Moving Object Segmentation Based on Video Coding Information in H.264 Compressed Domain

The paper presents a novel segmentation approach to extract moving objects in the H.264 compressed domain. In the proposed method, some H.264 video coding information besides motion vector is used to segment moving object more accurately. Firstly, motion vector is refined by the motion relativity in space and time. Secondly, a coarse segmentation is gotten through motion vector difference between motion vector and global motion. Finally, the coarse segmentation is refined by either intra prediction mode in intra-frame or expansion and contraction operation in inter-frame. Experimental results for several H.264 compressed video sequences demonstrate the good segmentation quality of the proposed approach.