Boundary Roughness Analysis of Skin Lesions Using Local Fractals and Wavelet Transforms

Boundary roughness of skin lesions is of clinical significance for early detection of malignant melanomas. An integrated approach of local fractals and multi-level wavelet analysis is proposed in the paper. Local fractal profile of a lesion contour is generated to describe boundary roughness while a tree structured wavelet packet is further processed to give behaviors of the local fractals at different frequency bands. Features extracted at each sub-bands are selected using Hausdorff distance for identification of skin tumors in order to reduce dimension in feature space. Experimental results show that features chosen are dominating clues to discriminate skin tumors.