Abstract:

To reduce the coding complexity of the rate-distortion optimization process of the latest video coding standard versatile video coding (VVC), a fast rate estimation model based on statistical modeling was proposed.Firstly, the quantization behavior in dependent quantization (DQ) and the context dependency in entropy coding were fully considered.Features that could describe context state transition in the coding process were proposed to estimate rate of some synatax elements in a TU preliminarily.Secondly, coefficient chaos and sparsity features were proposed to distinguish the influence of coefficient distribution difference on the rate cost based on the coefficient distribution characteristics which built a TU level rate model.Finally, large-size transform unit (TU) and small-size TU was modeling respectively according to the rate composition character to achieve more accurate rate estimation.A large number of parameters were trained by regression model through statistical methods, and the final linear rate model was obtained which was applied to the mode decision.Experimental results show that the proposed algorithm can achieve 16.289% complexity reduction with 1.567% BD-BR increase for RA configuration.

Key words: rate estimation, VVC, RDO, regression training