Research Studies Project Report (M.Eng.) - Asian Institute of Technology, 2002

A research study submitted in partial fulfillment of the requirements for the degree of Master of Engineering, School of Advanced Technologies

Effective video indexing method is a must when working with huge video databases. While manual indexing is considered as the most effective approach, it is not cost-effective and the speed is very slow. Thus, automatic indexing is desirable and recently various indexing tools for video databases have been developed. With the birth of MPEG-4 with object-based functionalities, the need to detect abrnpt scene changes is becoming a must. This research study introduces a new scene change detection method that would speedily and effectively detect abrupt scene transitions in MPEG-4 video sequences in the compressed domain utilizing features such as motion and texture information. Simulation results showed that the proposed MPEG-4 abrupt scene change detection method can detect abrupt transitions in the real-time scenario. Comparisons showed that 79.11 frames per second can be processed using the proposed method while for its uncompressed counterpart, only 0.588 frames can be processed in a second. Furthermore, using texture information to detect abrupt transitions resulted to effective scene transitions detection with a mean recall and precision values equal to 84.72% and 85.96%, respectively. On the other hand, using motion information to detect abrupt scene changes in MPEG-4 compressed domain resulted to a mean recall value of 86.57 percent and a precision value of 76.28 percent, an acceptable value for both recall and precision.