Study Of Content-Based Video Clip Retrieval

As the most complex media form of multimedia, video is applied and developed widely due to its diversiform representation, abundant semantic content, and convenient ways of recording. Secondly, the development in computing technologies, broadband communication networks, mass storage devices have resulted in large amount of video data being generated and made accessible in digital form throughout the world. As a result, the main tasks of video information processing are effective video content analysis and efficient video retrieval. Compared to current video information retrieval technologies, two key problems about video retrieval are studied: video shot boundary detection and video clip similarity measure.In terms of video boundary detection, video shot boundary detection and segmentation mainly are studied. this thesis proposes a shot boundary detection framework with collaboration of multiple detectors.The detector consists of abrupt boundary detection based on fuzzy logic theory and gradual boundary detection based on non-adjacent frame threshold method. Experimental results reveal the effectiveness of this framework.In terms of video clip similarity measure, this thesis presents an efficient measure method in which equivalence relations theory is applied to video clip retrieval. After a similarity class is defined, shot maximal match can be computed. In order to rank the similarity of the selected video clips, three different factors: visual similarity, interference and temporal order of shots are taken into consideration. Experimental results indicate that the proposed methods are effective and efficient in retrievaling and ranking similar video clips.Finally, the work of research and design in this thesis is summarized, and the future work is also discussed.