Research On Video Moving Object Segmentation Based On H.264 Coding Standard

Segmentation of video moving object is extraction of semantic area meet certain characteristics from video sequence. It is a difficult and hot issue in the digital video processing areas. It is widely used in video coding, video retrieval, pattern recognition, robot vision, etc. Video motion object segmentation has two branches. One is based on pixel domain; another is based on compressed domain. Because computational complexity is larger in pixel domain so as not to meet real-time requirement, research for video moving object is in compressed domain in recent years. Without decoding all compressed stream, computational complexity reduces largely, so the method of video moving object can meet our demands in the compressed domain.Research for video object segmentation in the compress domain is mainly based on the MPEG encoding standards and H series code standards. We can extract DCT coefficients and MV from MPEG compressed'stream to segment object, but we can only extract MV from H.264 compressed stream, because of.I frame's DCT Based on residual. That is Very important problem, How to ensure the correctness and reliability for motion vector field. Now research for video moving object extracting from the H.264 compress stream is very little.This paper mainly studies how to extract moving object from H.264 compressed stream. Main contributions are summarized as follows:1,We extract motion vector field from H.264 compressed stream that background is static. Firstly, motion vector field extracted from H.264 compressed stream is processed by noise processing and median filter. An iteratively backward projection scheme is then proposed to obtain an accumulated motion vector filed. Moving object region are finally extracted in turn based on k-means cluster. Simulation results show we can extract object of good quality from the compression stream.2,Research for video frame that it is not static for its background. Firstly, we need to do global motion estimation and compensation for motion vector field. Making use of inconsistency between object motion and scene motion, we extract moving object by few frames'motion vector field. Experiments show that algorithm can extract moving object of good quality from the compressed stream. In this paper, the research was verified and analyzed by Matlab7.0 simulation platform and JM8.6 platform.