The Fusion Algorithm Research Of Video And Audio Information

In recent years, with the development of the computer's informationization, more and more video equipments as well as the technology applied in people's daily life. Video conference, video search engine technology as well as video data inquiry technology and so on.which were including the massive non-text datum from the source of television, the meet in notes, the scientific literature and so on. To the one, the popularity of personal photographic equipment, and the improvement of Internet technology, so that ordinary people publish personal video recording has become extremely simple, and have consequently a large amount of video data. It is a severe test for the existing video processing technology that how to deal with so many multi-media information, how to organize data and make the index for it to retrieve.Early multimedia information retrieval algorithm has deviated from the original purpose of easy operations, future retrieval algorithms need to integrate low-level feature which contains more representative feature of the vision, hearing, semantic.The multi-modal property of video information provide a basis for information fusion. Video The most existing analysis tools for the single modal of video features, but the video is the special data which has multi-modal nature and there's a relevance about the multi-modal of the video features when it describes the same topics. Therefore need a kind of effective method to fusion video features for more accurately classify and retrieval. The major work at the process of video feature and the fusion process of video features are as follows: The current model of video data processing limited the application of news, advertising and other specific areas, and the technology used in processing is single and old. This article uses a video processing technology which is relative efficiency to define a relatively complete model of video retrieval pretreatment. In this model, used the multi-modal nature of the video low-level feature to extract the video time structure, to feature extraction for the content, and constructed the subset of video data from the original video. Based on this process to extract the key frames of video and extract audio features form video's audio streams. For simplify the calculations, to reduce dimensions for the low-level feature. We used the latest research of dimensionality reduction algorithm named the marginal fisher analysis algorithm in this paper. This method is superior to the commonly used PCA, LDA and other dimension reduction algorithms. According to obtained a variety of feature vectors, using robust classifier-support vector machine (SVM) to classify.To fusion the result of classification, we propose a fusion algorithm to improve the MGR fusion algorithm. Based on the output matrix of the sample number processed by feature vector which input classification and the integration framework of Melnik and others design, and to optimize preference and confidence, designed a fusion score function to improve MGR algorithm. The improved algorithm than MGR algorithm to reduce the computational cost, reduces the number of parameters and in the recognition rate also has some improvements.