Video Scene Oversegmentation Reduction by Tempo Analysis

Video scene segmentation is a key research problem in video content analysis. However, high oversegmentation lies in most existing video scene segmentation methods. This paper proposes an efficient approach of reducing scene oversegmentation. Based on film grammar, it models and utilizes the film tempo to group the scene units generated by oversegmentation into actual scenes. Experiments based on real-world movies verify the efficiency of the proposed method.