Segmentation and Classification of Skin Lesions from Dermoscopic Images

Abstract

Melanoma skin cancer has been one of the most risings of all cancer, especially among non-Hispanic white males and females, which has a high risk of spread. It should be treated earlier for effective treatment. Because of the expenses for dermatologists to screen each patient, it is necessary to develop an automated system to evaluate a patient's danger of melanoma by utilizing the images of their skin lesions which will provide reliable diagnosis. To segment the skin lesion from the digital image is the major challenge. A novel algorithm which is based on skin texture is proposed for segmenting the lesion in which a set of representative texture distributions are analyzed from a non-illumination image. Based on the presence of representative texture distributions by calculating the texture distinctiveness metrics the regions in the skin image is classified as either normal or lesion. The proposed algorithm has higher segmentation accuracy about 95% when compared with other state-of-art-algorithms.