Publication

Abstract : Skin cancer is a dangerous condition for which its early identification is essential to effective treatment. This work is a comprehensive evaluation of the dermoscopic skin lesions that are compressed using the Five Modulus Method (FMM) algorithm. Using the FMM technique, the original images are compressed, saving space and storage. The study aims to investigate the performance of a Deep learning model on the HAM10000 dataset consisting of 7 different types of skin lesions used for the study. A trained VGG16, CNN, Random Forest, and SVM were used to evaluate the performance. To conduct a fair comparison the models were trained on the HAM10000 dataset and the compressed dataset using a consistent training methodology and hyperparameter optimization techniques. Results showed that there is minimal difference in the accuracy of models with compressed and original images. PSNR values used as similarity measures show that important information is retained during the compression.