded", conducted by Amrita School of Engineering, Amritapuri Campus, Kollam on May 31, 2019.
  • Rendered a session on "Drones for Forestry Applications" at a monthly seminar conducted by the Institute of Forest Genetics and Tree Breeding (IFGTB), Indian Council of Forestry Research and Education, Coimbatore, India on May 30, 2019.
  • Delivered one day workshop on “Fundamentals of Deep Learning for Computer Vision – Hands-on using NVIDIA DIGITS” at KMEA College of Engineering, Aluva on May 14, 2019. This was certified and sponsored by NVIDIA as a part of “NVIDIA Deep Learning Instructor University Ambassadorship Award”.
  • Delivered a guest lecture on “Opportunities in Remote Sensing” at Avinashilingam Institute of Home Science and Higher Education for Women, Coimbatore on March 6, 2019.
  • Delivered one-day workshop on "Computational Tools Needed for Data Science (with hands-on in Matlab and Python)" during the 9th National Level Tech fest - Anokha 2019 organized by Amrita School of Engineering, Coimbatore during February 14-16, 2019.
  • Delivered a guest lecture on “Machine Learning” at Karpagam College of Engineering, Coimbatore on January 24, 2019.
  • Invited talk on "Deep Learning" in Two days IEEE workshop on Machine Learning held at Kalasalingam University on 2-3 Feb, 2018.
  • Invited talk on "Deep Learning for Bio-medical Application" in ICMR Sponsored Seminar on Deep Learning Techniqies and Tools for Medical Application organized by Department of Computer Science and Engineering, Mepco Schlenk Engineering College, Sivakasi on 17/01/2018.
  • Delivered a lecture on “Deep Learning for Medical Image Processing” in “TEQIP Sponsored Artificial Intelligence for Biomedical Applications” organized by TKM College of Engineering, Kollam, on 14th Dec, 2017.
  • Delivered a lecture on “Drone based Hyperspectral Imaging for Precision Agriculture” in “Refresher Course for Computer Science” organized by Bharathiar University, on 21st Nov, 2017.
  • Delivered a lecture on “Data Mining” for MBA students of Avinashilingam University on 18th March 2017.
  • Delivered a session on “Least square based image processing” as a part of short term training programme on Digital Signal Processing and its Applications held at Govt.Engineering College, Thrissur on 5th Dec, 2016.
  • Rendered hands on training in “Support Vector Machines using Libsvm and Weka” for M.Tech students of the Department of Electronics and Communication Engineering, Rajiv Gandhi Institute of Technology, Kottayam on 18-12-2015.
  • Delivered one day session on ‘PDE and Image Processing” in two days “National Level Workshop on Signal and Image Processing” conducted by Department of Information Technology, Sona College of technology, Salem during 4-5 Dec, 2015.
  • Rendered a lecture on “PDE based Image Processing” in “A training workshop on Advanced Optimization, Deep Learning Applications (AODA)” organized by organized by the Centre for Excellence in Computational Engineering and Networking (CEN), Amrita Vishwa Vidyapeetham (University), Coimbatore, during January 30 – February 2, 2014.
  • Rendered Invited Guest Lecture on “Remote Sensing and Applications of GIS” at Avinashilingam University, Coimbatore, 2012.
  • Rendered a talk on “Hyperspectral Image Processing” at “First National Workshop on “Sparse Image and Signal Processing (SISP-2011)”, organized by the Centre for Excellence in Computational Engineering and Networking (CEN), Amrita Vishwa Vidyapeetham (University), Coimbatore.

Awards

  • Awarded "Bharat Excellence" and "Best Indian Global Personalities"  by Friendship Forum of India at Delhi on July 28, 2019.
  • Awarded "Best Engineering College Teacher" by Society for Engineering Education Enrichment (SEEE) at Dr.N.G.P Institute of Technology, Coimbatore on July 20, 2019.
  • Awarded with "Excellence in Research for the academic year 2016-2017" by Amrita School of Engineering, Coimbatore, Amrita Vishwa Vidyapeetham on 23rd Institution day held on 10th January 2018.
  • Awarded with "Best Young Researcher Award" rendered by Integrated Intelligent Research Groups on the occasion of Republic Day Achievers award 2018, celebrated at Loyola-ICAM College of Engineering and Technology (LICET), Chennai.
  • Awarded with PG Merit Scholarship for University Rank Holders - Rs.40000 by University Grants Commission (UGC), New-Delhi in the year 2008.
  • Awarded with “Young Women Educator and Researcher” by National Foundation for Entrepreneurship Development (NFED), Coimbatore on 4th Women’s Day Awards celebrated on 8th March 2017 at Grand Reagent Coimbatore.
  • Awarded with “Young Women Achiever (in recognition of Your Role, Outstanding Contributions, Significant Achievements and Sustained Excellence in the field of Engineering) of the Women Awards - VIWA 2016” celebrated on 5 March 2016 at Radha Regent Chennai.
  • Received third prize in Essay Competition conducted by Amrita nature club on the occasion of International Women’s Day on March 8th, 2014.
  • Awarded with title “Associate of the month of January 2011” for project excellence by Cognizant Technology Solutions (CTS), Chennai.
  • Awarded with shield for securing University First Rank in M.TECH (Remote Sensing and Wireless Sensor Networks) (2008-2010) by Amrita Vishwa Vidyapeetham, Coimbatore.
  • Awarded with Gold Medal for University First Rank in B.Sc., (Physics) (2003-2006) by Avinashilingam University, Coimbatore.
  • Awarded with cash award of Rs.5000 for securing First Place in the paper presented in National Level Seminar on Signal Processing held at Sree NarayanaGuru Institute of Science and technology, Kochi.
  • Awarded with “Swami Vivekananda Award” for Excellence in Education by Yuva Kendra Association, Madurai.

Professional Activities

  • Actively Participated in the "Accenture Learning Symposium workshops on Deep Learning and DevOps", conducted on March 13-14, 2018 at Amrita School of Engineering, Coimbatore.
  • Successfully cleared the assessment test and completed a hands on workshop on "Artificial Intelligence and Deep Learning," held at Kongu Engineering College, Erode from July 28 - 30, 2018, conducted by leadingIndia.ai, a nation wide initiative by Bennett University, Greater Noida, India. (Listed as one of the toppers in the assessment)
  • Presented a paper tiled, "Inspiring stories from Indian Freedom Movement" in SADGAMAYA 5119, Cultural Camp organized by Amrita Vishwa Vidyapeetham, Coimbatore on 29-30 June 2017.
  • Event Coordinator - ‘Cook without Fire or Wire’ of Amrita Cultural Fest “Amritotsavam-2015’.
  • Participated in one day Seminar on “Projects in Signal & Image Processing, Communication, embedded, Robotics, Networks and VLSI”, organized by Department of Electronics and Communication Engineering, SNS College of Engineering, Coimbatore on 20th July, 2013.
  • Participated in Faculty Development programme on “Linear Algebra and Applications” organized by the department of Mathematics and Centre for Continuing Education held at National Institute of Technology, Calicut during 07-13 July 2013.
  • Participated in National Workshop on “Computer Vision and Image Processing Techniques” organized by Department of Computer Science and Engineering, Amrita Vishwa Vidyapeetham, Coimbatore on 15-16 March 2013.
  • Participated in two days workshop on “Geospatial Technologies for Coastal Resources Management”, organized by Department of Earth and Space Sciences, Indian Institute of Space Science and Technology (IIST), Thiruvananthapuram on 19-20 May 2012.
  • Participated in two days workshop on “Machine Vision and Image Processing using Labview”, organized by Department of Instrumentation and Control Systems Engineering, PSG College of Technology, in association with NI Systems (India) Pvt.Ltd, Bangalore on 18-19 May 2012.
  • Participated in “Second Edition of Amrita International Conference of Women in Computing” organized by Department of Computer Science and Engineering, Amrita Vishwa Vidyapeetham, Coimbatore on 9-11 January 2012.
  • Co-ordinator of the “First National Workshop on “Sparse Image and Signal Processing (SISP-2011)”, organized by the Centre for Excellence in Computational Engineering and Networking (CEN), Amrita Vishwa Vidyapeetham, Coimbatore during 23rd Dec-26th Dec,2011.

Presentations

  • Presented two papers entitled “Significance of Contrast and Structure Features for an Improved Color Image Classification System”, Improved Color Scene Classification System using Deep Belief Networks and Support vector Machines”, in 2017 International Conference on Signal and Image Processing, ICSIPA-2017, Kuching, Malaysia, 12-14 Sep.2017.
  • Presented a paper entitled ““Edge Detection Using Sparse Banded Filter Matrices” – Second International Symposium on Computer Vision and the Internet (VisionNet’15) held at SCMS School of Engineering, Aluva, Kochi on Aug 10-13, 2015. Publisher: Elsevier Procedia Computer Science Journal. (Published)
  • Presented a paper entitled “Role of Teachers in Nation Building” in the seminar for College Faculty on Swami Vivekananda’s thoughts in the modern context organized by Swami Vivekananda 150th Birth Anniversary Celebration Committee, Coimbatore Region in association with Hindusthan Arts and Science, Coimbatore on September 7, 2013.
  • Rendered a lecture on “Signal and Image Processing Application” on two days workshop on Sparse Image and Signal Processing-2013 organized by organized by the Centre for Excellence in Computational Engineering and Networking (CEN), Amrita Vishwa Vidyapeetham, Coimbatore.
  • Rendered a talk on “Hyperspectral Image Processing” at “First National Workshop on Sparse Image and Signal Processing (SISP-2011), organized by the Centre for Excellence in Computational Engineering and Networking (CEN), Amrita Vishwa Vidyapeetham, Coimbatore.
  • Presented a paper entitled “A Decision Tree Based Land Cover Image Classification Using Color Space and Texture” in 2011 IEEE International Conference on Computational Intelligence and Computing Research, at Cape Institute of Technology, Levengipuram, Kanyakumari, India.
  • Presented a paper entitled “An Effective pre-processing algorithm for detecting noisy spectral bands in hyperspectral imagery” in International Symposium on Ocean Electronics, SYMPOL 2011 at Cochin university of Science and Technology (CUSAT), Kochi.
  • Presented a paper entitled “A Decision Tree Based Land Cover Image Classification Using Color Space and Texture” in National Level Seminar on Signal Processing held at Sree Narayana Guru Institute of Science and technology, Kochi.
  • Rendered Invited Guest Lecture on “Remote Sensing and Applications of GIS” at Avinashilingam University, Coimbatore.

Achievements

  • Awarded with Cash Prize of Rs.2000 (each for a Semester) for securing First Rank in M. Tech. (I- IV Semester) Examination by Amrita Vishwa Vidyapeetham, Coimbatore.
  • Proficiency Holder in M.Sc (2006- 2008)
  • Short listed for the National level presentation in “Einstein’s Year of Physics - 2005” conducted by the members of “Indian Association of Physics-Mumbai”.
  • Published an article in Tamil in Amrita University Magazine, Amritadhwani 2013.

Social Activities

  • Active National Service Scheme (NSS) volunteer during the period 2003- 2006.
  • Actively participated in ten day Special Camping Programme organized at a village by Avinashilingam University Coimbatore.
  • Actively participated in three days Residential Youth Camp on Achieving Human Excellence organized by Ramakrishna Mission Vidhyalaya, Coimbatore.
  • Actively participated in Resources Mobilization for Leprosy eradication.
  • Actively served as volunteer on the occasion of the 59th and 60th birthday celebrations of Sri Mata Amnritanandamayi Devi during 26-27 September, 2012,2013.

Guided M. Tech. Dissertations

Year M. Tech. Dissertations Title
2019 Breast Cancer Classification Using Capsule Network with Pre-Processed Histology Images.
2019 Convolutional Neural Networks for Fingerprint Liveness Detection Systems.
2019 A Transferable Deep Learning Approach for the Classification of Cardiac Diseases Diagnosed Using Electrocardiogram and Phonocardiogram Signals.
2019  Convolutional Neural Networks for Placenta Cell Classification.
2019 Effect of Data Pre-Processing on Brain Tumor Classification Using CapsuleNet.
2019 Remote Sensing Image Super Resolution Using Residual Dense Network.
2019 Open Set Domain Adaptation for Hyperspectral Image Classification Using Generative Adversarial Network.
2018 Single Plane Scene Image Classification Using Deep Convolution Features
2018 Convolution and Recurrent Neural Networks for Disease Diagnosis
2018 Deep Learning Architecture for Land Cover Classification Using Red and Near-Infrared Satellite Images
2018 Dimensionality Reduction for Hyperspectral Image Classification Using Deep Learning and Kernel Methods
2018 Randomized Methods for Dimensionality Reduction of Hyperspectral Image Classification
2018 Effect of Denoising on Hyperspectral Image Classification Using Deep Networks and Kernel Methods
2018 Spatio-Spectral Compression and Analysis of Hyperspectral Images Using Tensor Decomposition Techniques
2018 Deep Rectified System for High-Speed Tracking in Images (DRSHTI)
2017 Color image Dehazing using Variational Mode Decomposition
2017 Scene classification using Deep Belief Network and Support Vector Machine
2017 Effect of denoising and dimensoinality reduction on Vectorized Convolutional Neural Network for Hyperspectral Image Classification
2017 Effect of Dynamic Mode Decomposition based dimension reduction technique on Hyperspectral Image Classification
2017 Dependency of various color and intensity planes on CNN based image classification
2017 Effect of denoising and Variational Mode Decomposition based dimensionality reduction on sparsity based hyperspectral unmixing
2017 Fusion of panchromatic image with low-resolution multi-spectral images using NIHS Transform and Decomposition techniques
2016 Impact of Denoising and Dimensionality Reduction Technique on Kernel Based Hyperspectral Image Classification
2016 Decomposition Techniques Applied for Pan Sharpening
2015 Impact of Various Denoising Techniques on Hyperspectral Image Analysis
2015 Hyperspectral Image Classification using  Kernel Methods
2015 Comparative Analysis of Variational Mode & Empirical Mode Features on Hyperspectral Image Classification
2015 Performance Enhancement of Minimum Volume based Hyperspectral Unmixing Algorithm by Empirical Wavelet Transform and Variational Mode Decomposition
2015 Image Fusion using Empirical Wavelet Transform
2015 Enhanced Features for Hyperspectral Image Classification
2014 Classification of Hyperspectral Image using Scattering Transform
2014 Image Classification using Deep Learning Features
2012 Spatial preprocessing for Improved Sparsity based Hyperspectral Image Classification

Publications

Publication Type: Conference Paper

Year of Publication Title

2019

V. V. Sajithvariyar, Sowmya V., Gopalakrishnan, E. A., Dr. Soman K. P., Sivanpillai, R., and Brown, G. K., “Opportunities and challenges of launching UAVs within wooded areas”, in ASPRS Annual Conference and International Lidar Mapping Forum 2019, Denver, United States, 2019.

2019

R. K. Renu, Sowmya V., and Dr. Soman K. P., “Spatio-Spectral Compression and Analysis of Hyperspectral Images using Tensor Decomposition”, in 2018 24th National Conference on Communications, NCC 2018, Hyderabad; India, 2019.[Abstract]


Hyperspectral images are large cubes of data which are commonly processed band-wise as two-dimensional image patches. This 2D processing might lead to loose the spectral efficiency contained in the image. Introducing Hyperspectral image as third-order tensors helps to preserve the spectral and spatial efficiency of the image. Multilinear Singular Value Decomposition (MLSVD) is an extension of Singular Value Decomposition (SVD) to 3D which can be used for compressing the image spatially and spectrally. The efficiency of compression is verified by reconstructing the image using Low Multilinear Rank Approximation (LMLRA). The proposed method has been validated with Signal to Noise Ratio (SNR), pixel reflectance spectrum and pixel-wise classification of the reconstructed image. © 2018 IEEE.

More »»

2019

V. V. Sajith Variyar, Dr. E. A. Gopalakrishnan, Sowmya V., and Dr. Soman K. P., “A complex network approach for plant growth analysis using images”, in Proceedings of the 2019 IEEE International Conference on Communication and Signal Processing, ICCSP 2019, Melmaruvathur; India, 2019.[Abstract]


The process of plant growth monitoring and analysis changed its perspective from the way it was. The recent farming practices demand vision based sensors for monitoring and analyses of the plant growth characteristics from images acquired by satellites and Unmanned Aerial Vehicles (UAV's). The advanced plant phenotyping systems are equipped with digital cameras to report the plant growth on a daily basis. The time determined images from plant monitoring system require a better computational representation to understand and study the plant life cycle, plant to plant interaction and correlations between plants with-in the community. This paper presents a new and yet simple approach towards plant growth analysis and its correlations in community by applying the theory of complex network on visible images from a plant phenotyping system. The method is highly promising in the area of precision agriculture when we have large area to monitor. © 2019 IEEE.

