Adaptive Detection Of Scene Change For MPEG Videos Based On Sliding Windows Modeling

There is a great amount of multimedia information related to many applications such as computer, TV, internet, DVD, and digital libraries. The multimedia information needs a lot of devices to store. On the other hand, it is difficult to find out the information which you want from considerable multimedia information. Therefore the automatic video indexing and retrieval have become more popular in video applications. Scene change detection plays an important role in video processing with many applications such as video indexing, video object tracking and video content management.Over the last decades, scene change detection has been widely studied and researched. As a result, many scene change detection techniques have been proposed and published in the literatures. For our convenience of surveying existing research in the field of this subject, all these algorithms and techniques can be broadly classified as pixel-domain-based and compressed-domain-based.In the dissertation, we propose an effective and fast scene change detection algorithm based on compressed domain. The proposed method of scene change detection exploits the MPEG motion estimation and compensation scheme by examining the prediction status for each macro-block inside B frames and P frames, and no feature extraction or histogram differentiation is needed. A sliding windows mechanism is employed to detect the precision location of scene change. The proposed algorithm can work in compressed domain, and especially suitable for real-time applications. The proposed algorithm achieves superiority measured by recall and precision rates compared with existing methods.