Statistical Analysis And Modeling Of Rate-Distortion Function In Svc Fine-Granular Snr Scalable Videos

Jun Sun,Wen Gao,Debin Zhao

DOI: https://doi.org/10.1109/ICME.2007.4284855

2007-01-01

Abstract:In this paper, we propose a simple and effective rate-distortion model of fine-granular SNR scalable (FGS) layer in JVT scalable video coding (SVC). First, we introduce generalized Gaussian distributions (GGD) to model the distributions of the 16 (4*4) integer transform coefficients of the SVC FGS residue picture. Then we analyze the quantization scheme of SVC FGS coding, under which we analyze the distortion-rate function of generalized Gaussian model and conclude that the derivative of the distortion-rate function usually decreases to the traditional number 6.02 dB/bit as the coding rate increases. For some special cases where the shape of GGD is smooth, the derivative of the distortion-rate function increases to 6.02 dB/bit as the rate increases in the actual coding rate range. Guided by the observations, an effective and flexible rate-distortion model is proposed to approximate the actual rate-distortion function of SVC FGS layer. The average estimation error is only 0.07 dB in our extensive experiments. And our analysis and models bring us much insight into the SVC FGS coding and its rate-distortion function.