| With the growth of the daily information, news program quantity also has increased dramatically. Therefore news video analysis, organization, and retrieve become an urgent need. The story unit segmrentation, as the basis of the news video, has certain significance to realize. This paper, taking news video as research object, discusses several problems in the process of story unit segmentation, including:anchor shot detection, topic caption extraction, audio change detection, and design news stories unit segmentation algorithm based on multi-modal.After shot boundary detection based on motion compensation, key frames in every shot are selected. This paper proposes a new detection algorithm, based on face detection and affinity propagation clustering, to obtain the anchor shots.Using the wavelet coefficients as the feature of shots, the topic caption is extracted based on support vector machine classifier.For audio of news video, Mel Frequency Cepstral Coefficients choosed as the audio features, here a detection algorithm based on Gaussian mixture model and the partition entropy is designed.This paper summarizes the existing story unit segmentation methods. Based on the anchor shots, topic caption and audio change detection results, a multi-modal segmentation algorithm is designed.Finally, using the Visual C 6.0 and Matlab R2007, this article achieves a news story unit segmentation system. Experiments show that this segmentation algorithm has satisfactory results. |
PDF Full Text Request