| With the rapid advances in communication and multimedia computing technologies, accessing mass amounts of multimedia data is becoming a reality. However, the needs for computerized content characterization of visual information have been strongly argued in recent years because of several limitations of the manual annotations and the ambiguous nature of visual information. The Content-based Video Retrieval (CBRV) technology can exactly meet the needs mentioned above and is a hot research topic in recent years,it mainly refers to getting the specific objective retrival according to shot detection and keyframe extraction,feature extraction and object retrival ect and the rusult of the retrival is not influenced by the personal's different understandings towards the video.In this thesis, several key technologies for CBVR have been investigated. For the purpose of video clips retrieval, automatic and semi-automatic methods for video shot location and shot content characterization are the two prior steps. In Chapter 3, firstly we discuss some common shot segmentation methods and study their advatanges and shotcomings through the experiments and the analysis of the retults . Meanwhile we introduce the traditional key-frame extraction methods and the methods used in this system. In Chapter 4, we study the relevance feedback algorithms used in the retrival process, and make a littlt improvement in terms of its weight-adjusting algorithm.Based on the algorithms mentioned above in this thesis we build a keyframe-based video retrieval system and discuss the composition and functions of the system modules, introduce the querying means which this system supports. A number of querying and retrieval experiments with real images are performed to validate theories above and some results are presented. In the search process, we adopt the relevance feedback retrieval methods, and through the participation of users, after repeated iterative search we get better search results to meet the needs of users. |
PDF Full Text Request