CoChat

Comparative Analysis of Image Quality Assessment Metrics: MSE, PSNR, SSIM and FSIM

Open access
Al Najjar, Yusra · 2024 · International Journal of Science and Research (IJSR)

The DOI matches its Crossref record; 11 of the 11 references it cites were found in a registry or an index; 1 of the 4 sources its claims cite were checked for support.

Abstract

Evaluating the quality of an image proves to be a multifaceted and intricate endeavor, given the nuanced nature of human perception influenced by an array of physical and psychological factors. Despite numerous proposed techniques aimed at measuring image quality, none emerge as flawless or universally applicable. In the realm of image processing, where precision is paramount, extensive research has explored diverse methodologies including point difference analysis, image correlation, edge detection, neural networks (NN), region of interest (ROI) analysis, and consideration of the human visual system (HVS). The essence of an effective image quality measure lies in its ability to furnish accurate and consistent predictions. This study concentrates on the full -reference image quality, which involves utilizing a reference image for comparison purposes. Additionally, the study directs its focus toward contrasting a range of comparison tools, each leveraging different metrics. Among the comparison tools scrutinized are those reliant on point -based measurements such as the mean square error (MSE) and the peak signal -to -noise ratio (PSNR). Conversely, others focus on the compositional aspects of images, exemplified by metrics like the Feature Index Matrix (FSIM) and the Structured Similarity Index Matrix (SSIM).

Overview

CoChat Summary

TL;DR

This paper compares full-reference image quality metrics MSE, PSNR, SSIM, and FSIM on distorted images, reporting that FSIM values are closer to 1 than SSIM values. It concludes that subjective image quality assessment still outperforms objective metrics, and objective IQM remains an open area.

Key findings

  • In Table 1, FSIM is closer to 1 than SSIM across the listed images; for example, Boat has SSIM 0.90 vs. FSIM 0.95, Building has SSIM 0.77 vs. FSIM 0.95, and Lighthouse has SSIM 0.83 vs. FSIM 0.99.
  • MSE and PSNR vary by image and distortion; examples include Boat MSE 58.36 and PSNR 30.47, Building MSE 351.95 and PSNR 22.67, and Parrots MSE 37.02 and PSNR 32.45.
  • The paper states that different image quality metrics differ in value according to distortion type, and that it is hard to get the same quality value even if the same distortion is implemented on different images.
  • The conclusion states that subjective IQM, despite being time-consuming and expensive, still does better than objective IQM, and that objective IQM needs more work to cooperate with subjective IQM.

Methods

  • The study evaluates full-reference objective image quality metrics: MSE, PSNR, SSIM, and FSIM.
  • It uses thirteen original and distorted images listed as building, cycle, girl, door, hats, house, parrots, window, plane, boat, statue, lighthouse, and woman, obtained from the “Lossless True Color Image Suite” provided by the LIVE Image Quality Assessment Database; the results text also refers to “nine images.”
  • Images were converted to grayscale using PIL’s convert('L'), blurred with OpenCV GaussianBlur using kernel size ((5, 5), 0), and then assessed with the four metrics in Python; results are summarized in Table 1.

Limitations

  • The paper states that subjective IQM is time-consuming and expensive but still performs better than objective IQM, and that objective IQM is still open and needs work to cooperate with subjective IQM.
  • Evident from the text: the results discussion says nine images while the methodology and Table 1 list thirteen; the distortion is described as contrast enhancement, but the table column is labeled “Gaussian Variance.”
  • Evident from the text: formulas are not decoded in the provided version, and no statistical significance testing or confidence intervals are reported; the comparison is descriptive.

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

Integrity signals

Registry

Crossref · 10.21275/sr24302013533

Cites

11 of 11 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

1 Valid · 0 Not valid · 5 Not checked · 4 Own results and interpretations

  • Probably not a substantial claim. 2
  • It may state two separate claims as one. 4

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.

1 of 1

Citations

Cites

Cited by

See all 37

Metadata

Authors
Al Najjar, Y.
Venue
International Journal of Science and Research (IJSR)
Published
2024-03-05
Volume
13
Issue
3
Pages
110-114
ISSN
2319-7064 (ISSN-L)
OpenAlex
W4396863072
Cited by
37
References
10
Citations by year
2026: 21 · 2025: 15 · 2024: 1
FWCI
5.15
Type
journal article
Language
en
Open access
Licence
other-oa
Locations
International Journal of Science and Research (IJSR) (publishedVersion, other-oa)
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, mean squared error, PSNR, structural similarity index, human visual system, image quality prediction

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.

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).

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.