Research On Sports Video Classification Based On Feature Fusion

Owing to the rapid development of network and multimedia technology, it becomes very convenient for users to watch various kinds of videos on TV phone, car television and other medium anywhere, which enriches people’s life greatly. However, it is also very difficult for users to seek out what they want from a mass amount of video data. Classification of video is conducive to increasing the speed of retrieval.Due to the high error rate of manual annotation classification, automatic video classification therefore becomes more and more important in the multimedia field area. As the part of videos, automatic sports video classification can not only help coaches find associative video data to train athletes, but also contribute to common user finding out their interested sports video clips. Therefore the automatic classification of sports video is an important branch of video field area.Video features play an vital role on classification. It is difficult to depict the whole property of video. Aiming at this problem this paper presents two new fusion models and a method of automatic classification algorithm, then the new features and support vector machine are used to classify sports video. The new method got an ideal effect. The detailed steps as follows:First, this paper proposes the advantages and disadvantages of the existing video classification algorithms after careful analysis.Second, in order to evaluating the effectiveness of the proposing algorithm, a database is built, which contains six type including ball games:tennis, badminton, table tennis, football, snooker and tennis. The duration of the database is480minutes and there are953video clips.Third, this paper proposes a new nonlinear fusion model of the color and texture of sports video. First, a generalized linear quantitative model is built on the HSV color space, and the color index matrix is used to represent the visual content of video frames, then, the color co-occurrence matrix texture is defined. The matrix contains the color and texture information of video frame.Compared with the single use of color, gray texture or a simple combination of them, it improves the ability to express visual content, as well as the classification precision of the designed algorithm.Finally, this paper addresses a new automatic classification algorithm of sports video using a nonlinear fusion feature of color and edge. The algorithm introduces the edge information based on the three color channels of RGB color space, and combine color into edge to construct a model of color edge intensity and direction. At last, the intensity and direction of color edge are used to classify sports video. The result of experiment shows that the color edge feature is effective in the classification of sports video.