Real-Time and Multi-Video-Object Segmentation for Compressed Video Sequences

Abstract

We propose a real-time object segmentation method based on Gaussian mixture model(GMM) for MPEG compressed video. Computational superiority and multi video objects are the main advantages of compressed domain processing. In the paper, first, we introduce the macro-block structure of the MPEG encoded video and the preprocession of video vectors, then we build a GMM of motion vectors and adopt the genetic-based expectation-maximization algorithm (GA-EM) to compute its multivariate parameters. It is able to estimate automatically the number of objects of the motion model using the minimum description length (MDL) criterion. At last, we give the steps of objects extraction. It is proved that the algorithm is real-time and effective from the experiment results.