Behavior key frame extraction using invariant moment and unsupervised clustering

Key frame extraction technique plays an important role in video analysis and content-based video retrieval. Key frame has been used to reduce the use of video indexing data greatly and it also provides a framework to video summaries and retrieval. This paper proposes a novel method based on invariant moments for key frame extraction, according to the changes of independent objectpsilas shape and brightness. We first extract a moving object from the video sequence and compute the invariant moments in the area. Then cluster consecutive frames of similar invariant moments by unsupervised clustering. Finally, extract the typical data from each cluster as the key frame. The experimental results of different scenes show the feasibility of this method.