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

Accurate distribution modeling for the DCT 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%.

► Using Laplacian Mixture Model (LMM) to model the residues instead of using Laplacian or Cauchy distribution.
► The LMM-based analytic rate and distortion models.
► The LMM-based frame-level rate control algorithm.
► Up to 0.85 dB PSNR improvement over H.264 reference software.