A model based method for the quantizer assignment of JPEG-like coders

Abstract

In most coding systems, for image or audio signals, the bit allocation or quantization assignment problem is one of the important issues to minimize the distortion for a given bit budget. By exploiting the empirical distortion-quantization and rate-quantization relations, we propose a model based bit allocation method for JPEG-like coders. The empirical relations have been established using the regression technique which provides a low computational complexity to construct the preprocessing for the bit allocation. From the empirical relations, we can establish the modeling for the distortion-quantization and rate-quantization relations. Using the model, we simplify the problem of the quantization assignment and then obtain an efficient method for solving the simplified problem.