Statistical model, analysis and approximation of rate-distortion function in MPEG-4 FGS videos

In this paper, the generalized Gaussian distribution is employed first to model the DCT coefficients of image data from MPEG-4 fine-granularity scalability (FGS) frame. Then, according to the quantization theory, the distortion-rate function of the generalized Gaussian model is analyzed and it is concluded that the derivative of the distortion-rate function first decreases, and then increases up to the boundary of 6.02 as the bit rate increases. For actual FGS coding, the derivative of actual distortion-rate function usually decreases as the rate increases, and then begins to increase slowly at a comparatively high bit rate. Finally, based on above observations, a rate-distortion (R-D) model is proposed to approximate the actual distortion-rate function. Experiments show that the proposed R-D model is accurate and flexible.