Abstract： In order to extract foreground object accurately and efficiently in natural scenes, a segmentation algorithm based on the graph cut frame was proposed. First, GMMs background models with the adaptability to the scene change were combined with shadows removal methods using color and local appearance cues to set the data cost in the energy function. Then, the difference based on local binary pattern operator was used to set the smoothness term in the energy function. After that, all pixels were assigned to binary labels by minimizing the energy function by dynamic graph cut algorithm. Finally, the proposed algorithm was verified on different videos. Experimental results prove that the proposed method can cut the foreground object accurately with realtime ability.