In video information retrieval, keyframe extraction has been recognized as one of the important researchissues. Although much progress has been made, the existing approachesare either computationally expensive or ineffective in capturing salientvisual content. In this paper, we first discuss the importance of keyframe extraction and then briefly review and evaluate the existingapproaches. To overcome the shortcomings of the existing approaches, weintroduce a new algorithm for key frame extraction based on unsupervisedclustering. Meanwhile, we provide a feedback chain to adjust thegranularity of the extraction result. The proposed algorithm is bothcomputationally simple and able to capture the visual content. Theefficiency and effectiveness are validated by large amount of real-worldvideos.