A New approach of Rate-Quantization modeling for Intra and Inter frames in H.264 rate control

Abstract

Video encoding rate control has been the research focus in the recent years. The existing rate control algorithms use Rate-Distortion (R-D) or Rate-Quantization (R-Q) models. These latter assume that the enhancement of the bit allocation process, the quantization parameter determination and the buffer management are essentially based on the improvement of complexity measures estimation. Inaccurate estimation leads to wrong quantization parameters and affects significantly the global performance. Therefore, several improved frame complexity measures are proposed in literature. The efficiency of such measures is however limited by the linear prediction model which remains still inaccurate to encode complexity between two neighbour frames. In this paper, we propose a new approach of Rate-Quantization modeling for both Intra and Inter frame without any complexity measure estimation. This approach results from extensive experiments and proposes two Rate-Quantization models. The first one (M1) aims at determining an optimal initial quantization parameter for Intra frames based on sequence target bit-rate and frame rate. The second model (M2) determines the quantization parameter of Inter coding unit (Frame or Macroblock) according to the statistics of the previous coded ones. This model substitutes both linear and quadratic models used in H.264 rate controller. The simulations have been carried out using both JM10.2 and JM15.0 reference softwares. Compared to JM10.2, M1 alone, improves the PSNR up to 1.93dB, M2 achieves a closer output bit-rate and similar quality while the combined model (M1+M2) minimizes the computational complexity. (M1+M2) outperforms both JM10.2 and JM15.0 in terms of PSNR.