Scene change detection method for MPEG video

Abstract

Indexing and editing large quantities of video material is becoming an increasing problem today, especially when video acquire technology made video archiving easy. Manually indexing video content is currently the most accurate method but it is a very time consuming process. An efficient video indexing technique is to temporally segment a sequence into shots. A shot is defined as a sequence of frames captured from a single camera operation. A small subset of frames can be used to retrieve information from the video and enable content-based video browsing. To minimize the cost of time for video indexing and editing, automated shot cut (scene change) detection is used. This paper will introduce two types of scene change detection method based on frame by frame comparison. Both methods apply for MPEG-1 video stream. First approach will use the nature of edge continuity within video. Second approach uses grayscale level of extracted frames from MPEG video stream. Both techniques use image processing tools for frames analysis and comparison. Scene change decision is made with the reference of several image processing operation and the threshold value. The performance of detection will be evaluated on detection precision and false alarm. At the end, both methods will produce the result of frame number where scene change is detected. These values will be use for video indexing process.