Research On Video Segmentation Technology

With the growth of internet and computer technologies, the interest in advanced interactivity with audio-visual contents is increasing rapidly. To meet these requirements, new video coding standards, MPEG-4 and MPEG-7 were established. The MPEG-4 standard defines the conception of VO (Video Object) to support its interactive function based on content, the segmentation of video sequence combined into arbitrary region is required by the standard. Segmentation of moving objects is also important for MPEG-7.Although object segmentation of video sequence into VO is an important research subject,it is not a normative part of the MPEG-4 or MPEG-7 video coding scheme. Although many researchers work in this field for many years, there is not any satisfying algorithm can process all sequence.At present, there are many classes of object segmentation methods,and the most effective class of segmentation method is spatial-temporal, so an spatial-temporal algorithm was proposed in this paper. First of all, in temporal domain video sequence were segmented with change detection algorithm to get segmented model; and then, using watershed algorithm based on gradient image segmented image, and over-segmentation was overcome by region merging base on spatial-temporal fusion.To improve speed of watershed algorithm and region merging, watershed transformation combined temporal filed was used. At last, target is extracted by spatial-temporal fusion. The simulation results show that the proposed algorithm can segment whole video object accurately.