An L1/2-norm based efficient block level rate estimation model for HEVC

In this paper, we propose a block level rate estimation model for HEVC in discrete cosine transformation (DCT) domain, to reduce the computation burden of entropy coding in mode decision. Rather than the widely used ℓ₀ and ℓ₁-norm, the proposed model is based on the root of ℓ_1/2-norm of the quantized DCT coefficients (r-qCoeffs) which presents a higher estimation accuracy. Furthermore, to adapt to the tree partition structured coding units in HEVC, weight matrixes of r-qCoeffs are developed for different sized transform units, where each weight is a precalculated linear function of quantization parameter (QP). Benefit from the proposed model, no parameter updating is required and high accuracy can be achieved. Experimental results show that compared with the HEVC reference software, 10.68% and 5.03% encoding time can be saved in average for intra and inter prediction mode respectively, with little rate-distortion performance loss. Besides, the proposed model that possesses a concise linear form of QP is quite qualified for the rate control mode. Results show that comparable performance to constant QP encoding mode can be achieved.