U-Net Based Segmentation of Skin Lesions Using Dermoscopic Images

Abstract

In this article, a new approach has been proposed to segment the skin lesion using Dermoscopic image. The hairs in the images are removed by Dull Razor's method with two iterations. Then the color artefacts in images are removed using standard color constancy methods such as color equalization and grey world method. Also, these two color corrected versions are blended using multiscale image fusion technique. Then the two color corrected versions and their fused version are given as input to a U-Net which is developed with three encoders. Then the feature maps obtained from the encoders are fused and concatenated with the decoder part to get the segmented skin lesion. This approach outperforms well compared with the existing works.