Adaptive modeling of macro block bit-quantization mapping for MPEG encoders

In this paper, we have presented our work on the adaptive modeling of quantization process at macro block layer of MPEG encoders. The difficulty of this modeling problem lies in two aspects: 1) the highly nonlinear nature of the relation between the quantization level and the corresponding number of bits; and 2) the nonstationaryness inherent to most of the practical video signals. We deal with the first aspect of the difficulty by using binary tree (B-tree) fuzzy interpolation algorithm for data approximation problems. The B-tree algorithm turns complicated high dimensional fuzzy decision making process into series of one dimensional fuzzy interpolations in a hierarchical manner. The difficulty of the nonstationaryness is dealt with by online adaptation which adjusts the parameters of membership functions according to domain data. Our experimental results have shown that the adaptive model has better performance than its non-adaptive counterpart as well as the lookup table method widely used and reported in the literature, which suggests that improvements can be well expected if the model is applied to the design of quantization control schemes for MPEG encoders