Video Shot Boundary Detection

The increasing use of multimedia streams nowadays necessitates the development of efficient and effective methodologies for manipulating databases storing this information. Moreover, in its first stage, content-based access to video data requires parsing of each video stream into its building blocks. The video stream consists of a number of shots, each one a sequence of frames pictured using a single camera. Switching from one camera to another indicates the transition from a shot to the next one. Therefore, the detection of these transitions, known as scene change or shot boundary detection, is the first step in any video-analysis system. A number of proposed techniques for solving the problem of shot boundary detection exist, but the major criticisms to them are their inefficiency and lack of reliability. The reliability of the scene change detection stage is a very significant requirement because it is the first stage in any video retrieval system; thus, its performance has a direct impact on the performance of all other stages. On the other hand, efficiency is also crucial due to the voluminous amounts of information found in video streams. This chapter proposes a new robust and efficient paradigm capable of detecting scene changes on compressed MPEG video data directly. This paradigm constitutes the first part of a Video Contentbased Retrieval (VCR) system that has been designed at Old Dominion University. At first, an abstract representation of the compressed video stream, known as the DC 194 Farag & Abdel-Wahab Copyright © 2005, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of Idea Group Inc. is prohibited. sequence, is extracted, then it is used as input to a Neural Network Module that performs the shot boundary-detection task. We have studied experimentally the performance of the proposed paradigm and have achieved higher shot boundary detection and lower false alarms rates than other techniques. Moreover, the efficiency of the system outperforms other approaches by several times. In short, the experimental results show the superior efficiency and robustness of the proposed system in detecting shot boundaries and flashlights — sudden lighting variation due to camera flash occurrences — within video shots.