Study On Shot Detection System In MPEG Compressed Video

Along with the rapid development in computer, multimedia and networking technologies, digital video applications become more and more popular. How to fastly and efficiently retrieve the extremely large amount of video data is the key problem. Content-based video retrieval is developed and becomes the issue of multimedia research.Shot detection is the most fundamental operation and very important step in content-based video retrieval. Shot detection is a technique that divides video data into physical units, generally called shots. Its precision directly affects the retrieval results in content-based video retrieval system. So the accuracy and execution speed of shot detection is critical. With the advance in data compression, digital video sequences are usually stored in a compressed format such as MPEG How to manipulate these compressed video becomes more important, In this thesis, shot detection is based on MPEG-1 video.Firstly, this thesis introduces the basic theory of video compression and MPEG-1 video compression standard. Secondly, shot detection is studied on uncompressed and MPEG compressed video streams and the experiment results are also given. Finally, a shot detection system based on MPEG-1 video streams is developed in Visual C++ 6.0 and DirectX SDK on windows system. The sliding window algorithm is chose to detect both abrupt and gradual shot hi system, and a fast abrupt shot detection algorithm using direct feature extracted from MPEG-1 compressed video is also adopted. The experiment results show the abrupt detection is excellent and the gradual shot such as sliding, fade in and fade out, wipe etc. can also be detected. The system has several features as follows:1) Abrupt and gradual shot can be detected on uncompressed video. The system need not the third software which used to decode video and store each frame in the hard disk, however, the DirectShow is introduced to deal with MPEG-1 video bit streams. So the system is universal.2) The shot can be detected directly on MPEG-1 compressed video without full frame decompression. The detection speed is improved obviously and it can achieve the real-time processing performance.3) The key-frame can be extracted and shown in system. Users can browse thesekey-frames in order to get the video content quickly.4) The interface is so good that it can meet requires of content-based multimedia retrieval system. The system can play video and give some useful detection feedback in detection process.Possible extension and new directions of the research is also given in thesis.