More »»

2019

M. A. Anupama, Sowmya V., and Dr. Soman K. P., “Breast cancer classification using capsule network with preprocessed histology images”, in Proceedings of the 2019 IEEE International Conference on Communication and Signal Processing, ICCSP 2019, Melmaruvathur; India, 2019.[Abstract]


Breast cancer is one of the most dangerous forms of cancer exists among women. The breast cancer is diagnosed using histology images. The purpose of this paper is to classify different types of breast cancer using histology images. The classification of histology images can be effectively done by image processing techniques. Among different image processing algorithms, deep learning gives the best performance for image classification applications. There are different convolutional neural network(CNN) architectures used for classification purpose such as AlexNet, Inception-Net, ResNet etc. Since conventional convolutional neural networks have lot of drawbacks, the architecture used for the current study is capsule networks, which captures the spatial and orientational information. The proposed work shows that the accuracy of Capsule Network model is improved due to the pre-processing of the histology images. © 2019 IEEE.

More »»

2018

A. Unnikrishnan, Sowmya V., and Dr. Soman K. P., “Deep alexnet with reduced number of trainable parameters for satellite image classification”, in Procedia Computer Science, 2018.[Abstract]


The emergence of machine learning algorithms enhanced the effectiveness of satellite image applications. On account of the high variability in-born in satellite data, a large portion of the present classification proposals are not sensible for handling satellite datasets. In this paper, we assessed profound learning systems in view of Convolutional Neural Networks for the precise order of multispectral remote sensing information. The Normalized Difference Vegetation Index (NDVI) is a solitary parameter for detecting landcover by utilizing the red and near-infrared band (NIR) information of the electromagnetic spectrum. NDVI is used to break down remote detecting images and survey the presence or absence of live green vegetation. The experiment is conducted on new publicaly available SAT-4 dataset, where the classes involved are types of vegetation. As NDVI computation require just two band data, it takes the benefit of both RED and NIR band information to classify diverse land covers. In the present work, the AlexNet architecture with two band information along with the reduced number of filters were trained and high-level features obtained from tested model managed to classify different land cover classes in the dataset. The proposed architecture is compared against the benchmark and results are estimated in terms of accuracy, precision and total number of trainable parameters. The proposed design brings about the aggregate diminishment of trainable parameters, while retaining high accuracy and precision, when compared against the existing architecture, which utilizes four bands. © 2018 The Authors. Published by Elsevier B.V.

More »»

2018

C. Jayaprakash, Damodaran, B. B., Sowmya V., and Dr. Soman K. P., “Dimensionality Reduction of Hyperspectral Images for Classification using Randomized Independent Component Analysis”, in 2018 5th International Conference on Signal Processing and Integrated Networks, SPIN 2018, Amity School of Engineering and Technology, Noida, India, 2018.[Abstract]


Independent Component Analysis (ICA) is a commonly used technique for the dimensionality reduction of Hyperspectral images (HSI) to capture the linear relationship in the original input features of the image. Even though kernel methods were introduced to capture the nonlinear features, they possess high computational complexity, while dealing with a large number of pixels of HSI. Recent research has introduced Random Fourier feature maps (RFF) to project high dimensional data to low dimension. In this paper, we propose a nonlinear component analysis for the dimensionality reduction of HSI based on RFF maps. The proposed method has experimented on two dataset namely Pavia University and Salinas scene. It is verified that the feature extracted using RFF maps outperforms the conventional and kernel methods, in terms of classification accuracy. © 2018 IEEE.

More »»

2018

V. S. Mohan, Sowmya V., and Dr. Soman K. P., “Deep Neural Networks as Feature Extractors for Classification of Vehicles in Aerial Imagery”, in 2018 5th International Conference on Signal Processing and Integrated Networks, SPIN 2018, 2018.[Abstract]


Detection of vehicles from aerial images have several real world implications in surveillance, military applications, traffic lot management, border patrol and traffic monitoring. The system proposed in this paper intends to automate the process of detecting vehicles from aerial images, rather than relying on a human operator. Here, we identify an optimum classification strategy for the proposed detection system, which is the initial stage of designing a vehicle detection pipeline. This research focuses on the feature extraction capabilities of standard neural network models like, Alexnet [6] and VGG-16 [7], which are compared against classic feature extraction techniques, like Histogram of Oriented Gradients and Singular Value Decomposition. The extracted features are benchmarked across standard machine learning algorithms such as Support Vector Machine and random forest. It is observed that the neural net extracted features gives an overall classification accuracy of 99% on the VEDAI dataset. The classification was treated as a binary class problem with vehicles as one class and rest everything as non-vehicles. © 2018 IEEE.

More »»

2018

N. Damodaran, Sowmya V., Govind, D., and Dr. Soman K. P., “Effect of decolorized images in scene classification using deep convolution features”, in Procedia Computer Science, 2018, vol. 143, pp. 954-961.[Abstract]


Scene classification is considered as an imperative issue for computer vision and has got extensive consideration in the recent past. Due to recent developments in high performance computing units such as GPUs, popularly known deep learning algorithm namely, convolutional neural networks (CNNs), exploits huge datasets to give powerful models. The paper proposes the use of transfer learning technique, by which a pre-trained model known as Places-CNN is used to generate feature vectors for each scene image of the dataset. The scene-classification experiments are conducted on the Oliva Torralba (OT) scene dataset, which consists of eight outdoor scene categories. The features were extracted from the fully connected layer of the pre-trained Places CNN architecture. The deep features were extracted from the input color images and the grayscale images converted using two different techniques based on singular value decomposition (SVD). The results obtained from classification experiments show that, models trained on SVD-Decolorized and Modified-SVD decolorized images give comparable performance to the input color images. Unlike the color images, which use three planes (RGB) of information, the grayscale images use only one plane of information. The grayscale images were able to retain the required shape and texture information from the original RGB images and, thus sufficient to categorize the classes of scene images. © 2018 The Authors. Published by Elsevier B.V.

More »»

2018

N. V. Jacob, Sowmya V., and Dr. Soman K. P., “A Comparative Analysis of Total Variation and Least Square Based Hyperspectral Image Denoising Methods”, in Proceedings of the 2018 IEEE International Conference on Communication and Signal Processing, ICCSP 2018, Adhiparasakthi Engineering CollegeMelmaruvathur, Chennai; India, 2018, pp. 58-63.[Abstract]


Hyperspectral image HSI with high spectral resolution will be always degraded by the noise accumulation. Therefore, image denoising is a fundamental preprocessing technique which improves the precision of successive processes like image classification, unmixing etc. In this paper, we compare least square LS weighted regularization in spectral domain with spatial least square and total variation TV denoising techniques. These methods are experimented on real, and noise simulated hyperspectral image datasets. The contrast and edges of the image are well preserved in the spectral LS. The image contrast varies in spatial LS, and edge informations are lost in TV. The experimental results show that, the spectral LS is superior to other two techniques in terms of visual interpretation, Signal-to-Noise Ratio SNR and Structural Similarity SSIM Index. © 2018 IEEE.

More »»

2017

S. Jose, Mohan, N., Sowmya V., and Dr. Soman K. P., “Least square based image deblurring”, in 2017 International Conference on Advances in Computing, Communications and Informatics, ICACCI 2017, Manipal, Mangalore, India, 2017.[Abstract]


An image can be basically defined as an object that represents visual observation, which can be created and stored in the electronic form, produced from an optical device. When we take a photograph, there can be many problems associated with that particular image. Among them, one of the main issue is the blur of the image. Blur can be defined as something which will become vague or less distinct. A blurred image looks sharper or more detailed, if we are able to perceive all the objects and their shapes correctly in it. The main cause for blur is the out of focus issue of the camera/sensor. An image which is in out of focus will appear in a blurred state. Even if, at the present time, with an auto focus facility, sometimes we will not get the image in the correct focus. Most probably, a part of the image will be crisp and clear, however rest will be ill-defined. Image deblurring is a common and important process in fields like digital photography, medical imaging and astronomy. Hence, removing or dropping the total amount of blur is the most important task before being applying to the image analysis techniques. In this paper, a colour image deblurring algorithm based on the concept of least squares is proposed. The 1D least square based deconvolution technique is extended to colour image deblurring. The proposed approach is experimented on standard test images and the results are compared with classical total variation image deblurring algorithm. The effectiveness of the proposed approach is evaluated in terms of standard quality metrics such as PSNR and SSIM. © 2017 IEEE.

More »»

2017

V. S. Dev, Rajan, S., Sowmya V., and Dr. Soman K. P., “Hyperspectral image denoising: A least square approach using wavelet filters”, in 2017 International Conference on Advances in Computing, Communications and Informatics, ICACCI 2017, Manipal, Mangalore, India, 2017.[Abstract]


An image is an artifact that depicts visual perception, having a similar appearance to an object or person, thus providing a depiction of it. Images accomodate various types of noises which are due to sensor defects, lens distortion, software artifacts, blur etc. Denoising an image not only aims at removing the undesired noise but also, at retaining the features of the original image. The same goes for hyperspectral images which have numerous bands (each band contains information of the same object or location taken under different wavelengths of light) as compared to the red, green and blue bands of a color image. The need for better denoising techniques have brought about the birth of different image denoising algorithms, each with its own unique characteristics. Total Variation Denoising (TVD) is an advent for noise removal developed so as to retain sharp edges in the underlying signal. It is characterised as an optimization problem. Denoising using Legendre-Fenchel transform also widely used. Least Square based denoising technique is computationally demanding and gives better results. This paper compares the efficiency of various image denoising techniques like, total variation denoising, legendre-fenchel transform and wavelet transform denoising with the proposed method of least square denoising. This paper focuses on the hyperspectral image denoising technique based on least square approach using different wavelet filters. The proposed technique gives satisfactory denoising output with less computational time when compared with existing methods. © 2017 IEEE.

More »»

2017

Sowmya V., Ajay, A., Dr. Govind D., and Dr. Soman K. P., “Improved color scene classification system using deep belief networks and support vector machines”, in 2017 IEEE International Conference on Signal and Image Processing Applications (ICSIPA), Kuching, Sarawak; Malaysia, 2017, pp. 12-14 .[Abstract]


In general, the three main modules of the color scene classification systems are image decolorization, feature extraction and classification. The work presented in this paper focuses on image decolorization and classification as two stages. The first stage or objective of this paper is to improve the performance of the color scene classification system using deep belief networks (DBN) and support vector machines (SVM). Therefore, color scene classification system termed as AGMM-DBN-SVM is proposed using the existing feature extraction technique called bags of visual words (BoW) derived from the dense scale-invariant feature transform (SIFT) and adapted gaussian mixture models (AGMM). The second stage of the presented work is to combine the proposed AGMM-DBN-SVM classification models obtained for the two different image decolorization methods called rgb2gray and singular value decomposition (SVD) based color-to-grayscale image mapping techniques to significantly increase the performance of the proposed color scene classification system. The effectiveness of the proposed framework is experimented on Oliva Torralba (OT) scene dataset containing 8 different classes. The classification rate of the proposed color scene classification system applied on OT 8 scene dataset is significantly greater than the one of the existing benchmarks color scene classification system developed using AGMM and SVM.

More »»

2016

S. Srivatsa, Ajay, A., Chandni, C. K., Sowmya V., and Soman, K. P., “Application of Least Square Denoising to Improve ADMM Based Hyperspectral Image Classification”, in 6th International Conference on Advances in Computing and Communications , ICACC-2016, Rajagiri School of Engineering & Technology, 6-8 September 2016, 2016, vol. 93, pp. 416-423.[Abstract]


Hyperspectral images contain a huge amount of spatial and spectral information so that, almost any type of Earth feature can be discriminated from any other feature. But, for this classification to be possible, it is to be ensured that there is as less noise as possible in the captured data. Unfortunately, noise is unavoidable in nature and most hyperspectral images need denoising before they can be processed for classification work. In this paper, we are presenting a new approach for denoising hyperspectral images based on Least Square Regularization. Then, the hyperspectral data is classified using Basis Pursuit classifier, a constrained L1 minimization problem. To improve the time requirement for classification, Alternating Direction Method of Multipliers (ADMM) solver is used instead of CVX (convex optimization) solver. The method proposed is compared with other existing denoising methods such as Legendre-Fenchel (LF), Wavelet thresholding and Total Variation (TV). It is observed that the proposed Least Square (LS) denoising method improves classification accuracy much better than other existing denoising techniques. Even with fewer training sets, the proposed denoising technique yields better classification accuracy, thus proving least square denoising to be a powerful denoising technique. © 2016 The Authors. Published by Elsevier B.V.

More »»

2016

R. Reshma, Sowmya V., and Soman, K. P., “Dimensionality Reduction Using Band Selection Technique for Kernel Based Hyperspectral Image Classification”, in 6th International Conference on Advances in Computing and Communications , ICACC-2016, Rajagiri School of Engineering & Technology, 2016, vol. 93, pp. 396-402.[Abstract]


Hyperspectral images have abundant of information stored in the various spectral bands ranging from visible to infrared region in the electromagnetic spectrum. High data volume of these images have to be reduced, preserving the original information, to ensure efficient processing. In this paper, dimensionality reduction is done on Indian Pines and Salinas-A datasets using inter band block correlation coefficient technique followed by Singular Value Decomposition (SVD) and QR decomposition. The dimensionally reduced images are classified using GURLS and LibSVM. Classification accuracies of the original image is compared to that of the dimensionally reduced image. The experimental analysis shows that, for 10% training sample the overall accuracy, average accuracy and kappa coefficient of the dimensionally reduced image (about 50% of the dimension is reduced) is i)83.52%, 77.18%, 0.8110 for Indian Pines and ii)99.53%, 99.40%, 0.9941 for Salinas-A dataset which is comparable to that of original image i)84.67%, 82.28%, 0.8247 for Indian Pines and ii)99.32%, 99.18%, 0.9916 for Salinas-A dataset. © 2016 The Authors. Published by Elsevier B.V.

More »»

2016

M. P, M, S., Sowmya V., and Dr. Soman K. P., “Low Contrast Satellite Image Restoration based on adaptive Histogram Equalization and Discrete Wavelet Transform”, in - 5th IEEE International Conference on Communication and Signal Processing-ICCSP'15, Adhiparasakthi Engineering College, Melmaruvathur , 2016.[Abstract]


Normally images obtained from satellites are of low-contrast type which hides major information carried by the image. Hence, image restoration is necessary in the image processing domain to extract all the information present in the images. The low contrast satellite image restoration based on adaptive histogram equalization combined with Discrete Cosine Transform (DCT) and Discrete Wavelet Transform (DWT) is proposed in this paper. The proposed technique is experimented on three different satellite images. The effectiveness of the method introduced in this paper is shown by comparing it against the existing techniques based on gamma correction and histogram equalization combined with DCT and DWT. The comparison is done based on the standard parameters called Peak Signal to Noise Ratio (PSNR) and Standard Deviation. The result and analysis on the basis of PSNR values shows that adaptive histogram equalization combined with DWT is more effective approach compared to adaptive histogram equalization combined with DCT.

More »»

2016

P. A, R, M., Sowmya V., and Dr. Soman K. P., “X-ray Image Classification Based On Tumor using GURLS and LIBSVM”, in International Conference on Communications and Signal Processing (ICCSP’16), Adhiparasakthi Engineering College, Melmaruvathur , 2016.[Abstract]


In today's world, X-ray imaging is the low cost diagnostic technique when compared with all other medical imaging techniques. In this paper, the proposed method is to classify X-ray images based on tumor. The features are extracted using Singular Value Decomposition (SVD) and classified using different kernels in Library for Support Vector Machine (Lib-SVM) and Grand Unified Regularized Least Squares (GURLS). The proposed method is experimented on X-ray image dataset which is approved by an Oncologist. The effectiveness of proposed method is validated based on classification parameters. The experiment result analysis shows that Gaussian-ho in GURLS provides 95% classification accuracy which is 5% higher than RBF kernel in LibSVM. The performance of the proposed system is validated by an Oncologist.

More »»

2015

A. Joy, Merlin, D., .K, D., Sowmya V., and , “Aerial Image Classification using GURLS and LIBSVM”, in 5th IEEE International Conference on Communication and Signal Processing-ICCSP'15, Adhiparasakthi Engineering College, Melmaruvathur , 2015.[Abstract]


Image classification using kernels have very great importance in remote sensing data. The goal of this work is to efficiently classify the large set of aerial images into different classes. This paper introduces a kernel based classification for aerial images. It uses Grand Unified Regularized Least Square (GURLS) and library for support vector machines (LIBSVM). This paper compares the performance of different kernel methods used in GURLS and LIBSVM. The experiment is performed on three sets of aerial image data sets which are obtained from electrical engineering department of Banja Luka University under the DSP laboratory, funded by the WUSAUSTRIA project of the European Union. From the experiment performed, it can be deduced that GURLS library is better compared to LIBSVM in terms of its prediction accuracy. The advantage of GURLS library package over LIBSVM is its automatic parameter selection.

