| In the field of computer vision, segmentation of video foreground objects from backgroundhas many important applications, such as human computer interaction, video compression,multimedia content editing and manipulation. The accuracy of segmentation has directlydetermined whether the follow-up identification process can succeed or not. Therefore the videosegmentation remains the hot topic in computer vision. In this paper, we propose a new methodwhich can accurately extract video object from a single video sequence through the combinationof motion detection and Markov random field model (MRF).First of all, an improved moving detection method has been adopted to obtain the region ofvideo object which can be gained through the max frame difference in the RGB components ofcolor images. Besides that, the fame differences will be identified in the continuous three framesso as to eliminate background which leads to revealing of the cause of the error area. It is farenough to accurately identify the target region only by the frame differences because themoving objects can come to a sudden still or partial statement of still. In this paper, the region oftarget object can be identified by the current frame with combination with the former frame.Then, based on the moving detection, the region of target objects can be set as theuncertain regionTU and backgroundTB, and this can be set as a prior into the Markovrandom field model. So the result can be obtained by MAP-MRF framework which thesegmentation is turned into energy optimization. In this paper, a novel maximum flow/minimum cut algorithm has been provide to extract the object from video, and in the meantimea multi-cue interaction segmentation has been proposed to correct the wrong segmentation.In the end, an interactive video object segmentation system is developed based onOpenCV2.1with MS Visual2008. The experiment results show that our algorithm can achievea highly segmentation, which proved the feasibility and accuracy of our algorithm. |
PDF Full Text Request