Improved algorithm for key frame extraction based on unsupervised clustering

Key frame extraction plays a very important role in content-based video retrieval. Since general clustering algorithm can only predefine a threshold in key frame extraction, this paper presents an improved method of adaptive threshold based on unsupervised clustering. The video frames texture feature is extracted based on regional segmentation. The adaptive threshold is determined by video content, and then the key frames are obtained through unsupervised clustering. This algorithm is simple and effective, it extracts key frames without predefined threshold. Experimental results of some videos with different traits demonstrate the good performance of the proposed algorithm.