More »»

2014

S. Santhosh, Abinaya, N., Rashmi, G., Sowmya V., and Dr. Soman K. P., “A novel approach for denoising coloured remote sensing image using Legendre Fenchel Transformation”, in 2014 International Conference on Recent Trends in Information Technology, ICRTIT 2014, Chennai, India, 2014.[Abstract]


Data acquired from remote sensing satellites are processed in order to retrieve the information from an image. Those images are preprocessed using image processing techniques such as noise removal. Satellite images are assumed to be corrupted with white Gaussian noise of zero mean and constant variance. Three planes of the noisy image are denoised separately through Legendre Fenchel Transformation. Later, these three planes are concatenated and compared with results obtained by Euler-Lagrange ROF model. Simulation results show that Legendre Fenchel ROF is highly convergent and less time consuming. To add evidence to the outcomes, quality metrics such as variance and PSNR for noisy and denoised images are calculated. The qualitative analysis of an image is analysed using MSSIM calculations, which clarifies the Structural Similarity between denoised images with original image. © 2014 IEEE.

More »»

2014

Sowmya V., “Image Classification using Deep Learning Features”, in International Conference on Emerging Trends in Electrical Engineering-ICETREE , Thangal Kunju Musaliar College of Engineering , 2014.

Publication Type: Journal Article

Year of Publication Title

2019

A. Simon, Vinayakumar, R., Sowmya V., and Dr. Soman K. P., “Shallow Cnn with Lstm layer for tuberculosis detection in microscopic image”, International Journal of Recent Technology and Engineering, vol. 7, pp. 56-60, 2019.[Abstract]


Tuberculosis or TB, a disease mainly affecting lungs is infected by bacterium mycobacterium tuberculosis and diagnosed by careful examination of microscopic images taken from sputum specimen. Diagnosis of disease using microscopy and computer vision methods are applied for many previous practical problems. Recently, deep learning is playing major role in computer vision applications producing remarkable performance. But, computational complexity always remains as an obstacle in the application of deep learning in many aspects. So in this paper, a shallow CNN with LSTM layer is used for detecting the tubercle bacillus, mycobacterium tuberculosis, from the microscopic images of the specimen collected from the patients. The specified model is producing better performance than state of the art model and also have reduced number of learnable parameters, which requires comparatively less computation than the existing model. © BEIESP.

More »»

2019

N. Damodaran, Sowmya V., Govind, D., and Dr. Soman K. P., “Scene classification using transfer learning”, Studies in Computational Intelligence, vol. 804, pp. 363-399, 2019.[Abstract]


Categorization of scene images is considered as a challenging prospect due to the fact that different classes of scene images often share similar image statistics. This chapter presents a transfer learning based approach for scene classification. A pre-trained Convolutional Neural Network (CNN) is used as a feature extractor for the images. The pre-trained network along with classifiers such as Support Vector Machines (SVM) or Multi Layer Perceptron (MLP) are used to classify the images. Also, the effect of single plane images such as, RGB2Gray, SVD Decolorized and Modified SVD decolorized images are analysed based on classification accuracy, class-wise precision, recall, F1-score and equal error rate (EER). The classification experiment for SVM was also done using a dimensionality reduction technique known as principal component analysis (PCA) on the feature vector. By comparing the results of models trained on RGB images with those grayscale images, the difference in the results is very small. These grayscale images were capable of retaining the required shape and texture information from the original RGB images and were also sufficient to categorize the classes of the given scene images. © Springer Nature Switzerland AG 2019.

More »»

2019

N. Damodaran, Sowmya V., Govind, D., and Dr. Soman K. P., “Single-plane scene classification using deep convolution features”, Advances in Intelligent Systems and Computing, vol. 900, pp. 743-752, 2019.[Abstract]


Scene classification is considered as one of the challenging tasks of computer vision. Due to the availability of powerful graphics processing unit and millions of images, deep learning techniques such as convolutional neural networks (CNNs) have become popular in the image classification. This paper proposes the use of a pre-trained CNN model known as Places CNN, which was trained on scene-centric images. In this work, the pre-trained CNN is used as a feature extractor. The features are then used as input data for support vector machines (SVMs). The effect of grayscale images on the performance of pre-trained CNN-based scene classification system is analyzed by means of classification accuracy and equal error rate. The dataset used for this purpose is Oliva Torralba (OT) scene dataset, which consists of eight classes. The classification experiments are conducted using the feature vector from the ‘fc7’ layer of the CNN model for RGB, RGB2Gray, and SVD decolorized images. The classification experiment was also done using a dimensionality reduction technique known as principal component analysis (PCA) on the feature vector. The results obtained from classification experiments show that RGB2Gray and SVD decolorized images were able to give results similar to that of RGB images. The grayscale images were able to retain the required shape and texture information from the original RGB images and were also sufficient to categorize the classes of scene images. © Springer Nature Singapore Pte Ltd. 2019.

More »»

2019

V. S. Mohan, Vinayakumar, R., Sowmya V., and Dr. Soman K. P., “Deep rectified system for high-speed tracking in images”, Journal of Intelligent and Fuzzy Systems, vol. 36, pp. 1957-1965, 2019.[Abstract]


Deep Rectified System for High-speed Tracking in Images (DRSHTI) is a unified open-source web portal developed for object detection in images. It aims to be a platform for the end user, where he/she can perform object detection on images without going through the hassles of debugging countless lines of code or setting up the right environment to perform computer vision tasks. By making the platform open-source, this work targets beginners in computer vision to form a basic understanding of object detection as an artificial intelligence task. This is made possible by releasing source codes, tools and tutorials on its usage via GitHub. This open-source portal offers two detection pipelines based on Faster-RCNN - a model to detect ground vehicles in aerial images and a model to detect everyday objects in 37 different classes in normal images. The former model is trained on VEDAI dataset, which gave 98.6% accuracy during testing and is offered as proof-of-concept that showcases the models ability to perform small target detections, but the latter model is trained on the PASCAL VOC dataset. Making the project open-source also aims at bringing in more development and tweaking to the existing vehicle detection module. The web portal can be accessed via https://drshti.github.io, where user can upload images and get annotations on objects present in it. Tutorials and source codes can be found at https://github.com/vyzboy92/Object-Detection-Net. © 2019 - IOS Press and the authors.

More »»

2019

K. S. Charmisha, Sowmya V., and Dr. Soman K. P., “Dimensionally reduced features for hyperspectral image classification using deep learning”, Lecture Notes in Electrical Engineering, vol. 500, pp. 171-179, 2019.[Abstract]


Hyperspectral images (HSIs) cover a wide range of spectral bands in the electromagnetic spectrum with a very finite interval, and with high spectral resolution of data. The main challenges encountered with HSIs are those associated with their large dimensions. To overcome these challenges we need a healthy classification technique, and we need to be able to extract required features. This chapter analyzes the effect of dimensionality reduction on vectorized convolution neural networks (VCNNs) for HSI classification. A VCNN is a recently introduced deep-learning architecture for HSI classification. To analyze the effect of dimensionality reduction (DR) on VCNN, the network is trained with dimensionally reduced hyperspectral data. The network is tuned in accordance with the learning rate and number of iterations. The effect of a VCNN is analyzed by computing overall accuracy, classification accuracy, and the total number of trainable parameters required before and after DR. The reduction technique used is dynamic mode decomposition (DMD), which is capable of selecting most informative bands using the concept of eigenvalues. Through this DR technique for HSI classification using a VCNN, comparable classification accuracy is obtained using the reduced feature dimension and a lesser number of VCNN trainable parameters. © Springer Nature Singapore Pte Ltd. 2019.

More »»

2019

A. Unnikrishnan, Sowmya V., and Dr. Soman K. P., “Deep learning architectures for land cover classification using red and near-infrared satellite images”, Multimedia Tools and Applications, 2019.[Abstract]


Classification of remotely sensed data requires the modelling of suitable image processing algorithms. The rise of machine learning systems upgraded the viability of satellite image applications. Using Convolutional Neural Networks (CNN), benchmark classification exactness can be accomplished for land cover grouping. Motivated by the concept of Normalized Difference Vegetation Index (NDVI), this paper utilizes only the red and near infrared (NIR) band information for classifying the publicly available SAT-4 and SAT-6 datasets. This is done, since NDVI computation requires only the two band (red and NIR) information and the classes involved in the dataset are types of vegetation. In this work, new deep learning architectures for three different networks (AlexNet, ConvNet, VGG) were proposed by hypertuning the network and the input as two band data. The modified architectures with the two band information along with reduced number of filters were trained and tested model manages to classify the images into different classes. The proposed architectures are compared against the existing architectures in terms of accuracy, precision and trainable parameters. The proposed architecture is found to perform equally efficient by retaining high accuracy with less number of trainable parameters, when compared against the the performance of benchmark deep learning architectures for satellite image classification. © 2019, Springer Science+Business Media, LLC, part of Springer Nature.

More »»

2019

N. V. Jacob, Sowmya V., and Dr. Soman K. P., “Effect of denoising on hyperspectral image classification using deep networks and kernel methods”, Journal of Intelligent and Fuzzy Systems, vol. 36, pp. 2067-2073, 2019.[Abstract]


Hyperspectral Image (HSI) store the reflectance values of a single scene or object in several continuous bands of electromagnetic spectrum. When the image is recorded, the information in some of the spectral bands gets mixed with noise. The classification accuracy of hyperspectral image varies inversely with the quantity and nature of noise present in the cluster of spectral bands. Thus, denoising is a fundamental prerequisite in image processing applications like classification, unmixing, etc. In this paper, we compare the effect of denoising via classification using Vectorized Convolutional Neural Network (VCNN), kernel based Support Vector Machine (SVM) and Grand Unified Regularized Least Squares (GURLS) classifiers. The classifiers are provided with raw data (without denoising) and denoised data using spectral and spatial Least Square (LS) techniques. The data given to the network are in the form of pixels, so we call the convolutional neural network (CNN) as VCNN. The experiments are performed on three standard HSI datasets. The performance of the classifiers are evaluated based on overall and class-wise accuracy. © 2019 - IOS Press and the authors.

More »»

2019

Sowmya V., Dr. Soman K. P., and Hassaballah, M., “Hyperspectral image: Fundamentals and advances”, Studies in Computational Intelligence, vol. 804, pp. 401-424, 2019.[Abstract]


Hyperspectral remote sensing has received considerable interest in recent years for a variety of industrial applications including urban mapping, precision agriculture, environmental monitoring, and military surveillance as well as computer vision applications. It can capture hyperspectral image (HSI) with a lager number of land-cover information. With the increasing industrial demand in using HSI, there is a must for more efficient and effective methods and data analysis techniques that can deal with the vast data volume of hyperspectral imagery. The main goal of this chapter is to provide the overview of fundamentals and advances in hyperspectral images. The hyperspectral image enhancement, denoising and restoration, classical classification techniques and the most recently popular classification algorithm are discussed with more details. Besides, the standard hyperspectral datasets used for the research purposes are covered in this chapter. © Springer Nature Switzerland AG 2019.

More »»

2019

R. K. Renu, Sowmya V., and Dr. Soman K. P., “Pre-processed hyperspectral image analysis using tensor decomposition techniques”, Communications in Computer and Information Science, vol. 968, pp. 205-216, 2019.[Abstract]


Hyperspectral remote sensing image analysis has always been a challenging task and hence there are several techniques employed for exploring the images. Recent approaches include visualizing hyperspectral images as third order tensors and processing using various tensor decomposition methods. This paper focuses on behavioural analysis of hyperspectral images processed with various decompositions. The experiments includes processing raw hyperspectral image and pre-processed hyperspectral image with tensor decomposition methods such as, Multilinear Singular Value Decomposition and Low Multilinear Rank Approximation technique. The results are projected based on relative reconstruction error, classification and pixel reflectance spectrums. The analysis provides correlated experimental results, which emphasizes the need of pre-processing for hyperspectral images and the trend followed by the tensor decomposition methods. © 2019, Springer Nature Singapore Pte Ltd.

More »»

2019

A. Unnikrishnan, Sowmya V., and Dr. Soman K. P., “A two-band convolutional neural network for satellite image classification”, Lecture Notes in Electrical Engineering, vol. 500, pp. 161-170, 2019.[Abstract]


The advent of neural networks has led to the development of image classification algorithms that are applied to different fields. In order to recover the vital spatial factor parameters, for example, land cover and land utilization, image grouping is most important in remote sensing. Recently, benchmark classification accuracy was achieved using convolutional neural networks (CNNs) for land cover classification. The most well-known tool which indicates the presence of green vegetation from multispectral pictures is the Normalized Difference Vegetation Index (NDVI). This chaper utilizes the success of the NDVI for effective classification of a new satellite dataset, SAT-4, where the classes involved are types of vegetation. As NDVI calculations require only two bands of information, it takes advantage of both RED- and NIR-band information to classify different land cover. The number and size of filters affect the number of parameters in convolutional networks. Restricting the aggregate number of trainable parameters reduces the complexity of the function and accordingly decreases overfitting. The ConvNet Architecture with two band information, along with a reduced number of filters, was trained, and high-level features obtained from a tested model managed to classify different land cover classes in the dataset. The proposed architecture, results in the total reduction of trainable parameters, while retaining high accuracy, when compared with existing architecture, which uses four bands. © Springer Nature Singapore Pte Ltd. 2019.

More »»

2019

K. S. Charmisha, Sowmya V., and Dr. Soman K. P., “Dimensionality reduction by dynamic mode decomposition for hyperspectral image classification using deep learning and kernel methods”, Communications in Computer and Information Science, vol. 968, pp. 256-267, 2019.[Abstract]


Hyperspectral images are remotely sensed high dimension images, which capture a scene at different spectral wavelengths. There is a high correlation between the bands of these images. For an efficient classification and processing, the high data volume of the images need to be reduced. This paper analyzes the effect of dimensionality reduction on hyperspectral image classification using vectorized convolution neural network (VCNN), Grand Unified Recursive Least Squares (GURLS) and Support Vector Machines (SVM). Inorder to analyze the effect of dimensionality reduction, the network is trained with dimensionally reduced hyperspectral data for VCNN, GURLS and SVM. The experimental results shows that, one-sixth of the total number of available bands are the maximum possible reduction in feature dimension for Salinas-A and one-third of the total available bands are for Indianpines dataset that results in comparable classification accuracy. © 2019, Springer Nature Singapore Pte Ltd.

More »»

2019

Sowmya V., Govind, D., and Dr. Soman K. P., “Significance of processing chrominance information for scene classification: a review”, Artificial Intelligence Review, 2019.[Abstract]


The primary objective of this paper is to provide a detailed review of various works showing the role of processing chrominance information for color-to-grayscale conversion. The usefulness of perceptually improved color-to-grayscale converted images for scene classification is then studied as a part of this presented work. Various issues identified for the color-to-grayscale conversion and improved scene classification are presented in this paper. The review provided in this paper includes, review on existing feature extraction techniques for scene classification, various existing scene classification systems, different methods available in the literature for color-to-grayscale image conversion, benchmark datasets for scene classification and color-to-gray-scale image conversion, subjective evaluation and objective quality assessments for image decolorization. In the present work, a scene classification system is proposed using the pre-trained convolutional neural network and Support Vector Machines developed utilizing the grayscale images converted by the image decolorization methods. The experimental analysis on Oliva Torralba scene dataset shows that the color-to-grayscale image conversion technique has a positive impact on the performance of scene classification systems. © 2019, Springer Nature B.V.

More »»

2018

S. Srivatsa, Sowmya V., and Dr. Soman K. P., “Least square based fast denoising approach to hyperspectral imagery”, Advances in Intelligent Systems and Computing, vol. 518, pp. 107-115, 2018.[Abstract]


The presence of noise in hyperspectral images degrades the quality of applications to be carried out using these images. But, since a hyperspectral data consists of numerous bands, the total time requirement for denoising all the bands will be much higher compared to normal RGB or multispectral images. In this paper, a denoising technique based on Least Square (LS) weighted regularization is proposed. It is fast, yet efficient in denoising images. The proposed denoising technique is compared with Legendre-Fenchel (LF) denoising, Wavelet-based denoising, and Total Variation (TV) denoising methods based on computational time requirement and Signal-to-Noise Ratio (SNR) calculations. The experimental results show that the proposed LS-based denoising method gives as good denoising output as LF and Wavelet, but with far lesser time consumption. Also, edge details are preserved unlike in the case of total variation technique. © Springer Nature Singapore Pte Ltd. 2018.

