titleRate-distortion optimization for full-reference image quality metrics via stochastic Hessian estimates
authorsSamuel Fernández-Menduiña, Eduardo Pavez, Antonio Ortega
date2026-09-24
venuearXiv (preprint, not peer reviewed)
idarXiv:2609.30077
affiliationUniversity of Southern California
upvotes5 (Hugging Face Daily Papers, 2026-11-01)
linksabs pdf hf code
abstractBlock-based video codecs select coding parameters based on the input by optimizing a rate-distortion trade-off. The conventional distortion choice, the sum of squared errors (SSE), simplifies parameter selection: the SSE is the sum of block-wise SSEs, so rate-distortion optimization (RDO) can treat blocks independently. Alternatively, full-reference image quality assessment (FR-IQA) metrics such as MS-SSIM or LPIPS often align better with the human visual system than SSE, but they cannot be used in-loop: they do not decompose block-wise and typically require the fully decoded image as input. Building on existing results in metric quadratization, we approximate a broad class of FR-IQA metrics by an input-dependent quadratic distortion (IDQD), whose quadratic form matrix is derived from the Hessian of the metric evaluated at the source video. To make the distortion computable block-wise, we propose two approximations of the Hessian matrix: 1) keeping the block-diagonal, and 2) keeping only its diagonal. We propose estimators for both that require only matrix-vector products with the Hessian obtained by automatic differentiation. Across five metrics for Kodak and CLIC in VVC, IDQD-RDO achieves 14.2-36.7 % BD-rate savings under the target metric with no decoder changes and incurs 10-30 % encoding complexity overhead.

cite

@misc{fernndezmenduia2026rate,
  title={Rate-distortion optimization for full-reference image quality metrics via stochastic Hessian estimates},
  author={Samuel Fernández-Menduiña and Eduardo Pavez and Antonio Ortega},
  year={2026},
  eprint={2609.30077},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/2609.30077}
}