Grey And Texture Feature Based Video Sequences Matching Algorithm Study

In the Image &Video retrieval and matching field, Text-based technique has been substituted by Content-based technique, which becomes the main methods and development trend. What's called Content-based means that query image &video by color, shape and texture of image, and combined with those technology of computer image processing, pattern recognization and database. Because video is a linear structure made up by consecutive images and be more complicated, it can those Image retrieval and matching method on base. While data of video is too huge that how distill efficient feature and put forward high-efficiency retrieval strategy, and make it work in real time has become the breakpoint of video retrieval.Video retrieval technique, such as scene &shot retrieval based on abstract semantic and Video Sequences retrieval and matching, has broad usage, and one typical usage of the latter is Advertising-video sequences matching. For market evaluation, Advertiser evaluate the quality(e.g. if on time and long enough) of advertising-video on TV and the situation of their rivals. This need to find matching videos of given advertising-video from TV program video stream and do some statistics. This job used to done by manual, so it's not automatical and can't obtain the most precise description.This paper based on advertising-video matching, studied several key techniques in Image &Video retrieval, and realized a Video retrieval and matching system with high-efficiency and robustness and work in real time.First of all, adopt block method for the structure distilling image feature, thereout can use the space-position information, which can avoid the complex holistic feature and lack of space-position information. Considering the main focus content of image is at the center, and the significance decrease from inner to outer, so put forward a block-order method to depict priority of significance.Secondly, as the core of the Content-based retrieval technique, this paper put forward a new method that combined grey and texture feature. Put forward the concept of Grey-gap-order, that can describe logic relation between sub domains of block image, and with its average grey value describes the grey characteristic of block image, and reduce sensitivity to noise and linear transform, has better robustness. The texture feature adopt Gabor transform that has better vision description ability to distill texture, and it's a description of multi-scale and multi-direction. In order to decrease the complexity in distilling texture, given that reserve the major information, put forward a new texture distilling method by combining Harr wavelet and Gabor transform, use their both advantage, reduce complexity and meet the need for real time.In order to enhance efficiency of video retrieval, put forward Multi-level filtration retrieval method, that built up multi-level filtration and debar the unnecessary computing feature for unmatching image, and so save cost. This method divide the process into two stage, Coarse-retrieval and Fine-retrieval, where filtrate search space by corresponding grey and texture feature. For two unmatching images, it can judge out them as unmatching in the midway computing process and skip out with saving cost.In addition, at the field of shot divide and key frame distill, this paper demonstrate the effectiveness and simpleness of whole histogram. It has wealthy usage in real time retrieval system. Otherwise, put forward usage of clustering in the process of build up the level relation of shots and key frames, which can work with Multi-level filtration retrieval.