ABSTRACT
Among the wide variety of skin abnormalities, malignant melanoma is the deadliest and causes a vast majority of deaths. Existence of dark dots/globules is a feature demarcating the melanocytic skin lesions from the others. In this reported research work, a methodical approach for the identification and segmentation of dark dots/globules from the dermoscopic images using mathematical morphology has been proposed and implemented. Different morphological gradient operations have been performed with varying size of the Structuring Element (SE) to identify the small circular structures present in the skin lesion area. In the segmentation stage the threshold value has been selected according to the range of intensity values present in those dot/globule regions. Subsequently, a number of dot/globule related features have been extracted for the classification of the dermoscopic images. Furthermore, it has been shown that the dot/globule related features have a significant effect on the classification of the melanoma, and has the potential for scoring even 100% sensitivity, provided it is used in conjunction with appropriately chosen classification algorithms.
