Lesion detection in dermatoscopic images using anisotropic diffusion and morphological flooding

A new technique for the unsupervised detection of multiple objects in noisy background is presented in this paper, with application to digitized dermatoscopic images. The proposed method uses anisotropic diffusion in the L*u*v* uniform color space to suppress background noise and spurious structures while preserving the object boundaries. A simple and fast hair removal scheme based on morphological operators and luminance thresholding is also introduced for the application to pigmented skin lesions. The luminance component is then used to extract iso-level closed contours by morphological flooding. These curves are considered as contour candidates and the selection is based on the minimization of a gradient based energy functional. The main advantages of this technique is the low processing time and the possibility, if necessary, to scroll through the different level curves without additional processing