Method and system for structural similarity based rate-distortion optimization for perceptual video coding

摘要

There is disclosed a system and method for video coding, and more particularly to video coding that uses structural similarity (SSIM) based rate-distortion optimization methods to improve the perceptual quality of decoded video without increasing data rate, or to reduce the data rate of compressed video stream without sacrificing perceived quality of the decoded video. In an embodiment, the video coding system and method may be a SSIM-based rate-distortion optimization approach that involves minimizing a joint cost function defined as the sum of a data rate term and a distortion functions. The distortion function may be defined to be monotonically increasing with the decrease of SSIM and a Lagrange parameter may be utilized to control the trade-off between rate and distortion. The optimal Lagrange parameter may be found by utilizing the ratio between a reduced-reference SSIM model with respect to quantization step, and a data rate model with respect to quantization step. In an embodiment, a group-of-picture (GOP) level quantization parameter (QP) adjustment method may be used in multi-pass encoding to reduce the bit-rate while keeping similar perceptual video quality. In another embodiment, a frame level QP adjustment method may be used in single-pass encoding to achieve constant SSIM quality. In accordance with an embodiment, the present invention may be implemented entirely at the encoder side and may or may not require any change at the decoder, and may be made compatible with existing video coding standards.

主权项

1. A computer-implemented method of video coding with rate-distortion optimization, comprising:
estimating, by a processor, a derivative of the structural similarity (SSIM) quality measure with respect to a quantization step Q using a reduced-reference SSIM model that utilizes a source video only; estimating, by the processor, a derivative of the data rate R with respect to the quantization step Q; minimizing, by the processor, a joint cost function defined as a sum of the data rate and one minus SSIM; utilizing, by the processor, a Lagrange parameter to control a trade-off between the data rate term and the SSIM term; determining, by the processor, an optimal Lagrange parameter based on a ratio between the estimated derivative of SSIM with respect to Q and the derivative of R with respect to Q; and utilizing, by the processor, the determined Lagrange parameter to encode the source video.