Research On Method Of Content-based Video Retrieval

With the development of multimedia and network technology, there are more and more multimedia information. However, it is becoming the emergent and important issues for efficient organization, representation, store and management of video data as well as rapid retrieval and browsing of video data. The traditional data management and retrieval methods can not do better to search the requiring information from the huge amount of video data because of the diversity, complexity of structure of video data, and space-time and multi-dimensional structure of video data. Therefore, the content-based video retrieval (CBVR) system emerges.CBVR is to extract the features that can represent the video content and to retrieval the relative video streams from the enormous video database through matching of patterns based on content and context of the video and video analysis. This paper focuses on the key techniques and methods of video retrieval.1) According to the analysis of video data, a tree structure model of video data is established. In this model, video is represented in a hierarchical structure with several layers from representative shot, shot clustering, shot, key frame, frame clustering to frame. The structure model makes up a basic for further processing.2) An improved adaptive shot segmentation method based on AIHFD(absolute intensity histogram frame difference) is proposed. The approach contains two processing modes. One is dedicated to detection of abrupt shot, and the other for detection of gradual shot. These two modes are chosen automatically by comparing intensity histogram frame difference and adaptive threshold. The next frame by frame difference detects flashlight in the process of the detection of abrupt shot. The variance of frame difference detects gradual transition. Experimental results show that the proposed algorithm obtains better results and computing complexity low, easy to implement.3) New approaches for representative shot and key frame extraction are proposed. The methods can control the quantity of the extracted representative shots and key frames. It is computed simply and has high exactness while avoids redundancy. In this paper, a new method of cluster is presented. It takes the extracted representative contents as the center of clusters. Then according to similarity estimation, it is classified as a cluster that has the largest similarity with representative contents. The experimental results show that this algorithm is better express the decomposition of the video content.4) Based on the video decomposition of tree structure, the methods of key frame-based video retrieval and shot-based video retrieval are proposed. The characteristic of video decomposition tree structure hierarchy is used for video retrieval to improve the accuracy of retrieval. Experimental results show that the proposed methods could effectively and efficiently retrieve.