Video Structure Analysis and Its Application to Creation of Video Summary

概要

論文の詳細を見る
In this paper, we propose scene decomposition and object image detection algorithms for video structure analysis and present its application to the creation of video summary. As for the scene decomposition, a shot which is conceived as minimum semantic unit of video is partitioned from MPEG video bitstream without any heavy decoding process. It is shown that more than 95% of decomposition accuracy is obtained in the experiment. Similar image detection algorithm from partitioned shots for stationary images like announcer shots in TV news is proposed as a simple object image detection. It is shown that only the announcer shots with subtitle can be detected by combining this similar image detection and subtitle detection. A video summary is created by compiling these shots together and about 20min news can be compress to 2min30sec abstract. We also have tried to perform automatic creation of summary by automatically extracting TV announcer shots and about 95% of news items are detected and their summaries are created.
一般社団法人電子情報通信学会の論文
1995-10-16