A keyframe extraction method based on transition detection and image entropy

With the massive growth of video data on the internet, users need a strategy to quickly browse video content. Extracting the key information from the redundant information of video data is a hot topic in recent research. There are two kinds of video summarization: static video summarization and dynamic video summarization. A set of keyframes is a form of a static video summary. Through a set of keyframes, users can understand the main content of the video. In addition, keyframes can also be used in video retrieval services. In this paper, shot segmentation based on transition frame detection is realized, and then, keyframes based on video shots are extracted. There are different processing strategies for long and short shots. The frame with the most information is selected from each segmented shot as the keyframe, and finally, the key frame is made nonredundant by using visual features. In this paper, the proposed strategy is tested on the OpenVideoProject dataset and compared with VSUMM and other keyframe extraction methods.