关键词: 关键帧, 聚类, k均值, 自适应

Abstract: To make up the deficiency of key frames extraction based on k-means clustering，an algorithm based on clustering of optimizing initial centers is presented.The initial clustering centers are determined by the distributing of video data and the stability of results can be strengthened.The number of clustering k is not determined only by given value，but chose the suitable one adaptively based on video content.Experimental results show that the approach has better adaptability and the key frames are able to represent the visual content more effectively.

Key words: key frame, clustering, k-means, adaptive