Key Technology Research For Content-based Video Abstract Generation

Along with the continuous development and progress of computer network and Internet industry,video data also show a trend of rapid growth.How to manage and organize large amounts of data of video effectively has become the research hot topics in the field of video retrieval issues.To solve this problem,content-based video retrieval technology is constantly developing to make people catch the key information through browsing the image sequence or label,increase the efficiency of browsing and searching.From the establishment of the video data to the image sequence is a process of video multi-stage stratified,segment the video data to clips and extract key frames are the important steps of video summary generated.Moreover,the research of these two steps are the key techniques in the field of content-based video retrieval technology.In this paper,the main work are video clips segmentation and key frame extraction,to split the video data to multi-level partition,the summary information is generated.After analyzed and summarized the typical algorithms,a new video clips segmentation method is proposed.Through the idea of multiple feature fusion,combine LBP feature and GIST feature for pre-fusion and post-fusion.Both global and local features are considered to describe the image.Intermediate descriptor is proposed,it comprehensively characterizes local temporal structure of the video,also apply the random forest in video clips segmentation.Turn the video clips segmentation problem into a classification problem and compared with static and adaptive thresholds approaches in video segmentation,the best performance can be achieved through the experiment.After segment the video into clips,key frames are extraction from each clips.In this paper proposed a method combined mutual information and the information entropy to extract key frames,segment the clips into sub-clips then find the candidate classes from sub-clips and extract key frames.Remove redundant frames by information entropy,to extract the appropriate number of key frames adaptable and represent the main content of video very well.The experiment proved the key frames we extracted have very good representative significance.Connected the key frames to achieve the summary information of the video.