Research On Video Moving Object Segmentation Technology In The H.264Compressed Domain

Video object segmentation is a key research issue in the field of digital videoprocessing, and it has been widespreadly used in video coding, retrieval, patternrecognition and robot vision. Because of the high computation cost of video objectsegmentation based on pixel domain and the real-time requirement in the practicalapplication, researchers start to focus on the video object segmentation based oncompression domain. Due to the lack of information which can be directly used asvideo objects segmentation in the H.264compression standard, and theinterference of noise during the segmentation processing, the objects aftersegmentation will be rough more or less. In order to make up the shortage, thisthesis proposes an algorithm which improves the accuracy of motion vectorsthrough the pretreatment and post-processing and improves the segmentationquality as much as possible. The main achievements are summarized as follows.(1) Specific to the video sequence of still background, a video objectssegmentation method based on fuzzy c-means clustering of H.264is proposed.Firstly, the moving vector field is extracted from compressed video streaming. Theaccuracy of motion vectors will be improved through filtering, calibration andbackward iteration cumulative. Secondly, the movement area is divided by usingmodified fuzzy c-means clustering and the different moving object is segmentedthrough part decoding. Experimental results show that this method can completelysegment the moving objects in video sequences.(2) Specific to the video sequence of motion background, a video objectsegmentation method based on H.264is proposed. The motion vectors areextracted from video sequences, which are composed of motion object motionvectors, noise motion vectors and background motion vectors. In order toaccurately segment the video motion object, we need to remove the noise motionvectors and background motion vectors. Then the filter is used to remove the noisemotion vectors, and the global motion compensation is used to remove the globalmotion vectors. After that, the background of video sequence can be considered asstill image. At last, we use the method in (1) to process the motion vectors andsegment the moving object. Experimental results show that this method cancompletely segment the moving objects in video sequences.The result is proved and analysised based on the experiment platform of matlab7.0and the JM86platform of Vc++6.0.