Classification, Content-based Sports Programs

The sportscast is one of most popular multimedia, which is in favor with great mass of spectators. The sportscast classification based on content is an important portion of the content-based video information classification and retrieval and main topic in the multimedia information research.In this paper, problems of the video information classification and retrieval are discussed, including video structuring, feature abstracting, feature optimizing and feature classifying, etc.Based on analyzing of recent researches, a video shot detection method, which is based on self-adapting dual-threshold compare, is proposed to solve the problem of video structuring, and then operations of sport video structuring are realized. The experimental results indicate that this method is efficient.Based on applying domain color, cylindrical distance and connectivity analysis to divide fields, static image features are presented for feature abstraction, such as color, texture, etc. it's also proposed dynamic video features to solve feature abstraction, which are based on applying the method of block motion estimation to build the field of video shot motion, such as motion texture, etc. These features are applied to program classification.Based on the research of abstracting typical frames of video shots, the method based on frame difference is presented to abstract all typical frames of sport video shots. And then the clustering of typical frames and the method of slow motion elimination are used to select sport shots. Finally, the Independent Component Analysis (ICA) is applied to optimize features and wipe off high-order dependency among features. The Support Vector Machine (SVM) is used to classify ball games, which achieves good experiment results.