R-D spatio-temporal adaptive quantization based on temporal distortion backpropagation in HEVC

Abstract

Nowadays, Rate-Distortion Optimization (RDO) is commonly used in hybrid video coding to maximize coding efficiency. Usually, the rate distortion tradeoff is explicitly computed in offline encoder implementations whereas R(D) model are used in live encoders to select the best decisions at a lower computational cost. For sake of simplicity, this (mathematical) modelling is often performed for each coding unit (CU) individually and independently, obliterating the spatial or temporal dependency between CUs. In this paper, we provide a new spatio-temporal algorithm to compute local quantizers, based on a theoretical framework able to describe the temporal distortion propagation from an R-D standpoint. In particular, we model the temporal distortion propagation making possible the retro accumulations of any (spatial) psycho-visually weighted distortion onto reference images. Using the R(D) Shannon bound, its high bitrate approximation, and a Lagrange optimization, analytical solutions are obtained for the local quantizers and the Lagrange multiplier. The proposed algorithm shows-4.4% BD-BR SSIM gains in average over state-of-the art algorithm in HEVC, using the same SSIM-based psycho-visual function.

Temporal Adaptive Quantization using Accurate Estimations of Inter and Skip Probabilities

“…Finally, in considering the independence of rates, the R-D Shannon bound and the high bitrate assumption d = Q 2 /12, the optimal local delta quantization parameter results in . The Inter probability estimator used in [12] is given by…”

“…The optimization problem (1) is solved in [12] assuming that the bitrate constraint is fulfilled or equivalently that the average quantizer over the GOP should remain unchanged against a configuration without adaptive quantization. However, the Skip mode is ignored in the initial solution which leads to an important TBD.…”

“…We sum up all distortions and derivate the total distortion according to the local quantizer q kτ . After some manipulations, we obtain (12).…”

“…Adaptive quantization has been optimized by modeling dependencies either at the pixel level [10], transform coefficient level [11] or CU level [12], [13]. In [10], Valenzise and Ortega estimate a temporal dependency tree, further used to design an adaptive quantization based on the tree depth.…”

“…Winken et al [11] measure the dependencies between coefficients levels after Discrete Cosine Transform (DCT)/Discrete Sine Transform (DST), leading to an optimization problem solved by an iterative approach. Ropert et al [12] propose the Rate Distortion Spatio-Temporal Quantization (RDSTQ), as a generalization of Macroblock-Tree framework designed for x264 open-source encoder [13]. The main interest of this approach is to model the temporal distortion propagation between CUs from an R-D standpoint.…”