Tree Representation of Soccer Shots for Event Based Video Retrieval

In this paper, novel method based on tree representation of video shots is proposed to retrieve soccer events. The first phase consists of extracting suitable features and key frames from video shots. The features include: types of view, slow-motion and caption detection, whistle, player gathering and referee extraction. Then, using the tree representation of soccer shots the events are extracted. The proposed system is an extended method in multimodal representation and the retrieval process. Experimental results show high accuracy of proposed feature extraction algorithms and satisfying retrieval process.