Qualification: 
Ph.D
b_rajathilagam@cb.amrita.edu

Dr. B. Rajathilagam is an Associate Professor, Department of Computer Science and Engineering, Amrita Vishwa Vidyapeetham, Coimbatore campus. She completed her Ph. D. in 2013 under Centre for Excellence in Computational Engineering, Amrita Vishwa Vidyapeetham, Coimbatore campus. Amrita Vishwa Vidyapeetham is a Zonal Partner in leadinindia.ai initiative. Dr. B. Rajathilagam is in charge of SiGMA research lab.

Projects

Sl. No Agency Project Title Status
1 ISRO-RESPOND Automation of Transport and Building Feature Extraction using Deep Learning with Super-Resolution Enhancement of Satellite Imagery Ongoing

Research Projects

  1. G-lets Signal Processing for Functional Approximation in Neural Networks
  2. Artificial Intelligence Systems for Intelligent Transportation Infrastructure on Indian Roads
  3. Graph Signal Processing Filters
  4. Geo-spatial 3D Modelling and Augmented Reality for Smart City Planning and Design
  5. Real-Time Traffic Incident Detection and Prediction with Machine Learning Tools on Indian Road Networks
  6. Multi-attribute Utility Model to Scrutinize Patient Data using Machine Learning and Deep Learning for Medical Decision Support System in Remote Healthcare

Ph. D. Students

  • Mrs. Vidhya Sudevan
  • Mrs. Parvathy Dharmarajan
  • Mrs. Sneha Kandacharam

Master’s Student Research Projects

  1. Object Shapes from Regular Curves through Sparse Representation

Undergraduate Student Research Projects

  1. Electrolaryngeal Speech Improvement through Phoneme Replacement
  2. Tracking and Radar Sensor models for Road Transport Safety Systems
  3. Road Traffic Incident Detection
  4. Obstacle Detection with Radar Image Sensing
  5. Facial Expression Improvement Recommendations for Individuals

Publications

Publication Type: Journal Article

Year of Publication Publication Type Title

2018

Journal Article

B. Nagarajan and Dr. Rajathilagam B., “Object Shapes from Regular Curves through Sparse Representations”, Procedia Computer Science, vol. 133, pp. 635 - 642, 2018.[Abstract]


Robotics coupled with image processing algorithms have led to more advanced control systems in varied applications. Object detection systems are used in multiple places like warehouses where detecting the object faster, makes efficient warehouse management. In this work, an object detection algorithm which uses shape as an object representation is presented. The algorithm uses semi-local contour grouping which is based on gradient maps generated from G-Let filters. The algorithm offers a computationally simpler solution as the original gradient map is reduced to a sparse representation using different thresholding techniques. The sparse representation of the object is modeled as a neighbor graph and the shape is constructed using alpha-hull of the nearest neighbor graph. The construction of a gradient map and working with its local curvatures simplifies the shape detection problem, from building hierarchical structures of local-to-global contour relationships to nearest neighbor calculations on a graph and fixing their boundary. More »»

2018

Journal Article

S. Suman and Dr. Rajathilagam B., “Outlier detection in time-series data: Specific to nearly uniform signals from the sensors”, Lecture Notes in Computational Vision and Biomechanics, vol. 28, pp. 697-704, 2018.[Abstract]


