Weighted Skin Color Segmentation and Detection Using Graph Cuts

In this work we propose a principal approach combining graph cuts, color weighting and texture information to segment skin regions from images. Successfully detected faces serve as foreground seeds, no assumptions about the background is made. Further, we introduce the concept of skin weights and skin weighted images. Performance is improved by efficiently adjusting pixel weights. Prior approaches do not provide a general skin detection when the information of foreground or background seeds is not available, e.g. when there are no faces present in the image. A concept for processing arbitrary images is introduced: We learn a universal seed to overcome the probable lack of successful seed detections. Experiments on a database of 300 images with annotated pixel-level ground truth show that the scheme outperforms widely used approaches. We also evaluate the effect of different edge detectors on graph cuts based skin segmentation. It is shown that a texture and color based weighting scheme improves skin segmentation compared to a color only weight-