Recently, very low bit rate coding schemes have been proposed by many researchers in video sequence coding applications. Most of them tried to compress the video data in order to maintain the high quality of a whole frame in video sequence. But they did not consider a relative importance of information between interest region (target region) and background region. So, if the capacity of transmission channel is very limited, it is difficult to use them in general tactical scene, which has global motion by camera movement. To overcome this problem, a new approach, selective coding, was introduced in 1993. In this scheme, they did discriminately encode target from background region considering degree of importance of each region in a frame.
In this paper, we propose a novel selective coding algorithm which is a two layer selective coding scheme for video signals as follows. First, we separate a frame into two regions which are target and background region by using the motion information between previous and current frames. This is performed by a moving target segmentation algorithm. In addition, we estimate global motion by camera movement in this stage. Next, we go to coding stage which is composed of two coders: The First Layer Coder(FLC) for background region and the Second Layer Coder(SLC) for target region. The coder for background region is called FLC which uses a motion-compensated predictive coding technique. Since background region does not include important information relatively, we assign fewer bits to the coding of the region by using a subsampling technique and controlling the quantization step size. Meanwhile, in SLC, we use a motion-compensated predictive coding technique. The motion of the target region is estimated between the current and the global motion-compensated previous frames at segmentation stage and it is a sort of region-based motion estimation. Moreover, since we can assign more bits to this region, the quality of the target region can be maintained higher than that of background region. SLC is controlled by a flag signal which indicates the existence of target region. And for prediction, the coder use the previous reconstructed frame from FLC.
Simulation result shows that the average PSNRs are obtained about 45.05 dB and 26.02 dB for target and background region respectively. Also the average number of bits generated in a frame is about 4780 bits: 2.16 bits and 41.184 bits per block are assigned to background and target region respectively. But if the bits assigned to target region are reduced and the reduced bits are assigned to background region, the quality of background region can be improved more or less. On the other hand, the quality of target region is apparently decreased. Another merit of our coding scheme is that it can be easily extended to multi layer coding. That is, when several target regions exist, we can discriminately encode the regions with a degree of importance of each region using multi layer coding.
