Features Extraction: A Significant Stage in Melanoma Classification

Abstract

Skin cancer is by far the most common cancer. Out of all the skin cancer types, Malignant melanoma is the deadliest one which can lead to mortality. Melanoma can be cured if detected at initial stage, so its early diagnosis is of quite importance. For this purpose Computer Aided systems are preferred which are based on image processing techniques. Among these techniques feature selection stage is of utmost significance as it directly affects the accuracy and robustness of CAD systems. In this work, it has been diagnosed that which kind of features out of color, texture and shape contribute admirable in melanoma classification task. In first phase, pre-processing steps are performed to refine images which involves hair exclusion and other noise removal techniques. Then segmentation is performed using otsu method followed by various morphological operations. After this color, texture and shape features are extracted and given to SVM classifier individually. Experimental results clearly show that color features are prominent in classification having accuracy of 93.5%. And among all color features, distinction ability of features obtained using HSV and CIE L * a * b color space is higher (91%).