Improving Retake Detection by Adding Motion Feature

Retake detection is useful for many applications of video summarization. It is a challenging task since different takes of the same scene are usually of different lengths; or have been recorded under different environment conditions. A general approach to solve this problem is to decompose the input video sequence into sub-sequences and then group these sub-sequences into clusters. By combining with temporal information, the clustering result is used to find take and scene boundaries. One of the most difficult steps in this approach is to cluster sub-sequences. Most of previous approaches only use one keyframe for representing one sub-sequence and extract features such as color and texture from this keyframe for clustering. We propose another approach to improve the performance by combining the motion feature extracted from each sub-sequence and the features extracted from each representing keyframe. Experiments evaluated on the standard benchmark dataset of TRECVID BBC Rushes 2007 show the effectiveness of the proposed method.