Research On Video Summarization By Clustering And Mining

With the rapid development of Internet and the advance of multimedia technology, the amount of videos On the Web increases dramatically. How to help users to browse their favorite videos from such large scale video collection has become a practical challenge to be urgently solved. Recently, some of video summarization approaches were proposed to help user quickly grasp the gist of video and find their interested video.This thesis focuses on video summarization over TV news. After the video structure is generated by text and visual analysis, we conduct single-video summarization and multiple-video summarization with clustering and mining technologies.As for single-video summarization, this paper introduces video structure, shot boundary detection algorithm and several common video summary presentation schemes. Then, this paper analyzes the most popular category of video summarization methods which using clustering and mining scheme. Several video summarization methods using different clustering methods are introduced and analyzed.As for multi-video summarization, in order to effectively tackle the problem of diversity of topics and utilized different types of media data in multiple video documents, this paper proposes an approach for video summarization by clustering of complex graph. That is to say, we get the keyframe from video shots and salient words from corresponding transcript, and then construct a complex graph for clustering. After the clustering result is obtained, a so-called circular storyboard which composes of some of keyframes and keywords is generated to denote different topics as video summarization. Experiments on a corpus of CCTV Broadcast News demonstrate the effectiveness of our proposed approach.