Segmentation of digitized dermatoscopic images by two-dimensional color clustering

In this paper a color-based segmentation scheme applied to dermatoscopic images is proposed. The RGB image is processed in the L* u* v* color space. A two-dimensional (2-D) histogram is computed with the two principal components and then smoothed with a Gaussian low-pass filter. The maxima location and a set of features are computed from the histogram contour lines. These features are the number of enclosed pixels, the surface of the base, and the height of the maximum. They allow for the selection of valid clusters which determine the number of classes. The image is then segmented using a modified version of the fuzzy c-means (FCM) clustering technique that takes into account the cluster orientation. Finally, the segmented image is cleaned using mathematical morphology, the region borders are smoothed, and small components are removed.