Research On Video Segmentation Based On Detection Of Motion-Changed Region

Nowadays, with the growing need of our material and cultural lives, thedigital video has become one of the most important information sources in themodern world. Especially, in the realms of industry, education, entertainment andsecurity etc., the digital video has been well used. At the same time, researcheshave obtained the substantial development to the related digital video processingtechnology.The video segmentation is a very important and difficult task in the digitalvideo processing. It is the basis for many video applications. For instance, theMPEG-4 and MPEG-7 use the video segmentation as the pre-processing moduleto generate the visual objects. In the content-based image and video retrievalsystem, the segmentation algorithm is the kernel part. Also, the segmentationtechnology is well used in such fields as surveillance, production control, videoediting and robotic vision.Based on the fused spatial-temporal feature of the digital video, videosegmentation is generally divided into two kinds: Temporal segmentation andSpatial segmentation. Temporal segmentation has Shot/Scene Change Detectionand their description. Shot is the basic structure layer for advanced structuredanalysis of the video data stream. The description of the scene mainly set up thestructural module and expression of the video data, such as timeline models andthe selection of the key frame. The objective of the temporal segmentation is thesegmentation of the interest object from the frames, such as leading actor(s) ofone film, automobiles on the road and so on. Spatial segmentation is also calledobject segmentation. Existing video segmentation algorithms are generally basedon the image segmentation by introducing the motive information of the object tocarry on the decision-making judgment in order to increase the precision of thesegmentation.Temporal segmentation of the video sequence is the basic process of theconstruction of the video sequence structure. The process of the construction ofthe video sequence structure is the segmentation of the frame sequence of thevideo stream, namely is to segment the video stream into several semantic units,according to the different content. In this paper, we reviewed and analyzed the algorithms of shot segmentation,and improved the algorithm of histogram. As the common histogram algorithmdoes not include spatial information, it is difficult to segment such two framesonly by histogram algorithm, which two frames have the approximate histogrambut completely different content . To overcome the shortcoming of this algorithm,we utilize region-histogram algorithm to segment. In this algorithm, we divide every frame image into 16 sub region, andcalculate the histogram of every sub region, then compare the corresponding subregion histogram of these two frame images and abandon the 8 biggest difference,the mean of the rest 8 difference as the histogram difference of the two frameimages. Thus, region-histogram algorithm introduced the spatial information, soimproved the common histogram algorithm. Technology of video object segmentation is the technical basis of manyvideo applications based on object oriented. In the video sequence, the objectswe are interested in are of self-movement , so most of video segmentation referto the segmentation of self-movement objects, such as moving people,automobile, boat and so on. While the background is not of self-movement, themovement of background is mainly generated by the movement of camera andthe change of illumination condition. From the information above, we can thinkup that a kind of effective segmentation algorithm is to segment the content intothe objects with self-movement and background. Existing video segmentationalgorithm mainly refers to the segmentation of self-movement objects from thevideo sequence. This paper also mainly discusses the segmentation ofself-movement objects, so video objects especially represent the moving objects. The video segmentation system designed in this paper utilizes thetechnology of motion-changed region detection. Due to the diversity of the videosequence , it is difficult to describe the motion-changed region by some specialfeature; therefore it is hard to segment motion-changed region directly. Asdifferent image of the two continual frames include motion-changed region andrelative noise region, and the distribution of the noise nk (x, y) and nk (x, y) +1follow some rule during the form of every frame, usually we can suppose the...