More »»

2018

M. Swarna, Sowmya V., and Dr. Soman K. P., “Effect of dimensionality reduction on sparsity based hyperspectral unmixing”, Advances in Intelligent Systems and Computing, vol. 614, pp. 429-439, 2018.[Abstract]


Interpretation of hyperspectral data is challenging due to the lack of spatial resolution, which causes mixing of endmember information in each pixel. Hyperspectral unmixing aims at extracting the information related to the fractional abundance of each endmember present in every pixel. The unmixing problem can be carried out by considering that the spectral signature of each endmember is a linear combination of the pure spectral signatures known in prior. In this work, sparse unmixing techniques such as, Orthogonal Matching Pursuit and Alternating Directional Multiplier Methods are applied along with dimensionality reduction of the hyperspectral image. Dimensionality reduction is obtained using the Inter-Band Block Correlation followed by singular value and QR decomposition (SVD-QR). Furthermore, we analyze the effect of dimensionality reduction on two different unmixing algorithms. Our experimentation is carried out on two real hyperspectral datasets namely ‘samson’ and ‘jasper ridge’ and the results comprises of a comparison between hyperspectral unmixing before and after dimensionality reduction using the standard metrics such as root mean square error, classwise-accuracy and visual perception. This provides a new outlook for the unmixing process as abundance estimation can be done with only the most informative bands of the image instead of using the entire data by using the dimensionality reduction technique. © Springer International Publishing AG 2018.

More »»

2018

R. Reshma, Sowmya V., and Dr. Soman K. P., “Effect of Legendre–Fenchel denoising and SVD-based dimensionality reduction algorithm on hyperspectral image classification”, Neural Computing and Applications, vol. 29, pp. 301-310, 2018.[Abstract]


This paper describes the importance of performing preprocessing techniques namely, denoising and dimensionality reduction to the hyperspectral data before classification. The two main problems faced in hyperspectral image processing are noise and higher dimension. Legendre–Fenchel transformation denoises each band in the data while preserving the edge information. To overcome the issue of high data volume, inter-band block correlation coefficient technique followed by singular value decomposition and QR decomposition is utilized to reduce the dimension of hyperspectral image without affecting the critical information. The preprocessed data are classified using kernel-based libraries, namely GURLS and LibSVM. Performance of these techniques is evaluated with accuracy assessment measures. The experiment was performed on five datasets. Experimental analysis shows that the proposed denoising technique increases the classification accuracy. In the case of Indian Pines data, with 10% of the training data, the classification accuracy is improved from 83.5 to 97.3%. And also, dimensionality reduction technique gives good classification accuracy even with 50% reduction in the number of bands. The classification accuracy of the Salinas-A and Pavia University data is 99.4 and 94.6% with the 50% dimensionally reduced (100 and 50 bands, respectively) number of bands. The bands extracted by the dimensionality reduction technique using the denoised hyperspectral data differ from that of the hyperspectral data without denoising. This emphasizes the importance of denoising the dataset before applying dimensionality reduction technique. In case of Pavia University, the band numbers above 50 (out of 100 bands) which were not informative bands before denoising are selected as informative bands by dimension reduction technique after denoising. © 2017, The Natural Computing Applications Forum.

More »»

2018

V. Ankarao, Sowmya V., and Soman, K. P., “Fusion of panchromatic image with low-resolution multispectral images using dynamic mode decomposition”, Lecture Notes in Electrical Engineering, vol. 490, pp. 335-346, 2018.[Abstract]


Remote sensing applications, like classification, vegetation, environmental changes, land use, land cover changes, need high spatial information along with multispectral data. There are many existing methods for image fusion, but all the methods are not able to provide the resultant without any deviations in the image properties. This work concentrates on embedding the spatial information of the panchromatic image onto spectral information of the multispectral image using dynamic mode decomposition (DMD). In this work, we propose a method for image fusion using dynamic mode decomposition (DMD) and weighted fusion rule. Dynamic mode decomposition is a data-driven model and it is able to provide the leading eigenvalues and eigenvectors. By separating the leading and lagging eigenvalues, we are able to construct modes for the datasets. We have calculated the fused coefficients by applying the weighted fusion rule for the decomposed modes. Proposed fusion method based on DMD is validated on four different datasets. Obtained results are analyzed qualitatively and quantitatively and are compared with four existing methods—generalized intensity hue saturation (GIHS) transform, Brovey transform, discrete wavelet transform (DWT), and two-dimensional empirical mode decomposition (2D-EMD). © Springer Nature Singapore Pte Ltd. 2018.

More »»

2018

M. Swarna, Sowmya V., and Dr. Soman K. P., “Band selection using variational mode decomposition applied in sparsity-based hyperspectral unmixing algorithms”, Signal, Image and Video Processing, vol. 12, no. 8, pp. 1463-1470, 2018.[Abstract]


In this work, a frequency-based dimensionality reduction technique using variational mode decomposition (VMD) is proposed. Dimensionality reduction is a very important aspect of preprocessing in case of hyperspectral image (HSI) analysis where this step helps in elimination of the lesser informative bands, thereby reducing the size of the data and making its processing computationally less challenging. In contrast to the standard dimensionality reduction methods such as inter-band block correlation (IBBC) where bands are eliminated based on their similarity with the consecutive bands, the proposed method uses frequency information of each band to categorize it as a less or more informative band. In this way, only the topmost informative bands of HSI are selected to form the reduced dataset. In our experiment, in order to verify the efficiency of VMD as a dimensionality reduction technique, the hyperspectral unmixed results obtained for IBBC reduced dataset is compared with those obtained for VMD reduced dataset. From the parametric measures such as classification accuracy, root-mean-square error (RMSE) and visual results obtained after unmixing for both IBBC and VMD reduced datasets, it is noticed that the VMD reduced dataset performs better by achieving higher classification accuracy and lower RMSE than that of the existing IBBC method.

More »»

2018

V. Ankarao, Sowmya V., and Dr. Soman K. P., “Multi-sensor data fusion using NIHS transform and decomposition algorithms”, Multimedia Tools and Applications, vol. 77, pp. 30381–30402, 2018.[Abstract]


Multi-spectral image fusion is to enhance the details present in multi-spectral bands with the spatial information available in the panchromatic image. Fused images have the effect of spectral distortions and lack of structural similarity. To overcome these limitations, three methods are proposed using intensity, hue, saturation (IHS) and nonlinear IHS (NIHS) transform along with the Dynamic Mode Decomposition (DMD) and 2D-Empirical Mode Decomposition (2D-EMD or IEMD). An intensity plane is calculated from the NIHS transform. The modes are constructed using DMD by considering the variations between the intensity plane computed using NIHS transforms of a low resolution multi-spectral image and a panchromatic image. Similarly, 2D-EMD is also used for image fusion. Modes are subjected to weighted fusion rule to get an intensity plane with spatial and edge information. Finally, the calculated intensity plane is concatenated along with the hue and saturation plane of low-resolution multi-spectral image and transformed into RGB color space. Thus, the fused images have high spatial and edge information on spectral bands. The experiments and its quality assessment assure that proposed methods perform better than the existing methods.

More »»

2018

P. Megha, Sowmya V., and Dr. Soman K. P., “Effect of dynamic mode decomposition-based dimension reduction technique on hyperspectral image classification”, Lecture Notes in Electrical Engineering, vol. 490, pp. 89-99, 2018.[Abstract]


Hyperspectral imaging has become an interesting area of research in remote sensing over the past thirty years. But the main hurdles in understanding and analyzing hyperspectral datasets are the high dimension and presence of noisy bands. This work proposes a dynamic mode decomposition (DMD)-based dimension reduction technique for hyperspectral images. The preliminary step is to denoise every band in a hyperspectral image using least square denoising, and the second stage is to apply DMD on hyperspectral images. In the third stage, the denoised and dimension reduced data is given to alternating direction method of multipliers (ADMMs) classifier for validation. The effectiveness of proposed method in selecting most informative bands is compared with standard dimension reduction algorithms like principal component analysis (PCA) and singular value decomposition (SVD) based on classification accuracies and signal-to-noise ratio (SNR). The results illuminate that the proposed DMD-based dimension reduction technique is comparable with the other dimension reduction algorithms in reducing redundancy in band information. © Springer Nature Singapore Pte Ltd. 2018.

More »»

2017

Sowmya V., Dr. Govind D., and Soman, K. Padanyl, “Significance of perceptually relevant image decolorization for scene classification”, Journal of Electronic Imaging, vol. 26, no. 6, 2017.[Abstract]


A color image contains luminance and chrominance components representing the intensity and color information respectively. The objective of the work presented in this paper is to show the significance of incorporating the chrominance information for the task of scene classification. An improved color-to-grayscale image conversion algorithm by effectively incorporating the chrominance information is proposed using color-to-gay structure similarity index (C2G-SSIM) and singular value decomposition (SVD) to improve the perceptual quality of the converted grayscale images. The experimental result analysis based on the image quality assessment for image decolorization called C2G-SSIM and success rate (Cadik and COLOR250 datasets) shows that the proposed image decolorization technique performs better than 8 existing benchmark algorithms for image decolorization. In the second part of the paper, the effectiveness of incorporating the chrominance component in scene classification task is demonstrated using the deep belief network (DBN) based image classification system developed using dense scale invariant feature transform (SIFT) as features. The levels of chrominance information incorporated by the proposed image decolorization technique is confirmed by the improvement in the overall scene classification accuracy . Also, the overall scene classification performance is improved by the combination of models obtained using the proposed and the conventional decolorization methods.

More »»

2017

Sowmya V., “Effect of Legendre-Fenchel Denoising and SVD based Dimensionality Reduction Algorithm on Hyperspectral Image Classification”, Special Issue in the Journal of Neural Computing and Applications, Springer, 2017.

2017

V. V. Pradeep, Sowmya V., and Dr. Soman K. P., “Application of M-band wavelet in pan-sharpening”, Special Issue in the Journal of Intelligent and Fuzzy Systems, IOS Press, Netherlands, vol. 32, no. 4, pp. 3151-3158, 2017.[Abstract]


Remote sensing satellites are proficient in taking earth images across various regions in visible part of electromagnetic spectrum. The images can be panchromatic image of a single band, multispectral image of three to seven different bands, and hyperspectral image taken from about 220 contiguous spectral bands. These images are used together or on its own, depending on the significance and usage of the preferred application. Pan-sharpening is one method which is used to improve the quality of a low resolution multispectral image by fusion with a high resolution panchromatic image. This paper proposes a method based on M-band wavelets for the pan-sharpening of a low resolution multispectral image. The method tries to improve the spatial characteristics while preserving the spectral quality of the data. The proposed technique uses weighted fusion rule and average fusion rule. The data used for the experiment were acquired by high resolution optical imagers onboard QuickBird, WorldView-3, WorldView-2 and GeoEye-1. A comparison with existing fusion techniques is done based on image quality metrics and visual interpretation. The experimental results and analysis suggests that the proposed pan-sharpening technique outperforms other compared pre-existing pan-sharpening methods. © 2017-IOS Press and the authors. All rights reserved.

More »»

2017

Sowmya V., Govind, D., and Soman, K. P., “Significance of incorporating chrominance information for effective color-to-grayscale image conversion”, Signal, Image and Video Processing, vol. 11, no. 1, pp. 129–136, 2017.[Abstract]


This paper provides an alternative framework for color-to-grayscale image conversion by exploiting the chrominance information present in the color image using singular value decomposition (SVD). In the proposed technique of color-to-grayscale image conversion, a weight matrix corresponds to the chrominance components is derived by reconstructing the chrominance data matrix (planes a* and b*) from the eigenvalues and eigenvectors computed using SVD. The final grayscale converted image is obtained by adding the weighted chrominance data to the luminous intensity which is kept intact for the CIEL*a*b* color space of the given color image. The effectiveness of the proposed grayscale conversion is confirmed by the comparative analysis performed on the color-to-gray benchmark dataset across 10 existing algorithms based on the standard objective measures, namely normalized cross-correlation, color contrast preservation ratio, color content fidelity ratio, E score and subjective evaluation.

More »»

2016

M. Swarna, Megha, P., Sowmya V., and Dr. Soman K. P., “Regularized least square approach for remote sensing image denoising using wavelet filters”, Indian Journal of Science and Technology, vol. 9, 2016.[Abstract]


Noise in remote sensing images (aerial and satellite) is caused due to various reasons such as atmospheric interference or lack of quality in sensors used to capture them. Removal of noise in an efficient way is a big challenge for researchers. In this paper, one dimensional signal denoising based on weighted regularized least square method is mapped to two dimensional image de noising. Objectives: This paper introduces a novel image denoising technique based on least square weighted regularization. Methods/Statistical Analysis: The proposed technique for image denoising based on Least Square (LS) approach is experimented on five different satellite and aerial images corrupted by gaussian noise with varying noise levels and regularization parameter lambda (λ) for different wavelet filter coefficients such as 'haar', 'symlet', 'daubechies' and coiflet. The effectiveness of the proposed method of image denoising is compared against the existing second order filter [based on LS] and conventional wavelet based image denoising technique based on the standard metric called Peak Signal to Noise Ratio (PSNR). Findings: From the experimental result analysis obtained it is inferred that the wavelet filters outperforms the second order filter and the conventional wavelet based image denoising. The complexity of the mathematics is low in our proposed method for image denoising. Applications/Improvements: The proposed denoising technique can be adopted as a faster pre-processing step in most of the image processing applications.

More »»

2016

Sowmya V., Divu, G., and Dr. Soman K. P., “Significance of incorporating chrominance information for effective color-to-grayscale image conversion”, Signal, Image and Video Processing, vol. 11, 2016.[Abstract]


This paper provides an alternative framework for color-to-grayscale image conversion by exploiting the chrominance information present in the color image using singular value decomposition (SVD). In the proposed technique of color-to-grayscale image conversion, a weight matrix corresponds to the chrominance components is derived by reconstructing the chrominance data matrix (planes a* and b*) from the eigenvalues and eigenvectors computed using SVD. The final grayscale converted image is obtained by adding the weighted chrominance data to the luminous intensity which is kept intact for the CIEL*a*b* color space of the given color image. The effectiveness of the proposed grayscale conversion is confirmed by the comparative analysis performed on the color-to-gray benchmark dataset across 10 existing algorithms based on the standard objective measures, namely normalized cross-correlation, color contrast preservation ratio, color content fidelity ratio, E score and subjective evaluation. More »»

2016

P. Megha, Swarna, M., K. Dixon, D. Merlin, Dr. Soman K. P., and Sowmya V., “Impact of least square denoising on kernel based hyperspectral image classification”, International Journal of Control Theory and Applications, vol. 9, pp. 4623-4630, 2016.[Abstract]


Hyperspectral sensors capture images in hundreds of spectral bands spanning almost all the regions in the electromagnetic spectrum. These images consists of many noisy bands. Hence, a strong diffusion scheme (denoising) is required to extract the meaningful spatial information present in the noisy spectral bands. In this paper, one dimensional signal denoising based on weighted regularized least square (LS) method is mapped to two dimensional hyperspectral image (HSI) denoising and the superiority of this method is determined on the basis of the classification accuracies of the hyperspectral image obtained using Grand Unified Regularized Least Square (GURLS) library and LibSVM, the support vector machines library. The proposed method brings out the efficiency of LS denoising based on the classification accuracies obtained on classifying the denoised image. The obtained results are also compared with the accuracy obtained on classifying the original image without denoising. The analysis is also extended to the classification of the images denoised using the conventional denoising techniques such as Total Variation (TV) and LF (Legendre Fenchel) based denoising. From the analysis, it is observed that the classification accuracies of the images denoised using the proposed method is much higher than the conventional denoising methods. Hence, showing that the LS based denoising is an efficient method which provides a denoised output almost similar to the original image. © International Science Press.

More »»

2016

Sowmya V., Praveena, R., and K. P. Soman, “Least Square based Signal Denoising and Deconvolution using Wavelet Filters”, Indian Journal of Science and Technology, vol. 9, no. 33, 2016.[Abstract]


Noise, the unwanted information in a signal reduces the quality of signal. Hence to improve the signal quality, denoising is done. The main aim of the proposed method in this paper is to deconvolve and denoise a noisy signal by least square approach using wavelet filters. In this paper, least square approach given by Selesnick is modified by using different wavelet filters in place of second order sparse matrix applied for deconvolution and smoothing. The wavelet filters used in the proposed approach for denoising are Haar, Daubechies, Symlet, Coiflet, Biorthogonal and Reverse biorthogonal. The result of the proposed experiment is validated in terms of Peak Signal to Noise Ratio (PSNR). Analysis of the experiment results notify that proposed denoising based on least square using wavelet filters are comparable to the performances given by deconvolution and smoothing using the existing second order filter.

More »»

2016

D. P. Kuttichira, Sowmya V., and Dr. Soman K. P., “Digit recognition using multiple feature extraction”, IIOAB Journal, vol. 7, pp. 37-43, 2016.[Abstract]


Digit Recognition is one of the classic problems in pattern classification. It has ten labels which are digits from 0-9 and each prototypes in the test set has to be classified under these labels. In this paper, we have used MNIST data for training and testing. MNIST database is a standard database for digit classification. A number of neural network algorithms have been used on MNIST to get high accuracy outputs. These algorithms are computationally costly. Here, we have used multiple feature extraction based on SVD and histogram to create testing and training matrix. To the feature vector formed by SVD, histogram values along x-axis and y-axis of an image is appended. These vectors are mapped to hyperplane using polynomial and Gaussian kernel. For classification open source software like GURLS and LIBSVM is used to obtain a fairly good accuracy. © 2016, Institute of Integrative Omics and Applied Biotechnology. All rights reserved.

