A graph cuts based human shadow extraction using entropy-expressed texture and color features

Human shadow is one of the most important aspects which can be used for gait recognition. However, human shadow extraction is a hard task due to the umbra-penumbra phenomenon and backlight effects. In this paper, a new Graph Cuts based human shadow extraction method using entropy-expressed texture and color features is proposed. In this method, the local area attributes are estimated totally both by the entropy-expressed texture and color features to build the data term, and the whole-optimized shadow region are extracted by using of the Graph Cuts theory. Some experimental results show that our method is stable and efficient under different conditions.