Torso Detection in Static Images

The existence of different clothing and cluttering backgrounds is the main difficulty for accurate torso detection in still images. This paper presents a novel dominant colors based method to tackle this problem. First, the image is segmented into uniform areas. Then, dominant colors of the torso are adaptively selected using a color probability model. Finally, the torso candidates can be extracted based on the dominant colors. Experiment results show that the proposed approach can accurately extract the location and orientation of the torso in most situations