More »»

2016

A. M, N, D., Sowmya V., Mahan, N., and Dr. Soman K. P., “Least Square Based approach for Image Inpainting”, Institute of Integrative Omics and Applied Biotechnology, vol. 7, pp. 44-59, 2016.[Abstract]


Images are widely used over various applications under the aegis of various domains like Computer vision, Biomedical, etc. The problem of missing data identification is of great concern in various fields involving image processing. Least square can be used for missing sample estimation for 1-D signals. The proposed system extends the missing sample estimation in 1-D using least square to 2-D, applied for image inpainting. The paper also draws a comparison between the Total Variation (TV) algorithm and the proposed method. The experiments were conducted on standard images and the standard metrics namely PSNR and SSIM are used to compare the image quality obtained using the proposed method (least square based) and TV algorithm. More »»

2016

H. T. Suseelan, Sudhakaran, S., Sowmya V., and Dr. Soman K. P., “Performance Evaluation of Sparse Banded Filter Matrices using content based image retrieval”, Institute of Integrative Omics and Applied Biotechnology, vol. 7, no. 3, pp. 11-18, 2016.[Abstract]


Content Based Image Retrieval (CBIR) is an extensively used application in the field of Image Processing. It is used to search through a massive database and retrieve the images that have similarity with the query image. In this paper, performance is evaluated for Sparse Banded Filter matrices (ABfilter) against the standard edge detection filters through Content Based Image Retrieval. Performance factor of ABfilter directly relates to its edge detection capabilities. Here, edge detection followed by the Singular Value Decomposition (SVD) is done for feature extraction for both the query and images in database. Query image feature and database image features are matched and those having similar values are retrieved. Similarity measurement is done by computing the distance between corresponding features. Experimental results indicate that retrieval results using ABfilter is much better than using standard edge detection filters for the same, which in turn establishes its superiority in edge detection.

More »»

2016

V. P.V, G, R. Devi, Sowmya V., and Dr. Soman K. P., “Least Square based image denoising using wavelet filters”, Indian Journal of Science and Technology, vol. 9, no. 30, 2016.[Abstract]


Background/Objectives: Noise in a digital image, is unwanted information that degrades the quality of an image. The main aim of the proposed method is to denoise a noisy image based on least square approach using wavelet filters. Methods/ Statistical Analysis: One dimensional least square approach proposed by Selesnick is extended to two dimensional image denoising. In our proposed technique of least square problem formulation for image denoising, the matrix constructed using second order filter coefficients is replaced by wavelet filter coefficients. Findings: The method is experimented on standard digital images namely Lena, Cameraman, Barbara, Peppers and House. The images are subjected to different noise types such as Gaussian, Salt and Pepper and Speckle with varying noise level ranging from 0.01db to 0.5db. The wavelet filters used in the proposed approach of denoising are Haar, Daubechies, Symlet, Coiflet, Biorthogonal and Reverse biorthogonal. The outcome of the experiment is evaluated in terms of Peak Signal to Noise Ratio (PSNR). The analysis of the experiment results reveals that performance of the proposed method of least square based image denoising by wavelet filters are comparable to denoising using existing second order sparse matrix. Applications/Improvements: Digital images are often prone to noise; hence, proceeding with further processing of such an image requires denoising. This work can be extended in future to m-band wavelet filters.

More »»

2016

D. Merlin K. Dixon, Ajay, A., Soman, K. P., and Sowmya V., “Aerial and satellite image denoising using least square weighted regularization method”, Indian Journal of Science and Technology, vol. 9, no. 30, 2016.[Abstract]


Remotely sensed images are subjected to various types of noises. Noise interrupts the image information; hence noise removal is one of the important pre-processing steps in every image processing applications. Since both noise and edges contain highintensity values, image denoising leads to smoothening of the edges thereby reducing the visual quality of the image. Hence, edge preserved image denoising is an ever-relevant topic. Over decades, several image denoising techniques were developed. Most of the denoising algorithms are very complex and time consuming. Background/Objectives: This paper introduces a novel image denoising technique based on least square weighted regularization. Methods/Statistical Analysis: The onedimensional signal denoising introduced by14 is mapped into two-dimensional image denoising. The proposed method is experimented on a set of colored aerial and satellite images. The column-wise denoising of the image is performed first, followed by row-wise denoising. The performance of the proposed method is evaluated based on the standard quality metric peak signal-to-noise ratio and computational time. Findings: From the experimental results, it is observed that the proposed method outperforms the earlier denoising methods on the basis of time and complexity. Applications/Improvements: The proposed denoising technique can be adopted as a faster pre-processing step in most of the image processing applications.

More »»

2016

V. V. Pradeep, Sowmya V., and Soman, K. P., “Variational mode decomposition based multispectral and panchromatic image fusion”, International Journal of Control Theory and Applications, vol. 9, no. 16, pp. 8051-8059, 2016.[Abstract]


The technique of fusing multispectral image with panchromatic image in order to get a resultant output image of relatively higher spectral resolution and higher spatial information is termed as pan sharpening. It is being used in many remote sensing tasks for different applications including classification, segmentation, change detection, etc. This paper proposes the usage of Variational Mode Decomposition (VMD) as a technique for fusing multispectral and panchromatic images. It also considers average fusion rule and weighing fusion rule during its procedural steps. The experiment is being done on datasets acquired by high resolution sensors on-board satellites such as QuickBird, WorldView-3, WorldView-2 and GeoEye-1. Quantitative assessment measures and visual perception evaluates the effectiveness of the method. The analysis from the obtained results suggest that the proposed method can be used as an image fusion technique and its performance is comparable to the pre-existing pan sharpening techniques like Multi-resolution Singular Value Decomposition (MSVD), Discrete Wavelet Transform (DWT) and Empirical Wavelet Transform (EWT). © International Science Press.

More »»

2016

Reshma R, Sowmya V., and Dr. Soman K. P., “Improvement in kernel based Hyperspectral image classification using legendre fenchel denoising”, Indian Journal of Science and Technology, vol. 9, no. 33, 2016.[Abstract]


Hyperspectral images have bulk of information which are widely used in the field of remote sensing. One of the main problems faced by these images is noise. This emphasizes the importance of denoising techniques for enhancing the image quality. In this paper, Legendre Fenchel Transformation (LFT) is used for preprocessing the Indian Pines Dataset. LFT reduces the noise of each band of the hyperspectral image without affecting the edge information. Signal to noise ratio is computed which helps to evaluate the performance of denoising. Further, the denoised image is classified using GURLS and LibSVM and the various accuracies are estimated. The experimental analysis shows that the overall and classwise accuracies are more for the preprocessed data classification when compared to the classification without preprocessing. The classification accuracy is improved with denoising of hyperspectral image.

More »»

2016

S. Se, Pradeep, D., Sowmya V., and Dr. Soman K. P., “Fourier Descriptor features for Shape Deformation Classification using Random Kitchen Sink”, Journal of Chemical and Pharmaceutical Sciences, vol. 9, pp. 554-558, 2016.[Abstract]


This paper deals with the Fourier descriptor features for shape deformation classification using Random Kitchen Sink algorithm accessed through GURLS library. Shape recognition is an important method used in all industrial environments which are mostly concerned with robots. It is a highly essential task to make the robot understand the shape of an object. The object may have many deformed shapes and so it is necessary to train the classifier accordingly. Recognition methods based on polar coordinates and probabilistic models are already developed, but its accuracy for finding the deformed shape of the object is low. In this context, Random Kitchen Sink algorithm is used and the classification is done through GURLS in which, regularized least square method is used, which leads to better shape recognition. More »»

2016

N. Nechikkat, Sowmya V., and Dr. Soman K. P., “Low dimensional variational mode features for hyperspectral image classification”, Journal of Chemical and Pharmaceutical Sciences, vol. 9, pp. 565-570, 2016.[Abstract]


High Dimensionality is always a great concern while working with hyperspectral images. The high dimension of hyperspectral image increases the computational complexity, creates data storage issues and decrease the performance and accuracy of hyperspectral image analysis algorithms. This paper focuses on low dimensional Variational Mode features for hyperspectral image classification. The proposed method consist of three stages: preprocessing using Inter Band Block Correlation (IBBC) technique, feature extraction using Variational Mode Decomposition (VMD) and dimensionality reduction using Singular Value Decomposition (SVD). The efficiency of the proposed method based on the low dimensional feature extraction using VMD is evaluated by one of the sparsity based classification algorithms namely Orthogonal Matching Pursuit (OMP). The proposed work is experimented on the standard dataset namely Indian pines acquired by the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). The experimental analysis shows that our proposed technique produces 90.88% overall accuracy with 40% of training which is greater than the classification accuracy obtained without feature extraction. More »»

2016

L. S. Kiran, Sowmya V., and Dr. Soman K. P., “Enhanced Variational Mode Features for Hyperspectral Image Classification”, Journal of Chemical and Pharmaceutical Sciences, vol. 9, pp. 502-505, 2016.[Abstract]


Variational Mode Decomposition (VMD) is a recent method and is gaining popularity in the area of signal and image processing. The use of this decomposition technique in hyper spectral image classification is discussed in detail in this paper. The role of VMD as a feature extraction technique is exploited here. The proposed method includes an initial stage of dimensionality reduction so as to reduce the computational complexity. A final stage of recursive filtering is also added to further enhance the results. Results obtained by the proposed method on two hyper spectral image datasets 'Indian Pines and Salinas-A, suggests that VMD is a promising method in the area of image analysis and classification. Quality indices used for experimental analysis include overall accuracy (OA), average accuracy (AA) and kappa coefficient. Notable classification accuracy has been obtained for both the datasets and a final stage of recursive filtering has further improved the results (more than 98% accuracy in the case of Indian Pines). More »»

2015

Sowmya V., S., M., and Soman, K. P., “Performance Comparison of Empirical wavelet Transform and Empirical Mode Decomposition on Pan sharpening”, International Journal of Applied Engineering Research, vol. 10, no. 73, 2015.[Abstract]


Pan sharpening is defined as the fusion of low resolution multispectral image with panchromatic image, which plays a significant role in the field of remote sensing. This work considers the fusion of multispectral and panchromatic images (Pan sharpening) using EWT (Empirical Wavelet Transform) and EMD (Empirical Mode Decomposition). EWT and EMD algorithm decomposes the input image into several modes. This paper focus on the performance evaluation of image fusion technique based on EMD and EWT. The image fusion methods based on EWT and EMD are experimented on five sets of panchromatic and multispectral images captured by high resolution earth observation satellites. The efficiency of the fusion methods are evaluated by visual perception and standard fusion metrics. The experimental result analysis based on computed fusion metrics and computation time shows that the fusion method based on EWT outperforms the EMD based fusion approach.

More »»

2015

N. Nechikkat, Sowmya V., and Soman, K. P., “A Comparative Analysis of Variational Mode and Empirical Mode Features on Hyperspectral Image Classification”, International Journal of Applied Engineering Research, vol. 10, no. 73, 2015.[Abstract]


Considering the fact that involving spatial information in feature extraction significantly improves the classification accuracies, this paper focuses on Variational Mode Decomposition (VMD) and Empirical Mode Decomposition (EMD) as the featureextraction algorithms. Both the algorithms decompose an input image into different modes with each mode including different regions of frequency with unique properties. Here, the proposed method includes processing the same set of data with two different decomposition methods to compare the effect of the methods on the standard dataset. The method incorporates a preprocessing technique for noisy band removal, processing technique for feature extraction, band selection methods for dimensionality reduction and classification technique for result validation

More »»

2015

A. C, Haridas, N., Sowmya V., and Dr. Soman K. P., “Effect of AB filter denoising on ADMM based Hyperspectral Image Classification”, International Journal of Applied Engineering Research (IJAER), vol. 10, no. 73, pp. 127-131, 2015.[Abstract]


In recent years, hyperspectral remote sensing has emerged as a prominent area of research. This has developed a lot of practical solutions to solve the various challenges faced in the field. Noise is one of such issues which deteriorate the quality of information present in the hyperspectral images. In order to address this problem, various preprocessing (denoising) techniques are applied prior to data analysis. In this paper, the proposed method evaluates the effect of Hyperspectral Image (HSI) denoising employing AB filter on optimization based classification which uses Basis Pursuit solved by Alternating Direction Method of Multipliers (ADMM). AVIRIS Indian Pines dataset is used for the experimental study. The efficiency of the proposed technique is proved by a comparative study with other existing preprocessing methods. The experimental result analysis based on visual interpretation and quantitative assessment shows that the proposed method provides better classification results compared to the existing methods. The classification results are assessed by Overall Accuracy, Average accuracy and Kappa coefficient.

More »»

2015

N. Haridas, Sowmya V., and Soman, K. P., “GURLS vs LIBSVM: Performance Comparison of Kernel Methods for Hyperspectral Image Classification”, Indian Journal of Science and Technology, vol. 8, no. 24, 2015.[Abstract]


Kernel based methods have emerged as one of the most promising techniques for Hyper Spectral Image classification and has attracted extensive research efforts in recent years. This paper introduces a new kernel based framework for Hyper Spectral Image (HSI) classification using Grand Unified Regularized Least Squares (GURLS) library. The proposed work compares the performance of different kernel methods available in GURLS package with the library for Support Vector Machines namely, LIBSVM. The assessment is based on HSI classification accuracy measures and computation time. The experiment is performed on two standard Hyper Spectral datasets namely, Salinas A and Indian Pines subset captured by AVIRIS (Airborne Visible Infrared Imaging Spectrometer) sensor. From the analysis, it is observed that GURLS library is competitive to LIBSVM in terms of its prediction accuracy whereas computation time seems to favor LIBSVM. The major advantage of GURLS toolbox over LIBSVM is its simplicity, ease of use, automatic parameter selection and fast training and tuning of multi-class classifier. Moreover, GURLS package is provided with an implementation of Random Kitchen Sink algorithm, which can easily handle high dimensional Hyper Spectral Images at much lower computational cost than LIBSVM.

More »»

2015

C. Aswathy, Sowmya V., and Dr. Soman K. P., “ADMM based hyperspectral image classification improved by denoising using legendre fenchel transformation”, Indian Journal of Science and Technology, vol. 8, no. 24, 2015.[Abstract]


This paper discusses about a sparsity based algorithm used for Hyperspectral Image (HSI) classification where the test pixel vectors are sparsely represented as the linear combination of a few number of training samples from a well-organised dictionary matrix. The sparse vector is obtained using Basis Pursuit (BP) which is a constrained l4 minimization problem. This problem is solved by using a simple and powerful iterative algorithm known as Alternating Direction Method of Multipliers (ADMM) which significantly reduces the computational complexity of the problem and thereby speeds up the convergence. The classification accuracy is considerably improved by including efficient preprocessing techniques to remove the unwanted information (noise) present in Hyperspectral images. This paper uses a fast and reliable denoising technique based on Legendre Fenchel Transformation (LFT) to effectively denoise each band of HSI prior to ADMM based classification (proposed method). A comparison of proposed technique with one of the convex optimization tools namely, CVX is given to exhibit the fast convergence of the former method. The experiment is performed on standard Indian Pines dataset captured using AVIRIS sensor. The potential of the proposed method is illustrated by analyzing the classification indices obtained with and without applying any preprocessing methods. With only 10% training set, an overall accuracy of 96.76% is obtained for the proposed method at a much faster rate compared to computation time taken by CVX solver.

More »»

2015

P. G. Mol, Sowmya V., and Soman, K. P., “Performance Enhancement of Minimum Volume based Hyper Spectral Unmixing Algorithms by Variational Mode Decomposition”, Indian Journal of Science and Technology, vol. 8, no. 24, 2015.[Abstract]


Hyper spectral unmixing of data has become an indispensable technique in remote sensing zone. Spectral Unmixing is defined as the source separation of a mixed pixel. The fundamental sources are termed as endmembers and percentage of the source content is known as abundances. This paper demonstrates the effect of Variational Mode Decomposition (VMD) on hyper spectral unmixing algorithms based on geometrical minimum volume approaches. The proposed method is experimented on standard hyper spectral dataset namely, cuprite. The effectiveness of the proposed method is subjected to evaluation, based on the standard quality metric namely, Root Mean Square Error (RMSE). The experimental result analysis shows that, the proposed technique enhance the performance of hyper spectral unmixing algorithms based on the geometrical minimum volume based approaches.

