Vector Based Classification of Dermoscopic Images Using SURF

Detection of melanocytic skin lesion at an early stage increases the probability of being cured. Dermoscopy is a widely used diagnostic tool that aids the diagnosis of skin lesions and is proven to increase the accuracy of melanoma diagnosis. In this paper, vector based pattern analysis and classification approach for dermoscopic images are proposed. Feature plays a vital role in pattern recognition system. The various features include color, texture and shape features. Texture is considered as a dominant feature. In this paper lesion is segmented using region based Statistical Region Merging (SRM) algorithm. Scale invariant based Speeded up Robust Features (SURF) technique is used for feature point detection and descriptions under texture analysis. SURF uses Hessian matrix approximation for feature point detection, haarwavelet response for feature descriptions. It uses l*a*b color space for describing color intensities. The patterns so detected are classified using multi-SVM classifier. The proposed system provides the cla3ssification accuracy of 86.37% and sensitivity, specificity rates as 86.53% and 96.42% respectively.