In this paper, the complexity of detecting an outlier has been shown. The importance of an outlier has been presented with the need to interpret these outliers. The sensors collect data with certain sampling time period and these data are stored which contribute to the huge database. These sensors can be electrocardiogram sensor which monitors electrical and muscular functions of the heart, they can be pollution monitoring stations at the airports, they can be heat sensors in a building, and they can be flight data recorders (FDR) and so on. Sometimes, these sensors miss to detect the signal due to some technical fault and hence the output is “Not Available (NA)”. These NA time stamps create unnecessary problems which lead to unwanted outputs when the data is processed. In this paper, an algorithm is presented which replaces these NA values with most probable values. When the data is ready with all NA values removed, the data is processed for detecting the outlier. In this paper, an outlier is being detected by analyzing the signal in the frequency domain along with the mean in the time domain. A large data set is divided into equal sized blocks. Each block is then converted to its frequency domain and mean is calculated in the time domain. These two parameters are considered to detect any outlier in the block. This approach removes the complexity in the algorithm without compromising in the efficiency of detecting an outlier. Hence, a large database of values is processed in relatively less time with appreciable accuracy. © 2018, Springer International Publishing AG.

More »»

2017

Journal Article

Dr. Rajathilagam B. and Dr. Murali Rangarajan, “Reducing the cold-user and cold-item problem in recommender system by reducing the Spectral representation of principal components in signals and images using G-lets decomposition of sub bands”, IEEE Region 10 Annual International Conference, Proceedings TENCON, pp. 3809 -3812., 2017.

2017

Journal Article

Dr. Rajathilagam B. and Rangarajan, M., “Edge detection using G-lets based on matrix factorization by group representations”, Pattern Recognition, vol. 67, pp. 1-15, 2017.[Abstract]


A new edge detection technique using transformation groups based G-lets filters is proposed in this paper. Discretizing gradients seem to produce discontinuity in classic edge detectors. No particular filter is capable of identifying meaningful edges at all scales and it increases computations with a multiscale approach. It is a challenge to get localized edges without spurious ones due to noise and integrate the obtained edges into meaningful object boundaries. Without breaking edge continuity and strictly localizing edges requires that filters do not blur the image during preprocessing. G-lets filters are found to be capable of performing well in most type of images including natural, noisy, low resolution and synthetic. In this paper, an edge detection algorithm using G-lets filters which are built by direct factorization of linear transformation matrices using irreducible representations is proposed. A multiresolution approach is shown to enhance the possibility of detecting faint edges. An edge tracing algorithm is presented to produce the edge image. The computational cost involved is comparatively lesser than existing filters. It is found that the geometries in the original image are preserved in the edge image. The edge tracing algorithm is capable of constructing object boundaries without the inner textures in a way that is not completely dependent on intensity thresholding. G-lets filters and the edge operator is found to be a promising algorithm for drastically bringing down the computations needed for realtime applications. The results are compared with BSDS500 boundary detection dataset using pb and global pb detectors. © 2017 Elsevier Ltd.

More »»

2015

Journal Article

S. C, Sandeep, R., and Dr. Rajathilagam B., “Comparison of Image Segmentation Algorithms”, International Journal of Applied Engineering Research (IJAER), 2015.

2015

Journal Article

R. Shwetha and Dr. Rajathilagam B., “Super resolution of mammograms for breast cancer detection”, International Journal of Applied Engineering Research, vol. 10, pp. 21453-21465, 2015.[Abstract]


Mammography has been the most popular method for the early detection of the breast cancer. Due to low contrast of mammograms typical diagnostic signs such as masses and micro calcification are difficult to detect. So to create a high resolution mammogram super resolution (SR) technique can be used. This technique will make a high resolution image from a series of low resolution images of the same scene. A novel algorithm with interpolation for super resolution reconstruction has been proposed here. It has taken a interpolation technique that preserves edges without introducing any artifact. This also avoids pixilation, over smoothing and blurring of images. In our method we have used denoising, deblurring and registration technique to improve the quality of low resolution images and fused them to produce a higher resolution image. The proposed algorithm is a hybrid of bilinear interpolation and FCBI method with edge detecting criteria. © Research India Publications.

