Research On Video Segmentation Using A Novel Local Binary Pattern Descriptor

Video segmentation is important research on the process and analysis of video information. It is the basis of an efficient processing system, such as video retrieve based on the content,video compression,object recognition and object tracking. Video segmentation is to segment the video to spatial-temporal homogeneous regions, so it can generate the meaningful foreground object.This thesis aims at to study the motion detection and video segmentation algorithms. In the basis of systemic analysis of conventional video segmentation methods, the motion detection algorithm using spatial-temporal information coding technology and the corresponding video segmentation scheme are carried out. The main contribution of the dissertation is summarized as the follows.Firstly, the classical LBP descriptors for image analysis and its generalized form in spatial-temporal domain for video processing are introduced, respectively. Secondly, the novel 3D LBP coding method with translation-invariant property for describing spatial-temporal information is proposed and applied to motion detection. Finally, the video segmentation scheme is provided by combining the new 3D LBP descriptor with morphological filters.The distinguished advantage of the method, proposed in the thesis, is that it does not need to extract the edge information, neither to construct the background models. Instead, it transfers the change information to different LBP values, and then sequentially performs thresholding and morphological filtering to segment the video. Meanwhile, the proposed method has the characteristic of low computation load and easy to realize. The segmented results show that using the present method can achieve satisfied segmentation.