Study On Jensen-inequality-based Key Frame Extraction Method

Key frames are the subsets of all the video frames that could represent video content and with less redundance. In this way, key frames keep product and human living from the heavy load of the large video data. This thesis proposes a generic approach in which the Jensen-Shannon divergence(JSD), Jensen-Renyi divergence(JRD) and Jensen-Tsallis divergence(JTD) are investigated for measuring the difference between neighboring video frames. By measuring the distribution difference of color histograms between two neighboring frames, and by setting thresholds of some certain image features, we detect shot and sub-shot boundaries, segment a video clip into shots and sub-shots, and then choose a key frame from each shot and/or sub-shot. The selected key frames are arranged by time sequence and could be used to summarize and retrieve a video. In this thesis, we studied the adaptiveness of different information-theoretic-based key-frame extraction methods applying on different single types of video sequences, and got a positive conclusion for the future work for key frame extraction task. Our novel Information-based approach is computationally inexpensive and yet effective, as shown by large amount of experimental results.