More »»

2012

Journal Article

Dr. Rajathilagam B., Dr. Murali Rangarajan, and P, S. K., “G-Lets: A New Signal Processing Algorithm”, International Journal of Computer Applications, vol. 37 , no. 6, pp. 1-7, 2012.[Abstract]


Different signal processing transforms provide us with unique decomposition capabilities. Instead of using specific transformation for every type of signal, we propose in this paper a novel way of signal processing using a group of transformations within the limits of Group theory. For different types of signal different transformation combinations can be chosen. It is found that it is possible to process a signal at multiresolution and extend it to perform edge detection, denoising, face recognition, etc by filtering the local features. For a finite signal there should be a natural existence of basis in it’s vector space. Without any approximation using Group theory it is seen that one can get close to this finite basis from different viewpoints. Dihedral groups have been demonstrated for this purpose.

More »»

2012

Journal Article

Dr. Rajathilagam B., Dr. Murali Rangarajan, and Soman, K. P., “G-Lets: Signal Processing Using Transformation Groups”, vol. arXiv:1201.2995v1, 2012.

2012

Journal Article

Dr. Rajathilagam B., Rangarajan, M., and Soman, K. P., “Frequency analysis of signals and images using G-Lets”, International Journal of Imaging and Robotics, vol. 8, pp. 30-48, 2012.[Abstract]


This paper presents a method of frequency analysis for discrete signals using G-lets. From a group of transformations and representation theory, a finite basis of the signal space is obtained. The projections of the signal onto this basis are called G-lets. G-lets, due to the nature of transformations used, contain oscillations in such a manner that the difference between consecutive G-let coefficients is proportional to the local frequency. The signal frequency, in turn, is proportional to the difference in amplitude of the signal at any point. A dilation operation is defined to capture the frequencies without use of a windowing function, by highlighting the highest frequency of the existing signal in a G-let. Considering features of a signal as a combination of frequencies, feature extraction of 1-D signals and images are examined. The beginning and end of each feature are identified by the spread of low frequencies in the neighborhood of a high frequency. Results are demonstrated using dihedral groups, for simple 1-D signals, an ECG signal, and 'Lena' image. A qualitative comparison is provided with wavelets and Fourier analysis. © 2012 by IJIR (CESER Publications).

More »»

Publication Type: Conference Proceedings

Year of Publication Publication Type Title

2018

Conference Proceedings

A. A. V, S, R., and Dr. Rajathilagam B., “Electrolaryngeal Speech Improvement through Phoneme Enhancement”, Workshop on Speech Processing for Voice, Speech and Hearing Disorders (WSPD) 2018, INTERSPEECH 2018, Mysore, India. . 2018.

2015

Conference Proceedings

Dr. Rajathilagam B., Murali, R., and Balaji, B., “Dynamic Context-specific User Profiles by Regression Modelling”, Proceedings of GSTF CGAT. 2015.

2014

Conference Proceedings

Dr. Rajathilagam B., XC, D. P., and Jayaraj, B., “Techniques for Secure Communication in Emerging Wireless Ubiquitous Networks”, The International conference on High Power Computing (HiPC), Bangalore, India, 2004. 2014.

2005

Conference Proceedings

Dr. Rajathilagam B., Jayaraj, B., and Rangan, D. P. Venkat, “Naming, Location tracking, Synchronizing and Aggregating Wireless Sensor Networks”, In the proceedings of SPIE, Optics East 2005, Boston, USA., vol. 5993. pp. 5993 - 17, 2005.

Publication Type: Conference Paper

Year of Publication Publication Type Title

2017

Conference Paper

P. Dharmarajan and Dr. Rajathilagam B., “Cloud-Based e-Healthcare Service System Design for On-Demand Affordable Remote Patient Care”, in Computer Communication, Networking and Internet Security, Singapore, 2017.[Abstract]


In this paper, a low-cost system design for e-healthcare service including software and hardware components is presented. Vital signs of the human body are measured from the patient location and shared with a registered medical professional for consultation. Temperature and heart rate are the major signals obtained from a patient for the initial build of the system. Data is sent to a cloud server where processing and analysis is provided for the medical professional to analyze. Secure transmission and dissemination of data through the cloud server is provided and an authentication system, a secure storage server for the cloud is included for control by the patient from a smart phone. A prototype of the system is built with all the components for testing and the challenges of implementing the system in real time have been discussed. More »»

