Multimodal Fusion News Story Unit Segmentation

Under the rapid increasing quantity of video data, it is becoming a burning practical question of dealing with, browsing, indexing and managing of video data effectively. Video content analysis technology aims to change the non-structuring video data to the structuring one, and to extract semantics from it. So it bridges the semantic gap between low-level features and high-level concepts, and to provide an accessible way to organize and manage video data by finally construction of the application systems as video abstract, indexing and retrieval, etc.In this dissertation, news video as the research object, through the caption, audio, video, three different types of information studies, implemented on the news video story unit segmentation. And to Visual C++ as the development platform, designed and implemented to achieve a News Story Segmentation System unit.This paper first the structure of video processing, an improved shot boundary detection method, that is the same pixels after motion compensation using adaptive dual-threshold point to determine the shear lenses and gradient lenses. On this basis, improve a caption detection algorithm using wavelet transform to extract image features, their fuzzy C means clustering, to determine the image frame in the caption information; the same time to extract the key frame-based A chair human lens detection method, using news video camera host the inherent characteristics of the main extraction region color characteristics of images to judge, will host the lens into camera lenses and general reports.In this paper, hidden Markov model-based audio classification algorithm to extract the video in the bottom of the audio information features, the first judge on the audio mute, and then use the Hidden Markov Model and classify, to realize the classification of information and audio information recognition.Finally, with Visual C++ 6.0 as the programming platform to realize the news story unit segmentation system, completed a news video shot detection, caption detection, audio classification, the host shot detection, news story segmentation, and database management unit and search.