Research On HMM Based Video Content Analysis And Representation

With the rapid development of multimedia and internet techniques, content-based video retrieval has become the focus of a number of research efforts in recent years. As a crucial element, video content analysis and representation attract more and more attentions. In this thesis, we mainly develop the methods for video content analysis as well as video content representation on a semantic level. First, we demonstrate several basic problems and research interests in this area, and compare the basic methods with the statistical approaches; especially the Hidden Markov Models based techniques. Then, we introduce the Transformed Hidden Markov Model, and after the improvement of transformation matrix selection, new algorithms are advanced for semantic events detection and key-frames selection. Finally, according to the new method, we propose a hierarchical video content description scheme under the international standard MPEG-7. Experimental results show that the new methods can access and represent the video content on semantic level effectively, match the people's subjective perception well, and have the promising practical value.