Based On Spatio-temporal Video Object Segmentation Algorithm

As an important part of the video coding standard MPEG-4, the technology of objected coding is the key point to advance the efficiency of coding, and the efficient video object extraction algorithm is also the basic of the objected coding. Due to various kinds of factors according to the environment, the algorithm can not be used in any condition but for specific video. Referring to the complicated video content, the results degraded distinctly, so the problem of improving the flexibility of the algorithm has turned into the most active research field.This thesis has systematically investigated the segmentation problem in different condition, such as the video whose background and foreground change little or which has lots of noise in the background and the object moves a lot in the foreground, and this paper proposes a new method of objection extraction based on the multi-frame gray value distance and the improved watershed segmentation. The temporal segmentation is based on change of multi-frames' gray value distance removing the background pixels by the gauss clustering algorithm and receives the temporal template using morphology operator. In the spatial segmentation, a improved watershed is performed on the low-resolution image from wavelet transform, by adding the counter information to the transform result, the problem of over-segmentation is reduced efficiently. In the combination, we obtain a temporary object by rough segmentation and remove the most part of background. Than we update the attribution degree value considering the neighboring areas' attribution degree value and get the results after accurate segmentation. Finally the integrated object is received after con-transforming. The experiments show that the algorithm has good result on the video whose objects change a lot and the background is filled with noise.The paper discusses the application of the algorithm in the area of gait classification and picture repairing. During the processing of gait classification, we receive the recognized objects by the spatio-temporal combination segmentation algorithm, and establish the template library by the object character vector. Finally we compute the distances between the test objects and templates and get the results by the minimal distance. In the picture repairing, we get the temporal template by the gauss clustering, and divide up the image by the spatio-temporal combination segmentation algorithm. Finally we search the matching areas for the waiting for repairing areas during the segmentation results and repair the whole destroyed areas. According to the results of experiments for gait classification and picture repairing, the algorithm can improve the quality of the segmentation and advance the performance of the gait classification and picture repair processing.