CoChat

A Survey and Tutorial on Image Quality Assessment with a Contrast-Weighted Structural Similarity Framework

Open access
Agaian, Sos S., Grigoryan, Artyom M., Ayunts, Hrach · 2026 · Information

The DOI matches its Crossref record; 65 of the 73 references it cites were found in a registry or an index, and 1 are not checked yet; 25 of the 77 sources its claims cite were checked for support.

Abstract

Objective IQA has advanced significantly through full-reference metrics designed to approximate human judgment. Standard measures such as PSNR, SSIM, and RMSE provide established benchmarks.

Verified 2026-11-19 · 1 reference not checked yet

Overview

CoChat Summary

TL;DR

This survey/tutorial reviews full-reference and no-reference image quality assessment and introduces ACSSIM, a lightweight contrast-weighted extension of classical FR-IQA using local Weber-law contrast and entropy weighting. On TID2013 and KADID-10k, ACSSIM improves correlation with MOS over SSIM but remains below stronger learned and multi-scale FR-IQA methods.

Key findings

  • ACSSIM improves over SSIM on TID2013 (PLCC 0.6995 vs 0.6558; SRCC 0.6709 vs 0.6269) and KADID-10k (PLCC 0.5984 vs 0.5755; SRCC 0.6778 vs 0.6188).
  • Distortion-wise SRCC on TID2013 shows strongest gains for contrast change (+0.1044 SRCC over SSIM) and quantization noise (+0.0850 SRCC), with positive gains for sparse sampling/reconstruction, JPEG/JPEG2000-related distortions, and multiplicative Gaussian noise; decreases occur for local block-wise distortions, impulse noise, additive noise in color components, and mean shift.
  • In a controlled cat-image experiment with approximately equal MSE distortions, MSE and PSNR produced nearly identical values across distortions, SSIM’s lowest score was for the noisy image, while ACSSIM gave a more consistent ordering; ACSSIM distortion scores increased from 0.0281 to 0.4312 across the five distorted versions.
  • Preliminary contrast-measure examples report, for thermal panel “dji-0769.jpg,” EME values of 2.98 (original), 9.00 (histogram equalization), 11.48 (0.80-rooting), and 11.66 (0.75-rooting), with the 0.75-rooting image highest across all eight listed measurements.

Methods

  • Design: a survey and tutorial providing self-contained mathematical derivations of classical error-sensitivity measures (MSE, RMSE, MAE, Euclidean distance, RMSLE, Huber loss) and ANN approaches, plus a taxonomy of pixel-based, histogram-based, feature-based, and deep-learning similarity methods.
  • ACSSIM: decomposes reference and distorted images into non-overlapping blocks of L1×L2 pixels (L1, L2 = 3, 5, 8, or 16); computes local contrast terms using AC = AC5 or AMEE; computes local similarity via SSIM or MSE; and averages local similarity responses with normalized contrast-adaptive weights.
  • Evaluation: TID2013 (25 reference images, 3000 distorted images, 24 distortion types) and KADID-10k (81 reference images, 10,125 distorted images, 25 distortion types, five degradation levels), using PLCC and SRCC via the pyiqa toolbox; a controlled experiment used synthetically generated distortions on a single reference image with approximately equal MSE.

Limitations

  • Stated in paper: parameter sensitivity (weighting constants for contrast and entropy; sub-optimal tuning may cause over-masking in high-texture regions); block-size dependency (too small makes entropy statistically insignificant, too large loses local HVS masking); and sensitivity to uniform global luminance shifts.
  • Stated in paper: ACSSIM does not outperform stronger FR-IQA methods such as MS-SSIM, FSIM, GMSD, PieAPP, and TOPIQ-FR; its value is simplicity and interpretability. Distortion-wise improvement is not uniform, with decreases for local block-wise distortions, impulse noise, additive noise in color components, and mean shift; the current grayscale formulation is less effective for highly localized, color-component-dependent, or globally uniform distortions.
  • Evident from text: the survey notes that entropy ignores spatial relationships between pixels, and that EMEE focuses only on local features and may miss broader image characteristics, though ACSSIM uses local entropy/contrast weighting.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/info17070632

Cites

65 of 73 references verified0 citation errors · 7 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

10 Valid · 9 Not valid · 13 Not checked · 100 Own results and interpretations

52 more claims

Create a free account to see all 132 claims and every source verdict.

See all claims — free account
  • It may state two separate claims as one. 36
  • Probably not a substantial claim. 54
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

5 of 6 · the rest aren’t loaded

1 more finding

Create a free account to see every finding of this check.

See all findings — free account

Citations

Cites

Metadata

Authors
Agaian, S. S., Grigoryan, A. M., & Ayunts, H.
Affiliations
  • Department of Computer Science, School of Engineering, City University of New York (CUNY), New York, NY 10031, USA
  • The University of Texas at San Antonio (US)
  • Yerevan State University (AM)
Corresponding author
Grigoryan, Artyom M.
Venue
Information
Publisher
MDPI AG
Published
2026-06-27
Volume
17
Issue
7
Pages
632
ISSN
2078-2489 (ISSN-L)
OpenAlex
W7166583081
Cited by
0
References
72
FWCI
0.00
Type
journal article
Language
en
Open access
Locations
  • Information (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Image and Video Quality Assessment
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Image Retrieval and Classification Techniques
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Digital Media Forensic Detection
Keywords
image quality assessment, full-reference quality metrics, structural similarity index, human visual system, Weber's law, entropy weight method, perceptual image quality, TID2013, PSNR, LPIPS

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-06, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-06 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.