Abstract
This paper was aimed to develop a computer-aided detection (CAD) system which may support dermatologists in their early diagnosis of skin cancer from dermoscopic images. Segmentation play key role in the automatic detection of skin lesions. In this paper Gaussian Mixture Model-based Hidden Markov Random Field (GMM-HMRF) region-based segmentation method was used. The technique is fast and automated for dermoscopic images and is used in quantitative experimental analysis for publicly available data base and compared with two other well-known methods. The results revealed that the proposed method was highly accurate. The proposed segmentation technique achieved remarkable results for nevus images (accuracy (ACC) = 97.21%, sensitivity (SE) = 95.43% and specificity (SP) = 91.76%l; while for benign images (ACC = 96.15%, SE = 96.26% and SP = 95.32) and basal cell carcinoma & Seborrheic keratoses images (ACC = 96.85%, SE = 96.85% and SP = 97.34) the results were almost same.