An Automatic Segmentation of Skin Lesion from Dermoscopy Images using Watershed Segmentation

Abstract

Among all cancers, there is a deadly disease that affects the outer layer of the body which skin cancer. There are many rules for analysis and detection of skin lesion which are provided in literature reviews. In the proposed research work, the main task is to identify the malignant lesion at the initial stage. In the initial pre-processing, the noise is isolated and a purely digital image for segmentation and Edge detection is prepared. The Sobel Operator filters the extracted cancer region as foreground region and the remaining part of the image as background regions. Finally, the desired diagnosis is extracted from the Gradient magnitude based on Watershed Transformation. Watershed segmentation segments or separates the adjacent different colors in RGB image. The proposed simulation measures the accurate diagnosis between Threshold image, gradient Image and Watershed Image and confirms the best-offered values of accuracy up to 90.46%, sensitivity up to 98.36% and specificity up to 82.95%.