Study Of News Video Automatic Abstraction Algorithm

The rapid Internet growth and the multimedia information explosion urge theresearch on effective techniques of information organization, generalization andanalysis. The key of the research is the automatic summarization and inquiry ofinformation. Compared with generic video data, news video has its own specialstructure, the way of managing and retrieval. Now treating with the news video hasbeen widely investigated, and then the technique of the news video abstraction is animportant content of research. Under supported by the National Natural ScienceFoundation of China, the techniques of the news video automatic abstraction aredeveloped.On the basis of the analysis of the special structure of news videos, thepossible modes of news video retrieval are investigated. Corresponded to it, fourkinds of abstractions are proposed. They are the content-summary, the imageabstraction based on key-frames, the audio abstraction based on anchorpersonframes and the text abstraction based on closed-captions. Aiming at the applicationfor the abstraction of key frames, a new method of the possibilistic C-patterns (PCP)clustering is proposed, which is typical and immune to noise. On the basis of thePCP, a new idea of the abstract based semi-semantic and based semantic is designed,which is combined with the features of video frames represented by the dominantcolors and the subordinate colors. Through the compositive utilization of theanalysis of news video structure, the analysis of anchorperson character, audiodetection, feature matching, clustering, a new measure of circular searchanchorperson frames is proposed. Finally, with the introduction of the mathematicalmorphology in the digital image processing, the closed-caption is extracted and iscreated the text abstraction of news video according to its features.As a result, the methods of the automatic extractions in this paper are very novel,and which can not only extract the corresponding information effectively but alsopartition the configuration of news video exactly with good performances.