Flow-guided hair removal for automated skin lesion identification

Abstract

In this paper, we propose a method for removing hairs automatically from skin lesion images. To achieve this, we employ an edge-detection technique based on edge-tangent flow. To detect only hair-like structures, rather than contour boundaries, we propose a novel, specialized method for detecting hairs. Regardless of the personal characteristics of the hairs, hairy regions are detected because our method detects coherent thin lines of consistent width. We then restore the hairy regions detected by the proposed method by using the texture synthesis method. Our method restores the regions occluded by hairs with very few remarkable artifacts, because we utilize pixels that actually exist in the source image to restore the occluded areas by searching for the best matching pixels.