Mutual information based video shot boundary detection

A video sequence is usually composed of multiple shots, which are the basic units for higher-level content analysis, indexing, editing and so forth. As such, shot boundary detection is usually the first step toward content-based video applications. Correspondingly, many researchers have worked in this field and proposed many good algorithms. However, there is still much room for improvement in gradual transition detection capability and computational efficiency with those algorithms. The proposed method improves upon the graph cut algorithm based on mutual information for better detection accuracy and efficiency. In addition, we employ hierarchical analysis both spatially and temporarily on the video data, leading to reduced computation and improved gradual transition detection accuracy. Experiments on the public test video data demonstrate that the proposed algorithm improves both the efficiency and the accuracy of shot boundary detection, when compared with related works.