Learning Hand-Crafted Features for K-NN based Skin Disease Classification

Dermoscopy imaging is a diagnostic method widely used by dermatologists for the diagnosis of skin cancer. Since dermatological evaluation is dependent on the expert, timeconsuming and subjective, automated systems are preferred by dermatologists to contribute to the decision-making processes. Detection of melanomas from skin lesion images accelerates the treatment process together with the early diagnosis of the disease and reduces the morbidity and mortality rates. For this purpose, in this paper, a novel approach has been developed to solve binary image classification problem of melanoma and, nevus and seborrheic keratosis by using the K-nearest neighbor algorithm with hand-crafted features such as color, shape and texture extracted from publicly online ISIC 2017 dataset consisting of images taken from the skin region.