The Distortion-Perception Tradeoff in Finite Channels with Arbitrary Distortion Measures

Abstract

Whenever inspected by humans, reconstructed signals should not be distinguished from real ones. Typically, such a high perceptual quality comes at the price of high reconstruction error. We study this distortion-perception (DP) tradeoff over finite-alphabet channels, for the Wasserstein-$1$ distance as the perception index, and an arbitrary distortion matrix. We show that computing the DP function and the optimal reconstructions is equivalent to solving a set of linear programming problems. We prove that the DP curve is a piecewise linear function of the perception index, and derive a closed-form expression for the case of binary sources.

PDF NeurIPSW OpenReview Semantic Scholar

Cite

Text

Freirich et al. "The Distortion-Perception Tradeoff in Finite Channels with Arbitrary Distortion Measures." NeurIPS 2023 Workshops: InfoCog, 2023.

Markdown

[Freirich et al. "The Distortion-Perception Tradeoff in Finite Channels with Arbitrary Distortion Measures." NeurIPS 2023 Workshops: InfoCog, 2023.](https://mlanthology.org/neuripsw/2023/freirich2023neuripsw-distortionperception/)

BibTeX

@inproceedings{freirich2023neuripsw-distortionperception,
  title     = {{The Distortion-Perception Tradeoff in Finite Channels with Arbitrary Distortion Measures}},
  author    = {Freirich, Dror and Weinberger, Nir and Meir, Ron},
  booktitle = {NeurIPS 2023 Workshops: InfoCog},
  year      = {2023},
  url       = {https://mlanthology.org/neuripsw/2023/freirich2023neuripsw-distortionperception/}
}