Experimental comparison of existing video shot detection techniques in compressed video

Video shot detection is an important first step in video parsing that has been applied in many recent content-based video indexing and retrieval systems. A number of algorithms address the problem of detecting video shots automatically. Some algorithms work with decompressed digital video, while other more state-of-the-art approaches try to detect shots in MPEG-1,-2 compressed video streams directly in order to reduce the processing time. This paper reviews the main approaches to detecting shots in MPEG-1,-2 coded video and compares these using experimental results derived using extracts from the National Institute of Standards and Technology (NIST) video test data-set. These results show that the approach using DC images outperforms other more complex approaches based on macroblock coding reviewed in this paper in terms of precision and accuracy.