Study On Video Segmentation Based On Gray Correlation And NS Spatio-Temporal Combination

Video object segmentation is an important part of MPEG-4video coding standard, and also foundation of some application system, such as video monitor, multimedia interaction and pattern recognition. It has been attracted particular attention by some researchers, and has been achieved many achievements. Due to the complexity of the video objects, there are still some problems, for instance, when the target motion speed is slower, or the contrast of target and background is lower, the segmentation results are not complete and prone to error segmentation. Therefore, it has important theoretical significance and practical value to study video object detection.In view of this, this paper analyzes and summarizes the related research work at domestic and abroad, and proposes a method for video segmentation based on gray correlation and NS spatio-temporal combination, and proves the validity using experiments. The main content of this paper is as following:(1) In the time domain, different motion detection methods are used for different video sequences. First of all, the paper uses background difference method based on video stream for video sequence which has static background. If the video sequence is simpler and has faster object movement speed, it is used median method to establish background, or used improved adaptive GMM. Secondly, an algorithm of frame difference based on edge and gray correlation is proposed for the video which has no complete background, to overcome the sensitivity to noise of traditional frame difference.(2) In the spatial domain, the paper uses the higher contrast characteristics of identified regions T in NS algorithm to carry watershed transform, which reduce the over-segmentation phenomenon, and improve segmentation accuracy combining with the edge information of video image.(3) The other difficulty of video segmentation is shadow detection and removes. In order to reduce the error detection of HSV color space, the paper proposes a method based on color space and edge information, which could ensure the integrity of the moving target.(4) Finally, the spatial segmentation result is mapped to the time domain based on the membership identification to judge the initial segmentation accurately. The paper also uses various evaluation criteria to illustrate its efficiency. In short, the paper uses all kinds of theoretical foundations and the latest achievements to solve the error segmentation when the object motion speed is slower, or the contrast of target and background is lower, and improves the generality and accuracy.