ABSTRACT

Skin cancer is now becoming a challenging issue to identify the exact location of affection on the skin tone. A novel hierarchical k-Nearest Neighbors (k-NN) classifier is more useful to find the affected level of skin and the type of cancer disease. The k-NN classifier is comparatively simple, quick and effective. The structure which is of hierarchical level mainly decomposes classification into a set of easy issues. At the first level of the classification, feature selection is done. The most relevant feature subsets such as color and texture features are extracted from skin lesions and passed to each node of the classification level. The efficiency of the proposed scheme is typically larger in discriminating cancer and pre-malignant cells from the benign cells and it reaches overall classification accuracy over a few common classes of skin lesions that include two non-melanoma cancer classifications. An automatic skin cancer classification system is developed and the relationships of skin image across different type of training networks are studied with different types of image preprocessing. To enhance the classification results, the image properties of the normal skin is eliminated from the skin affected area and the cancer cell is presented in the image..