Automated skin disease segmentation and classification using multi-class SVM classifier

Abstract

The aim and objectives of the study is to perform automated segmentation of different skin cancer images using C-means and watershed algorithm and features extraction by Gray Level Co-occurrence Matrix (GLCM) and Image Quality Assessment (IQA) method. The classification of different diseased state was done using multi-class SVM classifier. In the Proposed method, 45 digital images collected from MIT BMI database consists of warts, benign skin cancer and malignant skin cancer and normal skin images. These images are subjected to various pre-processing techniques such as resizing, conversion and contrast enhancement. Then these images are segmented using c-means and watershed algorithms individually. Feature extraction is performed using GLCM and IQA methods for examining texture which gave the statistical parameters of each algorithm respectively. In this work, different types of skin diseases are commonly classified as Benign Skin Cancer, Malignant Skin Cancer and Warts using multi-class SVM (Support Vector Machine). C-means algorithm produced better segmentation results with an accuracy of 98% compared to watershed algorithm (92% accuracy) in segmenting the skin cancer images. A computer aided diagnostic tool was implemented to diagnose the different diseased state of skin cancer.