Study on distortion-quantization models based on DCT coefficient distribution models

摘要

In H.264/AVC, rate control and mode decision play important roles for efficient video compression. Usually, rate control employs R-D model to select the quantization parameter for precise bit rate control. R-D model which includes rate quantization (R-Q) model and distortion quantization (D-Q) model can be derived from mathematic modeling based on distribution of DCT coefficients. Therefore, the DCT coefficients distribution models are very important for D-Q modeling. Currently, some D-Q models have been proposed based on these three models, such as Laplacian, Cauchy and Generalized Gaussian Distribution (GGD), but there is no systematic and rigorous comparison on their accuracy. In this paper, we will have an analysis on three DCT coefficients distribution models. Also, we will test several existing D-Q models and give fair comparison on these models. This work is meaningful for efficient video coding algorithm optimization in the future.