More »»
PDF iconperformance-enhancement-of-minimum-volume-based-hyper-spectral-unmixing-algorithms-by-variational-mode-decomposition.pdf

2015

S. Moushmi, Sowmya V., and Soman, K. P., “Multispectral and Panchromatic Image Fusion using Empirical Wavelet Transform”, Indian Journal of Science and Technology, vol. 8, no. 24, 2015.[Abstract]


Pan sharpening is the process of fusion of panchromatic and multispectral image to obtain an output image of high spatial and spectral resolution. It is very important for various remote sensing applications such as image segmentation studies, image classification, temporal change detection etc. The present work demonstrates the application of Empirical Wavelet Transform for the fusion of panchromatic image and multispectral image by simple average fusion rule. The Proposed method is experimented on panchromatic and multispectral images captured by high resolution earth observation satellites such as GeoEye-1, QuickBird, WorldView-2 and World View-3. The effectiveness of our proposed method is evaluated by visual perception and quantitative assessment measures. The experimental analysis shows that the proposed method performs comparable to the existing fusion algorithms such as Multi-resolution Singular Value Decomposition and Discrete Wavelet Transform.

More »»

2015

Sowmya V., Neethu Mohan, and Dr. Soman K. P., “Sparse banded matrix filter for image denoising”, Indian Journal of Science and Technology, vol. 8, no. 24, 2015.[Abstract]


Noise is one of the prime factors which degrade the quality of an image. Hence, image denoising is an essential image enhancement technique in the image processing domain. In this paper, we use low-pass sparse banded filter matrices for image denoising. Sparsity is the key concept in this filter design. We applied the designed low-pass filter both row-wise and column-wise to denoise the image. The proposed method is experimented on standard test images corrupted with different types of noises namely Gaussian, White Gaussian, Salt & Pepper and Speckle with noise level equals to 0.01, 0.05 and 0.1. The effectiveness of the proposed method of denoising is evaluated by the computation of standard quality metric known as Peak Signal-to-Noise Ratio (PSNR). The experimental result analysis shows that the proposed image denoising technique based on sparse banded filter matrices results in significant improvement in PSNR around 2dB to 8dB for different type of noises with noise level equal to 0.1 and is also aided by the visual analysis.

More »»

2015

M. Aswathi, M. Babu, J., Sowmya V., and Soman, K. P., “Smart meter data security based on cosine transform and chaotic theory”, International Journal of Applied Engineering Research, vol. 10, no. 20, pp. 19181-19185, 2015.[Abstract]


The advent of Smart Metering has facilitated the knowledge of fine-grained energy-consumption data. The data is transmitted over the internet which is prone to access by an undesired third party. The high resolution data gives information regarding the amount of electricity and the type of appliances used. Smart meters have also become potential surveillance devices that monitor the behavior of the customers. This has paved way for the need for powerful privacy preservation techniques. To ensure security, we introduce a technique of obfuscating the data before transmission by computing the DCT of the signal and adding a chaotic signal to it. The proposed approach is experimented on real household’s power signal data. © Research India Publications.

More »»

2015

S. Athira, Harikumar, K., Sowmya V., and Dr. Soman K. P., “Parameter analysis of random kitchen sink algorithm”, International Journal of Applied Engineering Research, vol. 10, no. 20, pp. 19351-19355, 2015.[Abstract]


Classification is one of the main stream areas in machine learning. Many a time we have to deal with data sets that are not classifiable using a linear separator. This paper presents a theoretical and an experimental study of random kitchen sink algorithm which makes it possible for such data sets to become linearly separable by efficiently, effectively and explicitly mapping it to an appropriate higher dimensional space. It explores the dependency of various parameters of the algorithm, mainly the dimension and variance of the concerned random variable on the accuracy of classification, and how to fine tune these parameters to obtain the best of results. It also presents an intuitive understanding of how these parameters actually affect the accuracy of classification by connecting it with the shift invariant RBF kernel. The effect of choice of these parameters on a two class classification problem is also included. © Research India Publicatio

More »»

2015

Y. C. Nair, Binsha, P., V. Pradeep, V., Sowmya V., and Dr. Soman K. P., “Spreadsheet implementation of random kitchen sink for classification”, International Journal of Applied Engineering Research, vol. 10, no. 20, pp. 19228-19233, 2015.[Abstract]


In order to excavate on large chunks of unstructured data to retrieve the nuggets of knowledge, a lot of technically sound algorithms have been developed. In this paper, the classification of data has been performed in an efficient manner by inheriting concepts from linear algebra and optimization theory. This paper demonstrates the implementation of the mathematical ideas behind mapping input data to higher dimension using Random Kitchen Sink (RKS) and implementation of classification algorithm using Regularized Least Square (RLS) estimation. In this context, computational thinking methodology is accomplished using the most versatile tool available for non-programmers, that is excel. This paper elucidates a number of excel utilities. © Research India Publications.

More »»

2014

Sowmya V., P., S. K., and Deepika, J., “Image Classification Using Convolutional Neural Networks”, International Journal of Scientific & Engineering Research , vol. 5, no. 6, p. 06/2014, 2014.[Abstract]


Convolutional Neural Networks (CNNs) have been established as a powerful class of models for image recognition problems. Inspired by a blog post [1], we tried to predict the probability of an image getting a high number of likes on Instagram. We modified a pre-trained AlexNet ImageNet CNN model using Caffe on a new dataset of Instagram images with hashtag ‘me’ to predict the likability of photos. We achieved a cross validation accuracy of 60% and a test accuracy of 57% using different approaches. Even though this task is difficult because of the inherent noise in the data, we were able to train the model to identify certain characteristics of photos which result in more likes.

More »»

2014

P. S. Ashitha, Sowmya V., and Soman, K. P., “Classification of hyperspectral images using scattering transform”, International Journal of Scientific & Engineering Research, vol. 5, no. 7, pp. 315-319, 2014.[Abstract]


In this paper, we applied scattering transform approach for the classification of hyperspectral images. This method integrates features, such as the translational and rotational invariance features for image classification. The classification of hyperspectral images is more challenging because of the very high dimensionality of the pixels and the small number of labelled examples typically available for learning. The scattering transform technique is validated with two standard hyperspectral datasets i.e, SalinasA_Scene and Salinas_Scene. The experimental result analysis proves that the applied scattering transform method provides high classification accuracy of 99.35% and 89.30% and kappa coefficients of 0.99 and 0.88 for the mentioned hyperspectral image dataset respectively.

More »»

2012

K. Balakrishnan, Soman, K. P., and Sowmya V., “Spatial preprocessing for improved sparsity based hyperspectral image classification”, International Journal of Engineering Research and Technology (IJERT), vol. 1, no. 5, 2012.[Abstract]


In this paper, we present that hyperspectral image classification based on sparse representation can be significantly improved by using an image enhancement step. Spatial enhancement allows further analysis of hyperspectral imagery, as it reduces the intensity variations within the image. Perona-Malik, a partial differential equation based non-linear diffusion scheme is used for the enhancement of the hyperspectral imagery prior to classification. The diffusion technique applied smoothens the homogenous areas of hyperspectral imagery and thereby increases the separability of the classes. The diffusion scheme is applied individually to each band of the hyperspectral imagery and it does not take into account the spectral relationship among different bands. Experiments are performed on the real hyperspectral dataset AVIRIS (Airborne Visible/IR Imaging Spectrometer) 1992 Indiana Indian Pines imagery. We compared the classification statistics of hyperspectral imagery before and after performing the spatial preprocessing step in order to prove the effectiveness of the proposed method. The experiments results proved that the hyperspectral image classification using sparse representation along with spatial enhancement step lead to 97.53% of classification accuracy which is high when compared with the classification accuracy obtained without applying the spatial preprocessing technique.

More »»

2012

K. P. Soman, Kumar, S., Sowmya V., and Shajeesh, K. U., “Computational Thinking with Spreadsheet: Convolution, High-Precision Computing and Filtering of Signals and Images”, International Journal of Computer Applications, vol. 60, pp. 1-7, 2012.[Abstract]


Modern day innovations in sciences and engineering are direct outcome of human’s capacity for abstract thinking thereby creating effective computational models of the problems that can be solved efficiently by the number crunching and massive data handling capabilities of modern networked computers. Survival of any economy now depends on innovating-capacity of its citizens. Thus capacity for computational thinking has become an essential skill for survival in the 21st century. It is necessitating a fundamental change in our curriculum in schools. Computational thinking need to be introduced incrementally along with standard content in a way that makes the standard content easier to learn and vice versa. When learners successfully combine disciplinary knowledge and computational methods they develop their identity as Computational Thinkers. The need for trainers, training content and training methodology for imparting computational thinking has become subject of discussion in many international forums. In this article the use of spreadsheet as a tool for developing computational-thinking -capabilities by integrating it with existing curricula is explored. Concept of convolution which everybody uses when one does any multiplication is taken as a vehicle to develop exercises that enhance computational thinking. It is shown how convolution is visualized and implemented and also discussed a wide variety of computational experiments that students at various levels can do with the help of spreadsheet

More »»

Publication Type: Conference Proceedings

Year of Publication Title

2018

Dr. Govind D., Sowmya V., Sachin, R., and Dr. Soman K. P., “Dependency of Various Color and Intensity Planes on CNN Based Image Classification”, Advances in Intelligent Systems and Computing, International Symposium on Signal Processing and Intelligent Recognition Systems SIRS 2017, vol. 678. Springer International Publishing AG 2018, Manipal, India, pp. 167-177, 2018.[Abstract]


Scene classification systems have become an integral part of computer vision. Recent developments have seen the use of deep scene networks based on convolutional neural networks (CNN), trained using millions of images to classify scenes into various categories. This paper proposes the use of one such pre-trained network to classify specific scene categories. The pre-trained network is combined with the simple classifiers namely, random forest and extra tree classifiers to classify scenes into 8 different scene categories. Also, the effect of different color spaces such as RGB, YCbCr, CIEL*a*b* and HSV on the performance of the proposed CNN based scene classification system is analyzed based on the classification accuracy. In addition to this, various intensity planes extracted from the said color spaces coupled with color-to-gray image conversion techniques such as weighted average, and singular value decomposition (SVD) are also taken into consideration and their effects on the performance of the proposed CNN based scene classification system are also analyzed based on the classification accuracy. The experiments are conducted on the standard Oliva Torralba (OT) scene data set which comprises of 8 classes. The analysis of classification accuracy obtained for the experiments conducted on OT scene data shows that the different color spaces and the intensity planes extracted from various color spaces and color-to-gray image conversion techniques do affect the performance of proposed CNN based scene classification system.

More »»

2018

Sowmya V., Aleena Ajay, and K. P. Soman, “Vehicle detection in Aerial imagery using Eigen features”, IEEE International Conference on Communication and Signal Processing-ICCSP'17. IEEE, Adhiparasakthi Engineering College, Melmaruvathur , pp. 1620-1624, 2018.[Abstract]


The invention of low cost optical sensors lead to a rapid exploitation of aerial images in numerous applications. Automatic vehicle detection in aerial images is remarkably employed in traffic safety, urban planning, military, parking lot management, aerial surveillance, catastrophe and disaster management. The huge amount of data collected in such tasks enforce to automate the detection process. The objective of this work is to detect small vehicles from aerial images in an unconstrained environment. The experiments are conducted on the VEDAI dataset. The architecture of the system includes two phases namely training and detection. The training phase includes cropping and extraction of the training samples, feature representation and classification. And the detection phase consists of extracting the regions of interest, feature extraction and classification. The vehicles occupy only less than 1 % of total pixels in the image. Hence to limit the search area, an edge detection is performed as a pre-processing step. The bounding boxes that are selected after applying size constraint are classified using support vector machine (SVM) classifier. Two feature extraction techniques are utilized in this work specifically singular value decomposition (SVD) and histogram of oriented gradients (HOG). From the experimental results analyzed based on overall classification accuracy, the proposed SVD features provide a comparable performance with existing HOG features. © 2017 IEEE.

More »»

2018

Sowmya V., V. Ankarao, and K. P. Soman, “Sparse Image Denoising using Dictionary constructed based on Least Square Solution”, International Conference on Wireless Communications Signal Processing and Networking (WISPNET) 2017. IEEE, SSN College of Engineering, Chennai, India, pp. 1165-1171, 2018.[Abstract]


Compressed sensing became a vital tool for image or signal reconstruction with less number of samples compared with the Nyquist rate. Among the existing algorithms for reconstruction of an image using compressed sensing, orthogonal matching pursuit algorithm is cost effective in terms of computational complexity. This algorithm provides a solution for overdetermined and underdetermined systems by minimizing the error functions using least square. This work concentrates on the construction of dictionary which can be used to solve the sparsity based image denoising problem. In this paper, we constructed the dictionary using least square solution subjected to thresholding conditions such as hard, soft and semi-soft. Orthogonal matching pursuit (OMP) algorithm avoids the selection of the same atom in every iteration, due to the existence of orthogonal property between the residue and the atom selected from the dictionary. Thus, OMP algorithm results in precise image reconstruction. The proposed method is validated on four standard test images, such as Lena, Boat, Barbara and Cameraman with different noises such as salt & pepper noise, Gaussian noise and speckle noise with varying the percentage of noise level from 5% to 40%. Obtained results are evaluated by the quality metric peak-to-signal-noise ratio (PSNR) and compared with the existing wavelet based sparse image denoising. The experimental evaluation shows that the proposed method is better applicable to remove the speckle noise and salt & pepper noise when compared with the existing wavelet based sparse image denoising. © 2017 IEEE.

More »»

2018

Sowmya V., Hima . T.Suseelan, and K. P. Soman, “Image Dehazing Using Variational Mode Decomposition”, International Conference on Wireless Communications Signal Processing and Networking (WISPNET 2017). IEEE, SSN College of Engineering, Chennai, India, pp. 200-205, 2018.[Abstract]


Haze is caused by the scattering of airtight in atmosphere and it deteriorates the contrast of the photographs. Here, we propose a novel approach for haze removal based on Two Dimensional Variational Mode Decomposition (2D VMD). Two dimensional VMD decomposes the input image into desired number of bands with different central frequencies. From this set of modes, an enhanced image is reconstructed by identifying and eliminating the hazy modes. Algorithm is applied for a wide variety of hazy images including both Full Reference and Non Reference image datasets. Image Quality Assessment techniques PSNR, SSIM, and MSE are used for full reference datasets and a completely blind image quality metric, Natural Image Quality Evaluator (NIQE) is used for non reference datasets. Experimental results and analysis shows that the proposed method outperforms few of the existing state-of-the-art methods such as He et al., Kolor Neutralhazer, Photoshop Auto Contrast (PSAC) and Tang et al. © 2017 IEEE.

More »»

2018

Sowmya V., Deepa Merlin Dixon.K, and K. P. Soman, “Effect of Denoising on Vectorized Convolutional Neural Network for Hyperspectral Image Classification”, International Conference on Nextgen Electronic Technologies: Silicon to Software (ICNETS2), Lecture Notes in Electrical Engineering, vol. 490. Springer Nature Singapore Pte Ltd. 2018, VIT University, Chennai Campus, India, pp. 305-313, 2018.[Abstract]


The remotely sensed high-dimensional hyperspectral imagery is a single capture of a scene at different spectral wavelengths. Since it contains an enormous amount of information, it has multiple areas of application in the field of remote sensing, forensic, biomedical, etc. Hyperspectral images are very prone to noise due to atmospheric effects and instrumental errors. In the past, the bands which were affected by noise were discarded before further processing such as classification. Therefore, along with the noise the relevant features present in the hyperspectral image are lost. To avoid this, researchers developed many denoising techniques. The goal of denoising technique is to remove the noise effectively while preserving the important features. Recently, the convolutional neural network (CNN) serves as a benchmark on vision-related task. Hence, hyperspectral images can be classified using CNN. The data is fed to the network as pixel vectors thus called Vectorized Convolutional Neural Network (VCNN). The objective of this work is to analyze the effect of denoising on VCNN. Here, VCNN functions as the classifier. For the purpose of comparison and to analyze the effect of denoising on VCNN, the network is trained with raw data (without denoising) and denoised data using techniques such as Total Variation (TV), Wavelet, and Least Square. The performance of the classifier is evaluated by analyzing its precision, recall, and F1-score. Also, comparison based on classwise accuracies and average accuracies for all the methods has been performed. From the comparative classification result, it is observed that Least Square denoising performs well on VCNN. © Springer Nature Singapore Pte Ltd. 2018.

More »»

2017

Sowmya V., Dr. Govind D., and Dr. Soman K. P., “Significance of contrast and structure features for an improved color image classification system”, 2017 IEEE International Conference on Signal and Image Processing Applications (ICSIPA). pp. 210-215, 2017.[Abstract]


In general, the three main modules of color image classification systems are: color-to-grayscale image conversion, feature extraction and classification. The color-to-grayscale image conversion is the important pre-processing step which must incorporate the significant and discriminative contrast and structure information in the converted grayscale images as in the original color image. All the existing techniques for color-to-grayscale image conversion preserves the significant contrast and structure information in the converted grayscale images in different manners. Hence, the present work is to analyze the significant and discriminative contrast and structure information preserved in the converted grayscale images using two different decolorization techniques called rgb2gray and singular value decomposition based color-to-grayscale image conversion (SVD) applied in the color image classification systems using the three different proposed features. The three different features for color image classification systems are proposed based on the combination of the existing dense SIFT features and the contrast & structure content computed using color-to-gray structure similarity index (C2G-SSIM) metric.

