A new rate model based on a cubic function, and its related control schemes, is proposed to estimate the quantization scales of frames for a specified target bit rate. Our scheme is based on a distortion-rate model, which can easily be extended to a higher order for different coding algorithms without requiring much effort to determine the quantization scales. The formulation of the cubic model is derived from the quadratic rate-distortion model and from simple curve fitting. Our model is implemented in H.264 JM8.1a, and the performance is compared to JM8.1a with different test sequences and bit rates. Experimental results show that our proposed model outperforms the quadratic model in terms of both PSNR and bit rate.
Published in:
Circuits and Systems, 2005. ISCAS 2005. IEEE International Symposium on
Date of Conference: 23-26 May 2005