An efficient all-zero block detection algorithm for high efficiency video coding with RDOQ

Abstract

Rate distortion optimized HEVC video coding is challenged by the extremely high complexity caused by hierarchical prediction structure and advanced coding tools. All-zero-block (AZB) detection is an efficient tool to decrease the complexity of mode decision in HEVC coders in the sense of rate distortion optimization (RDO). Traditional AZB detection schemes were usually designed by employing hard-decision quantization (HDQ), and the AZB detection thresholds were generally derived according to the individual block’s characteristic parameter such as SAD or SATD. However, soft-decision quantization (SDQ) is preferably employed in HEVC encoders for coding performance improvement, and the AZB detection thresholds are supposed to be determined according to the ensemble’s statistical characteristics in the sense of statistical inference. This paper proposes an adaptive AZB detection algorithm well-suited for SDQ, more specifically RD optimized quantization (RDOQ). Inspired by Bayesian inference, a more accurate coefficient-level zero-quantized threshold model in the RDO sense is proposed by fully simulating RDOQ to aid in the following AZB detection. In addition, a local parameter depicting the individual block’s characteristics regarding the inter-coefficient distribution is jointly combined with SATD to derive an adaptive threshold model for AZB detection. The experimental results demonstrate that the proposed algorithm detects 91.9% 4×4 AZB with smaller than 1.9% FAR and 8.1% FRR false alarm rate on average, at the cost of negligible rate-distortion performance loss. This work is well-suited for fast RD optimized HEVC coding.