More »»

2017

Chippy Jayaprakash, Naveen Varghese Jacob, Renu .R.K., Sowmya V., and K. P. Soman, “Least Square Denoising in Spectral Domain for Hyperspectral Images”, Procedia Computer Science, 7th International Conference on Advances in Computing and Communications , ICACC-2017, vol. 115. Elsevier, Rajagiri School of Engineering & Technology, Kochi, Kerala, India, pp. 399-406, 2017.[Abstract]


Denoising is one of the fundamental pre-processing tasks in image processing that improves the quality of the information in the image. Processing of hyperspectral images requires high computational power and time. In this paper, a denoising technique based on least square weighted regularization in the spectral domain is proposed. The proposed technique is experimented on standard hyperspectral datasets and also, the performance of the proposed least square denoising in spectral domain is compared with least square weighted regularization in the spatial domain and total variation based denoising method. The obtained results in terms of computational time, Signal-to-Noise Ratio calculations and visual interpretation depicts that the proposed technique performs comparably better than the existing methods such as least square and total variation based hyperspectral image denoising. © 2017 The Author(s).

More »»

2017

Swarna .M, Sowmya V., and K. P. Soman, “Effect of Denoising on Dimensionally Reduced Sparse Hyperspectral Unmixing”, 7th International Conference on Advances in Computing and Communications , ICACC-2017, vol. 115. Elsevier, Rajagiri School of Engineering & Technology, pp. 391-398, 2017.[Abstract]


In hyperspectral images, spectral mixing occurs when objects lying beside each cannot be distinguished as different entities due to its low spatial resolution. Other hurdles in hyperspectral imaging are its huge dimension and noisy bands. In this paper, a new approach for spectral unmixing is presented where, the data is reduced dimensionally and, the bands eliminated during this are denoised using the existing denoising methods. Then, dataset with these bands is dimensionally reduced and their presence after reduction is validated using spectral unmixing methods. The effectiveness of this method is evaluated using parametric measures such as RMSE and classification accuracy. © 2017 The Author(s).

More »»

2017

Vishnu Pradeep V, Reshma R, Sowmya V., and K. P. Soman, “Comparative Analysis of Sparsity based and Kernel based algorithms for Hyperspectral Image Classification”, IEEE International Conference on Circuit, Power and Computing, ICCPCT-2017. Baselious Mathews II College of Engineering, Kerala, pp. 19-20, 2017.

2017

M .Srikanth, K.S. Gokul Krishnan, Sowmya V., and K. P. Soman, “Image Denoising based on Weighted Regularized Least Square Method”, IEEE International Conference on Circuit, Power and Computing, ICCPCT-2017. Baselious Mathews II College of Engineering, Kerala, pp. 19-20, 2017.

2017

Sowmya V., Ashwini B, Neethu Mohan, Shriya se, and K. P. Soman, “Performance Evaluation of Edge Feature Extracted using Sparse Banded Matrix Filter Applied for Face Recognition”, IEEE International Conference on Circuit, Power and Computing, ICCPCT-2017. Baselious Mathews II College of Engineering, Kerala, pp. 19-20, 2017.

2017

Sowmya V., Megha .P, and K. P. Soman, “Effect of Dynamic Mode Decomposition Based Dimension Reduction Technique on Hyperspectral Image Classification”, International Conference on Nextgen Electronic Technologies: Silicon to Software (ICNETS2. LNEE Springer Proceedings, VIT University, Chennai Campus, India, pp. 23-25 , 2017.

2017

Sowmya V., V. Ankarao, and K. P. Soman, “Fusion of panchromatic image with low-resolution multispectral images using Dynamic Mode Decomposition”, International Conference on Nextgen Electronic Technologies: Silicon to Software (ICNETS2). (LNEE Springer Proceedings), VIT University, Chennai Campus, India, pp. 23-25 , 2017.

2016

Swarna .M, Sowmya V., and K. P. Soman, “Effect of Dimensionality Reduction on Sparsity Based Hyperspectral Unmixing”, 8th International Conference on Soft Computing and Pattern Recognition (SoCPAR). School of Information Technology and Engineering, VIT University, Vellore, India, pp. 19-21 , 2016.

2016

Sreelekshmy Selvin, S .GAjay, B.Ganga Gowri, Sowmya V., and K. P. Soman, “l1 Trend Filter for Image Denoising”, Procedia Computer Science, 6th International Conference on Advances in Computing and Communications , ICACC-2016, vol. 93. Elsevier, Rajagiri School of Engineering and Technology, Kochi, India, pp. 495-502, 2016.[Abstract]


The major problem in digital image processing is the presence of unwanted frequencies(noise). In this paper ℓ1 trend filter is proposed as an image denoising technique. ℓ1-trend filter estimates the hidden trend in the data by formulating a convex optimization problem based on ℓ1 norm. The proposed method extends the application of ℓ1 trend filter from one dimensional signals to three dimensional color images. Here the filter is applied over the image in a cascade, initially filtering along the rows followed by filtering along the columns. This identifies the hidden image information from the noisy image resulting in a smooth or denoised image. The proposed method is compared with the wavelet denoising technique using the quality metrics Peak-Signal-to-Noise-Ratio(PSNR) and Structural Similarity Index(SSIM). © 2016 The Authors. Published by Elsevier B.V.

More »»

2016

Srivatsa .S, Sowmya V., and K. P. Soman, “Least Square Based Fast Denoising approach to Hyperspectral Imagery”, 4th International Conference on Advanced Computing, Networking and Informatics, ICACNI-2016, Centre for Computer Vision and Pattern Recognition, . NIT-Rourkela, pp. 22 - 24, 2016.

2016

N. John, Viswanath, A., Sowmya V., and Soman, K. P., “Analysis of various color space models on effective single image super resolution”, Advances in Intelligent Systems and Computing, International Symposium on Intelligent Systems Technologies and Applications (ISTA-15), co-located with 4th International Conference on Advances in Computing, Communications and Informatics, ICACCI 2015, vol. 384. Springer Verlag, Kochi, India, pp. 529-540, 2016.[Abstract]


Color models are used for facilitating the specification of colors in a standard way. A suitable color model is associated with every application based on color space. This paper mainly focuses on the analysis of effectiveness of different color models on single image scale-up problems. Single image scale-up aims in the recovery of original image, where the input image is a blurred and down- scaled version of the original one. In order to identify the effect of different color models on scale-up of single image applications, the experiment is performed with the single image scale-up algorithm on standard image database. The performance of different color models (YCbCr, YCoCg, HSV, YUV, CIE XYZ, Photo YCC, CMYK, YIQ, CIE Lab, YPbPr) are measured by quality metric called Peak Signal to Noise Ratio (PSNR). The experimental results based on the calculated PSNR values prove that YCbCr and CMYK color models give effective results in single image scale-up application when compared with the other available color models. © Springer International Publishing Switzerland 2016.

More »»

2016

N. Haridas, Aswathy, C., Sowmya V., and Soman, K. P., “Hyperspectral Image Denoising Using Legendre-Fenchel Transform for Improved Sparsity Based Classification”, Advances in Intelligent Systems and Computing, International Symposium on Intelligent Systems Technologies and Applications (ISTA-15), co-located with 4th International Conference on Advances in Computing, Communications and Informatics, ICACCI 2015, vol. 384. Springer, SCMS School of Engineering, Aluva, Kochi , pp. 521–528, 2016.[Abstract]


A significant challenge in hyperspectral remote sensing image analysis is the presence of noise, which has a negative impact on various data analysis methods such as image classification, target detection, unmixing etc. In order to address this issue, hyperspectral image denoising is used as a preprocessing step prior to classification. This paper presents an effective, fast and reliable method for denoising hyperspectral images followed by classification based on sparse representation of hyperspectral data. The use of Legendre-Fenchel transform for denoising is an effective spatial preprocessing step to improve the classification accuracy. The main advantage of Legendre-Fenchel transform is that it removes the noise in the image while preserving the sharp edges. The sparsity based algorithm namely, Orthogonal Matching Pursuit (OMP) is used for classification. The experiment is done on Indian Pines data set acquired by Airborne Visible Infrared Imaging Spectrometer (AVIRIS) sensor. It is inferred that the denoising of hyperspectral images before classification improves the Overall Accuracy of classification. The effect of preprocessing using Legendre Fenchel transformation is shown by comparing the classification results with Total Variation (TV) based denoising. A statistical comparison of the accuracies obtained on standard hyperspectral data before and after denoising is also analysed to show the effectiveness of the proposed method. The experimental result analysis shows that for 10%% training set the proposed method leads to the improvement in Overall Accuracy from 83.18%% to 91.06%%, Average Accuracy from 86.17%% to 92.78%% and Kappa coefficient from 0.8079 to 0.8981

More »»

2016

N. Nechikkat, Sowmya V., and Soman, K. P., “Variational Mode Feature-Based Hyperspectral Image Classification”, Advances in Intelligent Systems and Computing, Second International Conference on Computer and Communication Technologies (IC3T-2015), vol. 380. Springer, CMR Technical Campus, Hyderabad , pp. 365-373, 2016.[Abstract]


Hyperspectral image analysis is considered as a promising technology in the field of remote sensing over the past decade. There are various processing and analysis techniques developed that interpret and extract the maximum information from high-dimensional hyperspectral datasets. The processing techniques significantly improve the performance of standard algorithms. This paper uses variational mode decomposition (VMD) as the processing algorithm for hyperspectral data scenarios followed by classification based on sparse representation. Variational Mode Decomposition decomposes the experimental data set into few different modes of separate spectral bands, which are unknown. These modes are given as raw input to the classifier for performance analysis. Orthogonal matching pursuit (OMP), the sparsity-based algorithm is used for classification. The proposed work is experimented on the standard dataset, namely Indian pines collected by the airborne visible/infrared imaging spectrometer (AVIRIS). The classification accuracy obtained on the hyperspectral data before and after applying Variational Mode Decomposition was analyzed. The experimental result shows that the proposed work leads to an improvement in the overall accuracy from 84.82 to 89.78 %, average accuracy from 85.03 to 89.53 % while using 40 % data pixels for training.

More »»

2016

L. S. Kiran, Sowmya V., and Soman, K. P., “Dimensionality reduced recursive filter features for hyperspectral image classification”, Advances in Intelligent Systems and Computing, Second International Conference on Computer and Communication Technologies (IC3T -2015), vol. 380. AISC Springer Series(SCOPUS, ISI proceedings), CMR Technical Campus, Hyderabad, pp. 557-565, 2016.[Abstract]


Dimensionality reduction techniques have been immensely used in hyperspectral image classification tasks and is still a topic of great interest. Feature extraction based on image fusion and recursive filtering (IFRF) is a recent work which provides a framework for classification and produces good classification accuracy. In this paper, we propose an alternative approach to this technique by employing an efficient preprocessing technique based on average interband blockwise correlation coefficient followed by a stage of dimensionality reduction. The final stages involve recursive filtering and support vector machine (SVM) classifier. Our method highlights the utilization of an automated procedure for the removal of noisy and water absorption bands. Results obtained using experimentation of the proposed method on Aviris Indian Pines database indicate that a very low number of feature dimensions provide overall accuracy around 98%. Four different dimensionality reduction techniques (LDA, PCA, SVD, wavelet) have been employed and notable results have been obtained, especially in the case of SVD (OA = 98.81) and wavelet-based approaches (OA = 98.87). © Springer India 2016.

More »»

2016

S. Moushmi, Sowmya V., and Soman, K. P., “Empirical Wavelet Transform for Multifocus Image Fusion”, International Conference on Soft Computing Systems (2015), Advances in Intelligent Systems and Computing, vol. 397. AISC Springer Series, Noorul Islam Centre for Higher Education, Kumaracoil; India, pp. 257-263, 2016.[Abstract]


Image fusion has enormous applications in the fields of satellite imaging, remote sensing, target tracking, medical imaging, and much more. This paper aims to demonstrate the application of empirical wavelet transform for the fusion of multifocus images incorporating the simple average fusion rule. The method proposed in this paper is experimented on benchmark datasets used for fusing images of different focuses. The effectiveness of the proposed method is evaluated across the existing techniques. The performance comparison of the proposed method is done by visual perception and assessment of standard quality metrics which includes root mean squared error, relative average spectral error, universal image quality index, and spatial information. The experimental result analysis shows that the proposed technique based on the empirical wavelet transform (EWT) outperforms the existing techniques.

More »»

2016

P. G. Mol, Sowmya V., and Soman, K. P., “Performance enhancement of minimum volume-based hyperspectral unmixing algorithms by empirical wavelet transform”, International Conference on Soft Computing Systems (ICSCS 2015), Advances in Intelligent Systems and Computing, vol. 397. AISC Springer Series, Noorul Islam Centre for Higher Education, Kumaracoil; India, pp. 251-256, 2016.[Abstract]


Hyperspectral unmixing of data has become one of the essential processing steps for crop classification. The endmembers to be extracted from the data are statistically dependent either in the linear or nonlinear form. The primary focus of this paper is on the effect of empirical wavelet transform (EWT) on hyperspectral unmixing algorithms based on the geometrical minimum volume approaches. The proposed method is experimented on the standard hyperspectral dataset, namely Cuprite. The performance analysis of proposed approach is eval- uated based on the standard quality metric called root mean square error (RMSE). The experimental result analysis shows that our proposed technique based on EWT improves the performance of hyperspectral unmixing algorithms based on the geometrical minimum volume approaches.

More »»

2015

J. M Babu, Sowmya V., and Soman, K. P., “Fast Fourier Transform and Nonlinear Circuits Based Approach for Smart Meter Data Security”, International Conference onSmart Grid Technologies (ICSGT’15). Elsevier Procedia Technology, Amrita School of Engineering, Coimbatore, pp. 287–294, 2015.[Abstract]


Smart meters take measurements at minute intervals of time. The energy disaggregation of smart meter data provides accurate information regarding the electricity consumption but also reveals detailed information pertaining to the customer. Load signatures-that is unique for every appliance can be used to declare the behavioral pattern of the customers such as their sleep-wake cycles, usage period of various appliances, time during which the house is empty and by far, which particular channel is being viewed on the television. The smart meter data is transmitted to the utility company at regular intervals of time over the internet. Data that has such potential to access private information can be tampered with, by an undesirable third party during transmission. This is a vital privacy threat to the customer and has hyped the researches pertaining to smart meter data security in the recent times. To assure security, we introduce an algorithm that aims at encryption of data prior to transmission. The algorithm employs the FFT algorithm and various nonlinear systems to generate chaotic signals. The obtained chaotic signal is amalgamated with a transformed version of the smart meter data and securely transmitted over a suitable network. The proposed algorithm is tested on a real household's power signal data.

More »»

2015

C. Aswathy, Sowmya V., and K. P. Soman, “Hyperspectral Image Denoising using Low Pass Sparse Banded Filter Matrix for Improved Sparsity Based Classification”, Second International Symposium on Computer Vision and the Internet (VisionNet’15). Elsevier Procedia Computer Science Journal, SCMS School of Engineering, Aluva, Kochi, pp. 26–33, 2015.[Abstract]


The recent advance in sensor technology is a boon for hyperspectral remote sensing. Though Hyperspectral images (HSI) are captured using these advanced sensors, they are highly prone to issues like noise, high dimensionality of data and spectral mixing. Among these, noise is the major challenge that affects the quality of the captured image. In order to overcome this issue, hyperspectral images are subjected to spatial preprocessing (denoising) prior to image analysis (Classification). In this paper, authors discuss a sparsity based denoising strategy which uses low pass sparse banded filter matrices (AB filter) to effectively denoise each band of HSI. Both subjective and objective evaluations are conducted to prove the efficiency of the proposed method. Subjective evaluations involve visual interpretation while objective evaluations deals with the computation of quality matrices such as Peak Signal to Noise Ratio (PSNR) and Structural Similarity (SSIM) index at different noise variance. In addition to these, the denoised image is followed by a sparsity based classification using Orthogonal Matching Pursuit (OMP) to evaluate the effect of various denoising techniques on classification. Classification indices obtained without and with applying preprocessing are compared to highlight the potential of the proposed method. The experiment is performed on standard Indian Pines Dataset. By using 10% of training set, a significant improvement in overall accuracy (84.21%) is obtained by the proposed method, compared to the other existing techniques.

