关键词: 关键帧抽取, 压缩视频流, 自反馈定位, 运动图像专家组（MPEG）

Abstract: Key frame extraction is a premise of video content analysis.But current video key frame extraction algorithms are not suitable for mass video data processing，since they consume too much calculation time.For Internet data stream monitoring applications，a new method of key frame extraction is proposed with a detailed analysis of characteristics of MPEG compressed videos.This method takes into account the coverage of key frames extracted and the need for video dynamic testing.Key frames are extracted by segmentation based on the video length.Therefore，the position of the first key frame in the segment will be located by the feedback information，while key frames in a segment will be extracted according to the sparse coefficient.Experimental results，from video library and IDC room Internet data stream，demonstrate that this method is much more fast and effective，and thus can be specialized for high-speed network detection.

Key words: key frame extraction, compressed video, self-feedback localization, Moving Pictures Experts Group（MPEG）