LMM-based frame-level rate control for H.264/AVC high-definition video coding

摘要

Accurate distribution modeling for the DOT coefficients is greatly important for us to analyze the rate-distortion (R-D) behavior of video encoders. From the experiment, we observed that most of the existing models, paying more attention to the standard-definition (SD) videos, tend not to work well for high-definition (HD) videos. Motivated by this, in this paper, we address the statistical characteristics of DCT coefficients of HD videos coded by H.264/AVC. The contributions of this paper are threefold: first, Laplacian Mixture Model (LMM) is proposed to model the residues instead of using Laplacian or Cauchy distribution; second, the LMM-based analytic rate and distortion models are derived; third, building on the proposed rate and distortion models, a frame-level rate control algorithm is developed. Experimental results show that the proposed rate control method achieves a PSNR improvement of up to 0.85 dB compared with the rate control scheme adopted in the H.264 reference software [1]. Apart from the average visual quality improvement, the temporal visual quality fluctuation is reduced by 17%.