More »»

2015

N. Haridas, Sowmya V., and Soman, K. P., “Comparative Analysis of Scattering and Random Features in Hyperspectral Image Classification”, Second International Symposium on Computer Vision and the Internet (VisionNet’15). Elsevier Procedia Computer Science Journal, SCMS School of Engineering, Aluva, Kochi, pp. 307 – 314, 2015.[Abstract]


Hyperspectral images (HSI) contains extremely rich spectral and spatial information that offers great potential to discriminate between various land cover classes. The inherent high dimensionality and insufficient training samples in such images introduces Hughes phenomenon. In order to deal with this issue, several preprocessing techniques have been integrated in processing chain of HSI prior to classification. Supervised feature extraction is one such method which mitigates the curse of dimensionality induced by Hughes effect. In recent years, new strategies for feature extraction based on scattering transform and Random Kitchen Sink have been introduced, which can be used in context of hyperspectral image classification. This paper presents a comparative analysis of scattering and random features in hyperspectral image classification. The classification is performed using simple linear classifier such as Regularized Least Square (RLS) accessed through Grand Unified Regularized Least Squares (GURLS) library. The proposed approach is tested on two standard hyperspectral datasets namely, Salinas-A and Indian Pines subset scene captured by NASAs AVIRIS sensor (Airborne Visible Infrared Imaging Spectrometer). In order to show the effectiveness of proposed method, a comparative analysis is performed based on feature dimension, classification accuracy measures and computational time. From the comparative assessment, it is evident that classification using random features achieve excellent classification results with less computation time when compared with raw pixels(without feature extraction) and scattering features for both the datasets.

More »»

2015

Sowmya V., Neethu Mohan, and Soman, K. P., “Edge Detection Using Sparse Banded Filter Matrices”, Second International Symposium on Computer Vision and the Internet (VisionNet’15). Elsevier Procedia Computer Science Journal, SCMS School of Engineering, Aluva, Kochi , pp. 10–17, 2015.[Abstract]


Edges are intensity change, which occur on the boundary between two regions in an image. Edges can be used as feature descriptors of an object. Hence, edge detection plays an important role in computer vision applications. This paper presents the application of sparse banded filter matrices in edge detection. The filter design is formulated in terms of banded matrices. The sparsity property of the designed filter leads to efficient computation. In our proposed method, we applied sparse banded high-pass filter row-wise and column-wise to extract the vertical and the horizontal edges of the image respectively. The proposed technique is experimented on standard images and the results are compared with the state-of-the-art methods. The visual comparison of the experimental results shows that the proposed approach for edge extraction based on sparse banded filter matrices produces result comparable to the existing methods. The advantage of the proposed approach is that the continuous edges are attained without any parameter tuning.

More »»

2015

N. Haridas, Sowmya V., and Soman, K. P., “Hyperspectral image classification using Random Kitchen Sink and Regularized Least Squares”, 4th IEEE International Conference on Communications and Signal Processing (ICCSP), 2015 . IEEE, Adhiparasakthi Engineering College , Melmaruvathur, pp. 1665 - 1669, 2015.[Abstract]


Kernel machines has gained considerable attention in the field of remote sensing for solving machine learning tasks, particularly in classification. Despite the fact that, kernel based methods produce comparatively better performance than traditional learning approaches, they are computationally expensive and requires large memory storage. In recent years, the concept of random features was introduced in kernel machines to solve this problem. This paper presents a new framework for hyperspectral image classification using Random Kitchen Sink (RKS) and Regularized Least Squares (RLS) classifier. The study shows that randomized features are economically powerful tool for hyperspectral image classification which produces significant improvement in classification accuracy. The proposed approach is tested on two standard hyperspectral datasets namely, Salinas-A and Indian Pines subset scene acquired by Airborne Visible Infrared Imaging Spectrometer (AVIRIS) sensor. A statistical comparison of the accuracies obtained on standard hyperspectral data with and without using Random Kitchen Sink algorithm for Regularized Least Squares classifier is analysed to show the effectiveness of the proposed method. The experimental results shows that the proposed method leads to improvement in Overall Accuracy from 85.12% to 98.58% and Kappa Coefficient from 0.8154 to 0.9822 for Salinas-A data scene. The analysis of Indian Pines subset scene shows that the proposed work results in significant improvement in Overall Accuracy from 62.76% to 93.79% and Kappa Coefficient from 0.5061 to 0.9160. The result analysis proves that random features of hyperspectral data as input to a standard linear classifier without the aid of any preprocessing produces better classification accuracy.

More »»

2015

K. Harikumar, Athira, S., Nair, Y. C., Sowmya V., and Soman, K. P., “ADMM based algorithm for spike and smooth signal separation using over-complete dictionary”, 4th IEEE International Conference on Communication and Signal Processing-ICCSP'15. IEEE, Adhiparasakthi Engineering College , Melmaruvathur , pp. 1617 - 1622, 2015.[Abstract]


In signal processing, many a time people deal with smooth stationary signals mixed with sharp spikes and most of the time their analysis demands separation of the smooth and spike elements. In this paper, we propose a methodology for this kind of separation based on the well-known notion of using an over complete dictionary to define an underdetermined system of linear equations and picking out its sparsest solution. The Alternating Direction Method of Multipliers (ADMM) framework is proposed for formulating and solving this optimization problem. The study of the performance of the algorithm with respect to certain signal parameters is also included. Performance of the algorithm was tested for different sparsity values at different signal energies and the results are reported. Cases involving group sparsity is also studied.

More »»

2015

C. Aswathy, Sowmya V., Gandhiraj R., and Soman, K. P., “Hyperspectral image denoising using legendre Fenchel Transformation for improved Multinomial Logistic Regression based classification”, 4th IEEE International Conference on Communications and Signal Processing (ICCSP), 2015 . IEEE, Adhiparasakthi Engineering College , Melmaruvathur, pp. 1670 - 1674, 2015.[Abstract]


The abundant spectral and spatial information in the hyperspectral images (HSI) are largely used in the field of remote sensing. Though there are highly sophisticated sensors to capture the hyperspectral imagery, they suffer from issues like hyperspectral noise and spectral mixing. The major challenges encountered in this field, demands the use of preprocessing techniques prior to hyperspectral image analysis. In this paper, we discuss the effective role of denoising by Legendre Fenchel Transformation (LFT) as a preprocessing method to improve the classification accuracy. Experimental time analysis shows that the computational efficiency of the proposed method is highly effective when compared with the existing preprocessing methods. LFT is based on the concept of duality which makes it a fast and reliable denoising strategy to effectively reduce the noise present in each band of the hyperspectral imagery, without losing much of the edge information. The denoising is performed on standard AVIRIS Indian Pines dataset. The performance of LFT denoising is evaluated by analysing the classification accuracy assessment measures. The denoised image is subjected to hyperspectral image classification using Multinomial Logistic Regression which learns the posterior probability distributions of each class. The potential of the proposed method is proved by the mean classification accuracy obtained experimentally without any post processing technique (94.4%), which is better when compared with the accuracies acquired by existing preprocessing techniques like Total Variation denoising and wavelet based denoising.

More »»

2014

Sowmya V., “Comparison of Performance Analysis of Image Classification by K-Means and Bisecting K-Means”, 4th National Conference on Advanced Trends in Information and Computing Sciences (NCATICS’14). KCT, Coimbatore, 2014.

2013

M. Suchithra, Sukanya, P., Prabha, P., Sikha, O. K., Sowmya V., and Dr. Soman K. P., “An experimental study on application of orthogonal matching pursuit algorithm for image denoising”, 2013 IEEE International Multi Conference on Automation, Computing, Control, Communication and Compressed Sensing, iMac4s 2013. IEEE, Kochi, Kerala, pp. 729-736, 2013.[Abstract]


<p>Signal or image reconstruction has now become a common task in many applications. According to linear algebra perspective, the number of measurements made or the number of samples taken for reconstruction must be greater than or equal to the dimension of signal or image. Also reconstruction follows the Shanon's sampling theorem which is based on the Nyquist sampling rate. The reconstruction of a signal or image using the principle of compressed sensing is an exception which makes use of only few number of samples which is below the sampling limit. Compressive sensing also known as sparse recovery aims to provide a better data acquisition and reduces computational complexities that occur while solving problems. The main goal of this paper is to provide clear and easy way to understand one of the compressed sensing greedy algorithm called Orthogonal Matching Pursuit (OMP). The OMP algorithm involves the concept of overcomplete dictionary that is formulated based on different thresholding methods. The proposed method gives the simplified approach for image denoising by using OMP only. The experiment is performed on few standard image data set simulated with different types of noises such as Gaussian noise, salt and pepper noise, exponential noise and Poisson noise. The performance of the proposed method is evaluated based on the image quality metric, Peak Signal-to-Noise Ratio (PSNR). © 2013 IEEE.</p>

More »»

2013

G. Aarthy, Amitha, P. L., Krishnan, T., Pillai, G. S., Sowmya V., and Dr. Soman K. P., “A comparative study of spike and smooth separation from a signal using different overcomplete dictionary”, 2013 IEEE International Multi Conference on Automation, Computing, Control, Communication and Compressed Sensing, iMac4s 2013. IEEE, Kochi, Kerala, pp. 590-595, 2013.[Abstract]


Most of the natural signals are complex and are highly time varying, since they are non stationary in nature. In this paper, a comparative study for separating spikes and smooth signal components from a non-stationary signal are performed based on different overcomplete dictionaries. The experiment is evaluated using the sparse representation with different bases such as the Discrete Cosine Transform (DCT), Walsh-Hadamard, Orthogonal and Biorthogonal wavelet basis. The primary focus of this paper is to use L1 minimization for retrieving the smooth and spikes component of the signal using different overcomplete dictionary. The experimental results reveals out the dictionary that delivers a better separation without distorting temporal and spectral characteristics. © 2013 IEEE.

More »»

2013

B. G. Gowri, Hariharan, V., Thara, S., Sowmya V., Kumar, S. S., and Dr. Soman K. P., “2D Image data approximation using Savitzky Golay filter - Smoothing and differencing”, 2013 IEEE International Multi Conference on Automation, Computing, Control, Communication and Compressed Sensing, iMac4s 2013. IEEE, Kochi, Kerala, pp. 365-371, 2013.[Abstract]


Smoothing and differencing is one of the major important and necessary step in the field of signal processing, image processing and also in the field on analytical chemistry. The search for an efficient image smoothing and edge detection method is a challenging task in image processing sector. Savitzky Golay Filters are one among the widely used filters for analytical chemistry. Even though they have exceptional features, they are rarely used in the field of image processing. The designed filter is applied for image smoothing and a mathematical model based on partial derivatives is proposed to extract the edges in images. The smoothing technique of SG filter offers an extremely simple aid in extracting the edge information. An approach using SG filter which can be applied in preserving edge information is one of the major tasks involved in the classification process in the domain of Optical Character Recognition. The paper is focused on designing the Savitzky Golay filter by using the concepts of linear algebra. The main objective of the paper is to portray a clear cut idea about Savitzky Golay filter and to study the design of Savitsky Golay filters based on the concepts of Linear Algebra. © 2013 IEEE.

More »»

2012

P. K. Indukala, Lakshmi, K., Sowmya V., and Dr. Soman K. P., “Implementation of ℓ 1 magic and one bit compressed sensing based on linear programming using excel”, International Conference on Advances in Computing and Communications, ICACC 2012. IEEE, Kochi, Kerala, pp. 69-72, 2012.[Abstract]


<p>Compressed sensing helps in the reconstruction of sparse or compressible signals from small number of measurements. The sparse representation has great importance in modern signal processing. The main objective is to provide a strong understanding of the concept behind the theory of compressed sensing by using the key ideas from linear algebra. In this paper, the concept of compressed sensing is explained through an experiment formulated based on linear programming and solved using l1 magic and One bit compressed sensing methods in Excel. © 2012 IEEE.</p>

More »»

2012

R. Anand, Prabha, P., Sikha, O. K., Suchithra, M., Dr. Soman K. P., and Sowmya V., “Visualization of OFDM using Microsoft Excel spreadsheet in Linear Algebra Perspective”, International Conference on Advances in Computing and Communications, ICACC 2012. IEEE, Kochi, Kerala, pp. 58-64, 2012.[Abstract]


Orthogonal Frequency Division Multiplexing (OFDM) is one of the leading technology that is ruling the communication field. But unfortunately, it is shrouded in mystery. A good knowledge in Linear Algebra is required to appreciate the technology in a better way. So the work focuses on explaining OFDM system from linear algebra point of view. Also, OFDM model communication system is simulated using Excel which makes ease for anyone experiment with OFDM and understand the underlying principle. The paper aims to provide strong foundation on the concept behind OFDM without the need of having much knowledge in electronics field. © 2012 IEEE.

More »»

2012

P. Prabha, Sikha, O. K., Suchithra, M., Sukanya, P., Sowmya V., and Dr. Soman K. P., “Computation of continuous wavelet transform using microsoft excel spreadsheet”, Proceedings - 2012 International Conference on Advances in Computing and Communications, ICACC 2012. IEEE, Kochi, Kerala, pp. 73-77, 2012.[Abstract]


Wavelet theory has become an essential and significant tool for signal and image processing applied in the analysis of various real time signals. It is thus necessary to include wavelet transform and its application in multifractal analysis as a part of the engineering curriculum. In this paper, we present simple and effective way of computing Continuous Wavelet Transform (CWT) using Microsoft Excel Spreadsheet which serves as an user friendly mathematical tool for beginners. The motivation of this paper is to prove the computational power of excel, using which students can have better understanding of the basic concept behind the computation of Continuous Wavelet Transform. The plot of Continuous Wavelet Transform of Brownian signal computation in Excel is compared with that of the result in the Matlab Toolbox. The singularities present in the signal can be inferred from the wavelet modulus maxima plot. The visual interpretation proves that Excel tool provides computational power comparable to that of the Matlab software. The codes for the implementation of CWT in Excel are available on nlp.amrita.edu:8080/sisp/wavelet/cwt/cwt.xlsm, nlp.amrita.edu:8080/sisp/wavelet/cwt/modmax.xlsm, nlp.amrita.edu:8080/sisp/ wavelet/cwt/thermo.xlsm © 2012 IEEE.

More »»

2011

Sowmya V., “A Decision Tree Based Land Cover Image Classification Using Color Space and Texture”, IEEE International Conference on Computational Intelligence and Computing Research. IEEE, Cape Institute of Technology, Levengipuram, Kanyakumari, India, 2011.

2011

B. D Bhushan, Sowmya V., M Manikandan, S., and Soman, K. P., “An effective pre-processing algorithm for detecting noisy spectral bands in hyperspectral imagery”, International Symposium on Ocean Electronics, SYMPOL 2011. IEEE, Kochi, pp. 34 - 39, 2011.[Abstract]


In this paper, we present an effective pre-processing algorithm for band selection approach which is an essential task in hyperspectral image analysis. The pre-processing algorithm is developed based on the average inter-band block-wise correlation coefficient measure and a simple thresholding strategy. Here, the threshold parameter is found based on the standard deviation of the average inter-band block-wise correlation coefficients. The performance of the proposed algorithm is validated using the standard hyperspectral database created by the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) sensor. By comparing the detected bands with ground-truth annotations, we observed that the proposed algorithm identifies the noisy and water absorption bands in the high-dimensional hyperspectral images. The proposed algorithm achieves the classification accuracy of 94.73%.

More »»

2010

B. D. Bhushan, Sowmya V., and Dr. Soman K. P., “Super resolution blind reconstruction of low resolution images using framelets based fusion”, International Conference on Recent Trends in Information, Telecommunication, and Computing ITC 2010. IEEE, Kochi, Kerala, pp. 100-104, 2010.[Abstract]


In this paper, we propose a fusion technique based on framelets to obtain super resolution image from sub-pixel shifted, noisy, blurred low resolution images. This method has high advantages over all existing methods. A Tight frame filter bank provides symmetry and has a redundancy that allows for approximate shift invariance which leads to clear edges, high spatial information with effective denoising which was lacked in critically sampled discrete wavelet transform. They are also shorter and results in smoother scaling and wavelet functions. The reconstructed super resolution image obtained by this technique has high peak signal to noise ratio (PSNR) and low mean square error (MSE) than that obtained by wavelet based fusion method, which is evident through the experimental results. © 2010 IEEE.

More »»

Publication Type: Newspaper Article

Year of Publication Title

2013

Sowmya V., “Voice Your Views -5 –Should the death penalty be abolished?”, The Hindu Metroplus, CBE Edition , Coimbatore, 2013.

2013

Sowmya V., “Voice Your Views -4- What do you think should be censored from movies?”, The Hindu Metroplus, CBE Edition, Coimbatore, 2013.

2013

Sowmya V., “Voice Your Views -3- Favorite yet unusual getaways at Coimbatore”, The Hindu Metroplus, CBE Edition, Coimbatore, 2013.

2013

Sowmya V., “Voice Your Views -2 – Is Coimbatore Safe for Women?”, The Hindu Metroplus, CBE Edition, Coimbatore, 2013.

2012

Sowmya V., “Voice Your Views -1 –Is there a way to segregate the volume of garbage we generate?”, The Hindu Metroplus, CBE Edition, Coimbatore, 2012.