Content Based Video Boundary Detection And Key Frame Extraction

With the rapid development of digital media, online communications and other information technology, video and other multimedia technologies emerged explosive growth. The data structure of video is three-dimensional, while with the feature of large amount and rich information resulting great demand of technology for video management and retrieval. Traditional video retrieval need manual labeling as it is based on keyword, so large time consuming is needed and it tends to be influenced by artificial subjective. As text cannot fully describe the content of video, so contend-based video retrieval is needed, video will be analyzed automatically for users to retrieval by image.Shot boundary detection and video key frame extraction is necessary for content-based video retrieval, shot is the basic unit for retrieval. Pre-processing is taken to detect shot boundary, adaptive threshold is used to select candidate segment which may contain shot boundary, begin frame will be detected firstly, then cut will be detected immediately, if cut is detected successfully, the detection process is completed; gradual transition will be detected if no cut exist in the candidate segment. Detection time can be reduced by compared with the traditional method.Image retrieval is the last stage for video retrieval, key frames represents the content of video. C lustering based time is taken in this paper, frames are clustered into different clusters, each frame is similar to other frame in the same cluster, one key frame will be selected by entropy, key frames of shot will be extracted from sub shot.Finally, experiments are carried out about different kinds of video, the result prove that pre-processing method is useful for video shot boundary, while the clustering method based time sequence is useful for key frame extraction.