Static video summarization - A minimum edge weight bipartite graph matching approach

A video summary is a series of static pictures that represent the information of a video in a manner that the viewer is quickly supplied with content in compact form preserving necessary message of the original video. There are many algorithms for video summarization they represent visual information of video in concise form. Depending on how this concise form is constructed and presented the summary is static or a dynamic summary. The static summaries are constructed as group of key frames extracted from video and presented in the form of a story board. Dynamic summaries are constructed with collection of extracted key frames or some smaller segments of video and significant events and presented in the form of small video clip. This paper presents an algorithm for constructing the static summary of videos by modeling the video of 100 sequential frames as bipartite graph. The nodes of the graph represent frames of video and edges connecting nodes denote the mutual information between frames. Then matchings in resulting series of bipartite graphs are found using minimum edge weight matching algorithm. From matching in every bipartite graph modeled for group of 100 frames the frames represented by nodes connected by edge/link with minimum weight are extracted as key frames to construct a static story board. The results obtained significantly match the system used in open Video project for online video collection and repository of results of static summarization for evaluation and comparision of similar solutions. The experiments are also conducted on TRECVID specified videos. The values of performance metrics namel 97.8% compactness and 92% informativeness reveal the suitability of proposed method for static summarization of videos.