2017

Conference Paper

Dr. Rajathilagam B. and Rangarajan, M., “Spectral representation of principal components in signals and images using G-lets decomposition of subbands”, in IEEE Region 10 Annual International Conference, Proceedings/TENCON, 2017, pp. 3809-3812.[Abstract]


This paper presents a spectral subband decomposition using G-lets in time-domain for 1-D and 2-D signals. The decomposition is achieved through successive filtering and decimation steps ending up in a decomposition tree. At each node of the tree, the parameters of the corresponding subband signal are estimated using high gradients obtained at the first node. The resulting subbands are found to highlight the components of the signal. The proposed method using G-lets enables one to reduce the processing time and makes the choice of decomposition levels easier, comparatively to the case where the whole signal is processed at once. The advantage of G-lets based subbands is demonstrated using 1-D and 2-D signals. It is seen that a synthetic signal generated from a sine and cosine signal is separated into exactly the same two signals and the performance is good for monocomponent and multicomponent signals. © 2016 IEEE. More »»

2016

Conference Paper

S. S. and Dr. Rajathilagam B., “Intelligent System for Wayfinding through Unknown Complex Indoor Environment”, in Intelligent Systems Technologies and Applications 2016, Cham, 2016, vol. 530, pp. 759-770.[Abstract]


For a complex indoor environment, Wayfinding is knowing the environment and navigating within it. It is only possible if the person knows the place very well. If a person goes into an unknown environment, he/she may need assistance for wayfinding. GPS technology which works very well and is popularly used in outdoor navigation cannot be relied for indoor wayfinding as the signal strength is weak. This paper proposes a system which can be used as an assistance for navigating through an unknown environment with the aid of Wi-Fi, visual landmarks including corridors, staircase and others. This work aims at creating a custom route map for an unfamiliar complex indoor setting through visual perception and graphic information.

More »»

2007

Conference Paper

Dr. Rajathilagam B. and Prakash, K. S. S., “An Adaptive Communication Model for Large Wireless Sensor Networks”, in 2007 2nd International Conference on Pervasive Computing and Applications, 2007.[Abstract]


One of the most challenging aspects of ad hoc wireless networking is its scalability. Traditionally, using clusters has been an immediate solution. These algorithms do not promise that a network can scale well beyond 10,000 nodes. In this paper, we analyze different algorithms for various scenarios of an ad hoc wireless sensor network in terms of node distribution and propose an adaptive communication model. Using simulations, we study performance of the model in terms of density, size of network, payload and end-to-end delay. Our simulation results show that the proposed model can expand a sensor network up to 1,00,000 nodes. More »»

2004

Conference Paper

Dr. Rajathilagam B., Rangan, D. P. Venkat, and , “Techniques for Secure Communication in Emerging Wireless Ubiquitous Networks”, in The International conference on High Power Computing (HiPC), Bangalore, India, 2004.

Publication Type: Thesis

Year of Publication Publication Type Title

2013

Thesis

Dr. Rajathilagam B., “G-lets: A New Signal Processing Algorithm”, 2013.[Abstract]


Different signal processing transforms provide us with unique decomposition capabilities. Instead of using specific transformation for every type of signal, we propose in this paper a novel way of signal processing using a group of transformations within the limits of Group theory. For different types of signal different transformation combinations can be chosen. It is found that it is possible to process a signal at multiresolution and extend it to perform edge detection, denoising, face recognition, etc by filtering the local features. For a finite signal there should be a natural existence of basis in it’s vector space. Without any approximation using Group theory it is seen that one can get close to this finite basis from different viewpoints. Dihedral groups have been demonstrated for this purpose.

More »»