A superpixel based post-processing approach for segmenting dermoscopy images

Malignant melanoma is among the most rapidly increasing cancers in the world. Image border detection is often the first step to characterize skin lesion for the follow-up computer-aided diagnosis. As the existing segmentation techniques tend to find the sharpest pigment change in the dermoscopy images, the detected lesion borders are mostly contained inside the borders delineated manually by the dermatologists. Therefore, post-processing steps are needed to smooth and expand the segmented borders. In this paper, we propose a novel post-processing approach to enhance the accuracy of segmentation results, by merging superpixels which intersect with skin lesion area based on skin color consistency. The experimental results on the real dermoscopy image set show that the proposed method can improve the overall performance in terms of both accuracy and robustness.