Research Of Video Objects Segmentation Based On Active Contour Model

With the development of multimedia technology, video object segmentation has become a hot topic in the video research field. It has a broad application prospect in video encode and retrieval, video database browsing, intelligent video monitor and virtual reality, multimedia interaction, computer vision and pattern recognition. All the above application must be realized based on video object segmentation. Active contour model could generate closed curve by self through appropriate initialization, and then drive the closed curve tend toward the real contour of object continually under relevant power or function. It is the effective algorithm to resolve the problem of video object segmentation.How to improve the accuracy of video object segmentation based on active contour model is researched in this paper, including segmentation of single video using parameter active contour model and segmentation of multi-video using geometric parameter model. The paper emphasizes the research of multiple video objects segmentation; it also solve other problems in video objects segmentation such as partially occlusion effectively. The main research of the paper is as following:(1) In order to improve the accuracy of video object segmentation based on parameter active contour model, an algorithm of video object segmentation based on difference multiplication methods and GVF Snake is proposed. These two methods make up the shortcomings of each other in algorithm. It firstly eliminates most background boundaries through difference multiplication of four consecutive frames. Then median filer is used to remove residual strong noise in the image. Finally, it operates the GVF Snake model to obtain the moving object contour.(2) With regard to multiple video object segmentation, an algorithm of video object segmentation based on Chan-Vese model and grid-based object detection is proposed. It eliminates most background boundaries through the difference multiplication of four consecutive frames firstly, minimizes an energy to propel the level set evolution, making the active contours toward objects approximately. Then, grid-based object detection is used in difference multiplication image to overcome spatial disconnection. The resulting motion regions with noise are obtained. Finally, pixels in the motion regions are determined whether they belong to objects or background through motion estimation. With the minimal Bayesian error probability criterion, it also can extract partially occluded objects.In summary, this paper analyzes the video object segmentation in the pixel domain, focusing on research of some problems and key issues, and proposes corresponding novel and improved algorithms. The simulation results using standard test sequence and video filmed by ourselves both verify the proposed methods in this paper.