Semantic spatio-temporal segmentation for extracting video objects

To improve flexibility and interactivity, the latest video coding and indexing methodologies address content-based processing of video sequences and therefore necessitate the segmentation process to extract video objects. We propose a segmentation method, called semantic spatio-temporal segmentation, which exploits motion and spatial information together with user's semantic input. Combining this information for extracting video objects is viewed as a fusion process, and a fuzzy-set approach is proposed in this paper. Simulation results have demonstrated that semantic video objects could be effectively identified by exploiting our method.