Technology Research, Based On The Motion Characteristics Of Video Object Segmentation

This dissertation focus on one of key techniques related to MPEG-4 standards — video object segmentation technique. Video object segmentation refers to the division of video sequence into semantic entities (objects) according to related criterions, where the entities are processed as objects and called video object (VO.). Many applications related to video compression and transmission, and pattern recognition rely on video object segmentation. Video object segmentation techniques are also important tools for content-based video coding and manipulation, and interactive multimedia application. These semantically segmented objects can be coded independently, so that object-based manipulation of video content can be realized in interactive multimedia application. For example, in the context of the MPEG-4 standard, a video is considered to consist of independently moving objects and is encoded every object.After analysis the character of video sequence, the motion-based video object segmentation technique is studied particularly. At first, the video object segmentation technique is discussed, including the Current situation, classification and basic methods of video object segmentation. Then, the basic theory of video object segmentation is reviewed and analyzed, including static image segmentation and mathematics morphologic.Temporal video object segmentation is developed. To remove the influences of tiny motion in background on the video object segmentation, a method for video object segmentation based on MDBP (minimum distance between pixels) is proposed, improving the precision. First, a reliable background image is reconstructed, using the technique for background construction based on background confidence image and background label image. Then, after video objects are obtained by background subtraction, fault detection of video object caused by small change of background is overcome, using MDBP (the minimum distance between pixels) and the whole MDBP rules.The spatio-temporal based video object segmentation technique is studied here. As is known, spatial methods for video object segmentation achieve high segmentation accuracy with too garrulous segments inside video objects and background regions. On the other hand, temporal segmentation method has the advantage of high segment speed but coarse boundaries of video objects. Combining both the advantage of spatial method and that of temporal one, a new video object segmentation method based on comparative histogram is proposed. This method makes use of the video's spatial relativity and temporal relativity to get the video object. Color information in spatial segmentation and accumulation frames difference in spatial segmentation is employed to overcome the problems that outline of video object is not accurate by using framedifference and spatial segmentation method is difficult to realize.Finally, the global motion (camera motion) segmentation technique is discussed. To realize the content-based retrieval and recovery of video, video segmentation into shots and its description are necessary. Usually the global motion refers to the video motion caused by cameras motion, which has a strong integrity, a small computation, stable results and convenient feature representation. For most video sequences, motion of camera always follows important persons and affairs. Therefore, it is believed that the global motion information reflects the semantic contents of the video. In this paper a method for video's global motion (camera motion) segmentation is put forward based on SVD (singular value decomposition). Slice images of the video are gained firstly, which includes the global motion information. Then, global motion information is extracted, using SVD. In the end, according .to the global motion, video is segmented into shots with each shot having only one camera motion.