Multi-Video-Object Segmentation Based on SOFM Network for Compressed Video Sequences

This paper proposes a real-time object segmentation method based on SOFM network for MPEG compressed video. First, we introduce the macro-block structure of the MPEG encoded video and the preprocession of motion vectors, then the motion vectors are given as input to the self-organizing feature maps (SOFM) models which can automatically estimate the number of objects of the motion model, and the motion vectors are divided into several sorts. Each sort belongs to one object, so we can extract object . Finally we give the steps of object extraction. It is proved that the algorithm is real-time and effective from the experiment results.