Study On A Mpeg-4-Oriented Video Object Segmentation And Tracking Algorithm

The distinct feature of MPEG-4,the second generation object-based coding standard, is content description and coding based on video objects. So a video sequence must be segmented into a set of video objects first. Video object segmentation is also an opening part in the standard. The performance of the segmentation algorithm is crucial to the final MPEG-4 coding products, so the author chooses this question for discussion.Many methods segment with the space-time information of video sequences. For the static background sequences the author presents a segmentation algorithm based on difference and intersection. It has three virtues: First it can obtain accurate moving object information; Second it makes threshold adaptively; Third it combines space-time segmentation results to make better. The author also proposes a video object segmentation algorithm based on background registration. The background region doesn't depend on the characters of moving objects to obtain easily because of it's reliability. And we don't want to know the moving object's shape , number and so on. Experiments show the two algorithms can separate objects from stationary background and have some feasibility. And for the moving background sequences, we should make the global motion estimation and compensating first, and then segment objects in terms of the static ones. After the initial segmentation, the author proposes an improved algorithm based on Hausdorff Distance to track objects. It can obtain the new objects accurately even if they are sheltered or their shapes have slow changes.There is no all-purpose segmentation theory yet up to now. The hardware realization of algorithm is a challenging issue which must be solved. It is also the future study direction that the author should make great efforts.