SSIM based perceptual distortion rate optimization coding

The current rate distortion optimization (RDO) coding schemes usually use the Sum of Absolute Difference (SAD) or Sum of Square Difference (SSD) as the distortion metric. However, neither SAD nor SSD correlates with the human visual system (HVS) very well. To develop a perceptual distortion based video encoder, we employ the Structural Similarity (SSIM) Index as the distortion metric and propose a SSIM based Lagrangian perceptual distortion rate optimization (PDRO) method in this paper. Furthermore, to adapt the different input sequences dynamically, we present an adaptive Lagrange multiplier selection scheme based on the properties of the input sequences. By modeling the transformed residuals with Laplace distribution, the statistical SSIM and rate models are deduced to obtain the adaptive Lagrange multiplier. Extensive experiments show that the proposed scheme can achieve better perceptual distortion rate performance and provide better visual quality than the SAD/SSD based RDO coding scheme.