Pose-Invariant Clothing Segmentation Based on Saliency Detection and Graph Cuts

A novel algorithm of clothing segmentation,which can handle variable human poses,is proposed in this paper.The segmentation performance is significantly improved by combining saliency detection and graph cuts methods.Firstly,we propose a sliding window based visual saliency detection method to locate initial foreground background seed regions,which are invariant to poses.Secondly,the seed regions are further adjusted by a graph based segmentation method.Finally,the segmentation results of clothing regions are acquired by an iterated graph cuts method called GrabCut,with the seed regions as the input.The experimental results demonstrate that our proposed algorithm can achieve promising results and is feasible in practical applications.