Research On Video Shot Segmentation Technology And Content-Based Video Retrieval System

With radip development of multimedia and computer network technologies, capacity of digital and video data is increasing at an amazing speed, so how to organize, manage and utilize the video data in an effective way is of great importance. Accordingly, content-base video retrievalhas become one of the mot active research subjects in the multimedia field. Unfortunately, a number of problems with content-based video retrieval still remain to be overcome up to now.As a first step in video processing, video shot segmentation acts as a basis of subsequent analysis, classification, indexing and searching operations on content of higher levels. Accuracy of the video shot segmentation will have a direct influence on effect of the subsequent processing operations. Therefore, in view of abrupt changes and gradual changes of video shots, this paper proposes a method for adaptive real-time camera shot detection based on interframe difference and difference between non-adjacent frames. More specifically, this method accomplishes effective detection of abruptly changing shots by calculating difference between adjacent frames, and accomplishes effective detection of gradually changing shots by calculating difference between non-adjacent frames. These two operations in combination allow for detection of nearly all abruptly changing shots and gradually changing shots, thus achieving better comprehensive detection effect.Further, this paper also proposes a video retrival method based on histogram and key frame technologies, a core concept of which is to classify all frames in a clip of video into camera shots, scenes and story units and derive a structure chart in tree view. When desired to retrive a certain video frame or shot from a video library, the user may follow a top-down video analyzing method instead of the traditional frame-by-frame searching. The user is allowed to start the retrieval from the highest level of video, i.e., the story units. In more detail, a story unit where the target image lies is searched at first, then the scene, and at last the corresponding shot is found, from which the desired video frame can be retrieved. This technology is advantageous in that it overcomes the blindness of traditional retrieving technologies by setting the target to the scene level directly, thus reducing the time necessary for the retrieval and improving the retrieval efficiency.This paper has four sections. The first section discusses the current status of researches on content-based video retrival and the primary targets of the subject. The second section introduces relevant knowledge of the histogram technology, and on basis of this, introduces the color matching method which improves the retrieval effect by reflecting the spatial characteristics of video images. The third section focuses mainly on the method for adaptive real-time camera shot detection based on interframe difference and difference between non-adjacent frames proposed by the author. The fourth section focuses mainly on a content-based video retrieval system proposed by the author. At last, a conclusion is made on the research of this paper and a prospect is made on the further plan.