Evaluating quality metrics through the lenses of psychophysical measurements of low-level vision

Dounia Hammou 1 , Yancheng Cai 1 , Pavan Madhusudanarao 2 , Christos G. Bampis 2 , Zhi Li 2 , Rafał K. Mantiuk 1 · 2025

Read Evaluating quality metrics through the lenses of psychophysical measurements of low-level vision by Dounia Hammou 1 , Yancheng Cai 1 , Pavan Madhusudanarao 2 , Christos G. Bampis 2 , Zhi Li 2 , Rafał K. Mantiuk 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-16264.txt.

Description

Scores that grade image and video quality are trusted to see as humans see. Yet few are built from real models of vision. Dounia Hammou and co-authors put thirty-four such metrics to a different test in this 2025 paper. They used classic experiments on low-level vision: how faint a contrast we can spot, how a strong pattern masks a weak one, and how contrast matches across strengths.

Two learned metrics predicted masking well, and a multiscale structural metric fixed its parent metric's overweighting of fine detail. Almost no metric handled the constancy of contrast above threshold. The authors offer their test set as an extra lens for judging quality metrics, beyond the usual benchmark scores.

Catalog record

CREATORDounia Hammou 1 , Yancheng Cai 1 , Pavan Madhusudanarao 2 , Christos G. Bampis 2 , Zhi Li 2 , Rafał K. Mantiuk 1
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE32 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-16264

Browse the full library · About the Tower of Babel