A Novel Video Summarization Framework for Document Preparation and Archival Applications

With the rapid growth of network bandwidth and high-capacity storage devices, videos have become an important way of communication in the aerospace industry and many other entities. However, browsing and managing huge video databases are quite tedious. To solve the problem, in this paper, we propose a novel video summarization framework, and discuss its potential usage in the document preparation and archival applications. The proposed framework generates video skimmings that guarantee both the balanced content coverage and the visual coherence. First, we segment the raw video into video shots, analyze the structure of the video, find the boundaries of semantic scenes, then calculate each scene´s skimming length by its structure and content entropy. Second, we define a spatial-temporal dissimilarity function between video shots, model each video scene as a graph, and find each scene´s optimal skimming shots in the graph with dynamic programming. Shot arrangement patterns are analyzed to improve the coherence of the video skimming. Finally, the whole video´s skimming is obtained by concatenating the skimmings of the scenes. Our proposed framework generates video skimmings that guarantee both the balanced content coverage and the visual coherence. Experiments are conducted to evaluate the effectiveness of our proposed approach