Tools
Tools
Karami, Afsaneh 
ORCID: https://orcid.org/0009-0003-4156-2827 and Finlayson, Graham D.
(2025)
A Comparative Study: Assessing Image Quality for the Jacobi and Gauss-Seidel Retinex Algorithms.
In: The 16th Congress of International Colour Association, 2025-10-19 - 2025-10-24.
Abstract
We present a comparative image quality experiment for the Jacobi and Gauss-Seidel Retinex, which are, from an algorithmic perspective, improved versions of the McCann99. In previous work, we reformulated the path-based McCann99 into an iterative convolutional center-surround framework, which offers advantages. First, it aligns with the human visual receptive field and unifies three common Retinex categories—path-based method, lightness processing, and center-surround mechanisms—into a coherent framework. Furthermore, from a computational standpoint, it demonstrates faster convergence rates and better rendition of images with fewer artefacts compared to McCann99. The Gauss-Seidel algorithm further reduces Retinex’s computational costs. While our earlier research focused on the mathematical formulation and convergence behaviour of the Jacobi and Gauss-Seidel Retinex methods, the current work emphasises image quality assessment and halo artefact evaluation. We compared our methods with McCann99 and TMO-Net, a deep learning tone mapper algorithm. We conducted our experiment on the LVZ-HDR dataset using objective quality metrics, including TMQI, FSITM, and LOE, as well as subjective preference experiments. Our objective results demonstrate that the Jacobi and Gauss-Seidel variants outperform the original McCann99 algorithm in terms of visual quality. The lower LOE scores of our methods compared to McCann99 indicate better preservation of lightness order and reduction of halo artefacts. The preference experiment shows that user preferences remain similar across all three Retinex methods. Notably, our methods also perform competitively against TMO-Net in both objective metrics (TMQI and FSITM) and subjective assessments, all while requiring no training on large datasets.
| Item Type: | Conference or Workshop Item (Paper) |
|---|---|
| Uncontrolled Keywords: | retinex, image quality, halo artefacts, hdr tone mapper |
| Faculty \ School: | Faculty of Science Faculty of Science > School of Computing Sciences |
| UEA Research Groups: | Faculty of Science > Research Groups > Colour and Imaging Lab |
| Related URLs: | |
| Depositing User: | LivePure Connector |
| Date Deposited: | 26 Jun 2026 10:57 |
| Last Modified: | 26 Jun 2026 10:57 |
| URI: | https://ueaeprints.uea.ac.uk/id/eprint/103505 |
| DOI: |
Actions (login required)
 |
View Item |