|
Part I: Object-Based Video Coding Nowadays, conventional approaches are reaching a saturation point so that object-based approaches arise more and more attention and interest. We have developed a novel method toward very low bit rate application which is object-based. Decomposing a image into several objects simulates the function of human vision and helps to decrease the number of the motion vectors compared to MPEG. we have also proposed a method to solve some drawback of object-based motion compensation in this part. As to residue coding, that is, prediction error coding, block-based transform method in conventional algorithm is replaced by a segmentation method to eliminate the artificial blocky effects. At last part of this part, some simulation results are shown and the performance is discussed. Part II: Compressed Image Scene Analysis In the near future, efficient indexing methods will be required to handle rapidly increasing visual information, especially when visual information systems such as video databases are established. Video analysis that partitions the video into clips or extracts interesting frames is a very important initial step for video indexing. In this part, we have developed a novel method for video analysis using the macroblack type information in the MPEG framework. This method exploits the comparison operations performed in the motion estimation procedure in the MPEG framework for detecting the scene changes resulted from motion estimation will follow some specific pattern when scene changes occurs or some effects are applied. Experiments show that our approach can perform very fast scene change, flashlight and caption detection.
|