Study On The Technology Of Video Segmentation For Video Objects Generation

Traditional video coding standards such as MPEG-1, MPEG-2, H.261, H.263 and so on, adopted block-based coding methods. These methods not only are of low coding efficiency, but also don't consider the real composition of video scene in terms of video content. With the increasing of multimedia applications and services, especially for the increment of content-based manipulation and interactive multimedia applications, such as content-based video data indexing and querying and so on, traditional coding methods can not satisfy the new demands of multimedia applications, so it is necessary to adopt a new method to code the video in terms of content.The new coding standard MPEG-4 adopts content-based coding method. In the MPEG-4, each frame of a video sequence is segmented into semantically meaningful objects which are called video object planes (VOP's), and different encoding tools are applied to each VOP. Object-based coding or content-based coding not only can greatly improve the coding efficiency, but also allows users to manipulation multimedia data in terms of content, such as content-based control and bit stream's manipulation and so on. The generation of video object is the basis of object-based coding and interactive manipulation, so studies on the generation of video object have important significance and application values.In video sequences, the objects we are interested in are those which are of self-movement, so most of video segmentation refer to the segmentation of self-movement objects. This paper also mainly discusses the segmentation of self-movement objects.This paper discusses and establishes theory model of video segmentation. Based on the model of video segmentation, through analyzing the composition of video scene from different aspects, we discuss two different video segmentation methods, one is based on the motion-changed region and edge features and the other is based on spatial-temporal information. For the two methods, we establish their system model, discussthe realization steps and point to some important technologies concerned. In addition, for the special case of stationary background, this paper introduces an idea of video segmentation through background estimating and proposes an adaptive method for background estimating.In the video segmentation process based on the motion-changed region and edge features, an automatic motion changed-region detection algorithm is proposed which detects motion-changed region through filtering relative noise based on the estimated relative noise's parameters in difference image. First, the current frame in a video sequence is globally motion compensated after the global motion estimation, and the difference operation is taken between the globally motion compensated current frame and the previous frame to get difference image. Then, in difference image, an appropriate initial relative noise's mean and variance is estimated and based on the initial estimated value an iterative and weighting algorithm is used to get the estimated optimization mean and variance. Through estimating the noise's parameters, noise is filtered adaptively and the motion-changed region is detected. In addition, another automatic motion-changed region detection algorithm is proposed from the point of fuzzy clustering. In this algorithm, a fuzzy clustering rule is firstly established and motion-changed region is then detected through using a fuzzy classifier to distinguish between relative noise and motion-changed region in difference image after global motion compensation. Finally, video object can be generated through combining motion-changed region and edge features in the current frame.In the process of video segmentation based on spatial-temporal information, a spatial segmentation algorithm based on hierarchical partition is proposed which is in accordance with visual properties. Regions after spatial segmentation are merged based on motion similarity and spatial connectivity and then video objects are generated. Considering that moving objects exist in...