The Research of the Shot Boundaries Detection Based on Textual Features
|School||Beijing University of Posts and Telecommunications|
|Course||Communication and Information System|
|Keywords||Content-based video retrieval system Shot boundary detection Critical frame detector Differential box method|
With the rapid development of computer technology and network communication technology, video and other multimedia data has gradually become the main form of media in the field of information processing. And the continuous expansion of the network and the continuous development of the access technology, provides a broad space for the wide dissemination of video and other forms of multimedia. Increasingly rich multimedia data, especially video data, is not only limited to the operation of video data storage and transmission, retrieval application is accepted by more and more people. Retrieve the required information from the massive video data is becoming increasingly important, but the inefficiencies and limitations of traditional retrieval method makes it needed a new retrieval method that can handle video data. The traditional video retrieval method of the video through manual annotation, then label content to be retrieved. This method is the face of the rapid emergence of video content has a lot of difficulties, so the content-based video retrieval technology (CBVR) came into being. This paper studies - shot boundary detection based on key technologies in video retrieval. From the uncompressed domain and compressed domain presentation summarizes the existing shot boundary detection method and analysis of the main problems which exist and texture features applied to the establishing shot boundary detection system; then introduced now some variety of texture feature extraction methods, analysis and comparison of fractal dimension as texture features used in this paper, shot boundary detection system; subsequently proposed an improved keyframe detection method based on unsupervised clustering, last a video shot boundary detection system based on texture features. The video shot boundary detection system consists mainly of three parts - the image pre-processing, image feature extraction, shot boundary detection and key frame extraction. The image preprocessing main complete image feature extraction part filtering and sub-blocks of each frame of the video clips; differential box each frame texture feature extraction; video shot boundary detection and key frame extraction part by a sliding window and The dynamic threshold extract video clips shot boundary and key frame extraction using an improved unsupervised clustering method. Finally, simulation experiments of this paper, the video shot boundary detection system, experiments show that the boundary of the system for all kinds of video, especially gradient has a good robustness, while different efficient extraction of video content keyframes.