Imaging method for Noise Removal and segmentation of Skin Lesions from Dermoscopic Images

Abstract

Melanoma is a fatal skin anomaly which can be treated if diagnosed under benign condition. The accuracy of cancer detection depends directly on the accuracy of lesion segmentation. This work proposes an imaging method for lesion segmentation from dermoscopic images using inpainting, edge detection and intensity based threshold. The use of mathematical morphological operations has been done to remove noisy pixels post segmentation. The use of proposed techniques has made the method less complex and computationally efficient and can work in real-time. An average Jaccard index value of 89.2383% and correlation coefficient of 92.5271% has been achieved using the proposed method. The results are convincing and guides in the direction of usage of proposed algorithm as subset of some real time application for melanoma detection.