Practical R-Q and D-Q models for video rate control

Abstract

Practical and accurate R-Q (rate-quantization) and D-Q (distortion-quantization) models are presented to describe the R-D (rate-distortion) relationship before encoding a frame. The R-Q model is based on a linear relationship between non-zero level count of the DCT coefficients and the generated bits, while the D-Q model comes from the observation that the ratio Q(P)(2/)D(Q(P)) can be very accurately approximated by a quadratic function of Q(P), where Q(P) is the quantization parameter used for quantization of DCT coefficients in H.263 video coding standard. Simulation results show that the proposed models estimate the real coding results very accurately.