In the implementation of image coding system, the buffer design is very important as well as the coding algorithm. The buffer in the coding system has a variable rate data input and a constant rate data output. Therefore, to prevent buffer overflow and underflow we have to control data generated in the coder. In general, the desired operating conditions for the buffer are to converge rapidly and staying as close to the half full position from overflow and underflow status as possible.
In this thesis, the scene activities of each sub-block of image are classified into four categories (classes) based on the consideration of the probability and the entropy for activities in image sequences. In coding application, the normalization factors for a DCT coefficients block are determined by the class of scene activities and the buffer status.
By simulation, it is shown that the system with this buffer control method sustained a very stable buffer status, and scarcely the block effect of the reconstructed images by the system appeared in low bit rate (0.2 bit/pel). By having the stable buffer status for transmission, there was little the fluctuation of quality (NMSE) for the reconstructed images, and then this resulted in an excellent scene visually. By these results, the small buffer size can be used in hardware implementation, and the buffer control algorithm proposed in this thesis can be implemented easily.
