Research Of Content-Based Video Retrieval's Algorithm

With the rapid development of multimedia and network technique, efficient and effective video analysis and retrieval become more and more important due to teeming of vast video information. Hence, it comes forth a new research field in the information technology, i.e. Content-Based Video Retrieval (CBVR).CBVR means video retrieval from enormous video database through matching of patterns from video analysis and feature extraction to represent its content. In a video, the contents include its structure, low-level visual and aural features, and high-level semantic information, which form the base for further video comprehend and analysis. The main work in this thesis is video temporal segmentation, each video sequence could be hierarchically segmented as: Scenes,shots and key frames.Shot detection is the most fundamental operation and very important step in CBVR. Shot detection is a technique that divides video data into physical units, generally called shots. So the accuracy and execution speed of shot detection is critical. Approaches to shot segmentation can be classified as those that operate in MPEG compressed domain and those in uncompressed domain, shot change can be classified abrupt change and gradual change.The main tasks of this paper are:(1) Introducing the main technical characteristics of content-based video retrieval system, and the highlights of the video content-based data model and a common video structure data model.(2) Introducing the pixel domain and the compression domain shot segmentation method. In the pixel domain, on the basis of traditional change-step shot segmentation algorithm, we give an improvement of this algorithm, which can achieve the best effect of shot segmentation.Abrupt change detection method is unchanged, integrating bidirectional threshold way into the process of gradual change. About the expression of the shots, this paper present several key frame extraction methods, and gives the key frame extraction results of the evaluation criteria. (3) On the basis of the shot segmentation and key frame extraction, this paper, we researched on Scene segmentation method, which is based on the shot cluster. Based on the border in the shot clustering algorithm, we proposed the method of using the camera frame color image characteristics and the overall situation of movement to the scene segmentation. And using Visual C++ 6.0 achieved a simple video retrieval system.