Automated computer vision method for lesion segmentation from digital dermoscopic images

Abstract

Melanoma is one of the fatal skin cancers. Lesion segmentation is a crucial step in the analysis and diagnosis of a skin cancer from digital images. This work proposes a technique for automatic segmentation of lesions from digital dermoscopic images using adaptive threshold. The use of many image processing techniques, such as average filtering for removal of hair and skin scales, mathematical morphology to reject the false positives, texture and geometrical features to correctly segment the lesion, have been successfully implemented to accurately segment the lesion from the images. The segmentation results from the proposed work are compared with ground truth. The results are convincing and show that the method has good accuracy. A minimum correlation of 91.7% and a maximum overlapping score of 97.03% has been obtained for the digital images.