An integrated baseball digest system using maximum entropy method

摘要: In this paper, we propose a novel system that is able to automatically detect and classify highlights from baseball game videos in TV broadcast. The digest gives complete indexes of which cover all the status changes game. We achieve by seamlessly integrating image, audio speech clues using maximum entropy based method. What distinguishes our previous ones emphasize on integration multimedia features acquisition domain knowledge through machine learning process. Integration important because with current state-of-the-art image analysis techniques, most can extract are very low level, detecting/classifying sports single medium doomed yield poor performances. Acquiring process preferred over heuristic rules more powerful for discovering expressing knowledge. perform extensive experiments including various stadiums, teams broadcasted different stations.