Research On The Algorithm Of All Zero Quantization Block Detection In H.265/HEVC

HEVC video coding based on rate distortion optimization adopts multi-level prediction structure and advanced entropy coding to improve the coding performance,but also causes the increase of computational complexity.From the perspective of rate distortion optimization(RDO),all zero block(AZB)detection is an effective tool to reduce the multi-scale transformation encoding and mode selection complexity in the HEVC encoder.AZB detection is also considered to be an accelerated tool for fast quantification,therefore the quantitative algorithm plays a key role in AZB detection.Different quantitative algorithms,such as SDQ and HDQ,use different working mechanisms to provide a different trade-off between computational complexity and rate distortion performance.Therefore,the design of AZB detection algorithm is constrained by specific quantization algorithm.HDQ is the first choice of video encoder because of its simplified algorithm and the flexibility of hardware parallel processing.In contrast,RDOQ has been significantly improved in RD performance,but the cost is the increase of computational complexity and data dependence.In this paper,an adaptive quantization thresholding algorithm for SDQ is proposed,which is more specifically applicable to the rate distortion optimization quantization(RDOQ).Inspired by the Bayesian classification,a more accurate coefficient level zero-quantization threshold model of RDO is provided by completely simulating RDOQ.This quantitative threshold model not only has the flexibility of HDQ simple processing,but also is more convenient for the mathematical analysis and derivation modeling of AZB detection,which can be used to assist the subsequent proposed AZB detection.The traditional AZB detection scheme is usually designed with a hard decision quantization(HDQ)video encoder,and the AZB detection threshold is derived from the parameters of each block,such as SAD or SATD.However,in order to improve performance,soft decision quantization(SDQ)is preferred in HEVC encoders,and the statistical properties of statistical sets are assumed to directly replace the individual characteristics of blocks to determine AZB detection threshold.However,this assumption is not accurate.In this paper,the adaptive threshold model for AZB detection is derived by combining the local parameters of coefficient distribution characteristics with the global feature SATD of individual block.The experimental results show that the average positive judgment rate of the proposed AZB detection scheme for 4 × 4 AZB is 91.9% and 87.3% for 8 × 8 AZB,and the average FAR misjudgment rate is less than 1.9% and 3.1% respectively.The loss of rate-distortion performance is almost negligible.Therefore,the AZB detection scheme of this paper is very suitable for accelerating the HEVC coding of RD optimization.