A region growing based imaging method for lesion segmentation from dermoscopic images

Abstract

Melanoma is a fatal skin cancer. Correct localization and segmentation of a lesion is decisive in proper detection of a skin cancer from dermoscopic images. This work proposes an image processing technique for automatedlesion segmentation using a region growing technique. Low pass filtering techniques like median filter, are employed to remove the edges generated by the hair, which reduce the accuracy of correct segmentation. Multi-threshold and geometrical features as solidity and extent makes the proposed method adaptive. The ground truth, provided with the database used, has been used as a benchmark to determine the accuracy of segmentation using the proposed work. An average correlation and overlapping score are found out to be 93.79% and 91.13% between the ground truth and segmented lesion using proposed method.