Automatic melanoma detection in dermatological images

Malignant melanoma is one of the most dangerous types of skin cancer. A very important aspect of this type of cancer is that, if detected early, it can be completely removed from the body. These characteristics make the research on automated melanoma detection systems a field with high potential. In this thesis, a system for automatic detection of melanoma is designed, developed and studied. The system is composed of five stages; image acquisition, illumination correction, lesion segmentation, feature extraction and classification. The techniques implemented in illumination correction are based in morphological operators and the Retinex algorithm. The four proposed methods for lesion segmentation include Otsu's method thresholding, GVF Snakes, and two novel methods based in Mean Shift clustering using color and texture information. The classification stage makes use of linear discriminant analysis and SVMs. In addition, a GUI tool that takes advantage of the mentioned techniques is created and presented