Extracting Color Name Features Utilized for Skin Disease Characterization and Comparing It to Other Representations Describing the ABCD Dermatological Criteria for Melanoma Inspection.

Recent development in image resolution has led to the excessive use of medical images in creating automated systems for inspecting and early detection of deadly diseases. Some computer vision diagnosis systems has focused on the inspection and analysis of skin cancers, such as melanoma. It is considered as one of the most fatal skin condition if it is caught at an advanced stage. Thus, it is important to regularly check every skin lesion for an early detection of melanoma. Many systems have been introduced in the last years. They mainly rely on visual features describing the color, border and texture to inspect the malignancy of skin lesions. The performance of such systems is impressive but still lacks the integration of many other different clues invented by dermatologist. Studying the color variegation can shed the light on some criteria already adopted by doctors of malignant lesions. Thus, extracting the different colors skin lesion contain is of high relevance. This paper presents a novel method for extracting and quantifying the different colors in a skin lesion in a supervised way that will ensure high specificity and sensitivity in classifying different skin conditions simultaneously. This paper aims to prove the effectiveness of our proposed method by making a comparative analysis of different state-of-the-art methods held on the ISIC 2017 and SD 198 datasets. Moreover, we employ a collected dataset to explore the performance of our proposed method. The application of our hand-crafted color features results in better classification performance of different skin illnesses.