CoChat

Median Filter Aided CNN Based Image Denoising: An Ensemble Approach

Open access

The DOI matches its Crossref record; 36 of the 37 references it cites were found in a registry or an index; 3 of the 5 sources its claims cite were checked for support.

Abstract

Image denoising is a challenging research problem that aims to recover noise-free images from those that are contaminated with noise. In this paper, we focus on the denoising of images that are contaminated with additive white Gaussian noise. For this purpose, we propose an ensemble learning model that uses the output of three image denoising models, namely ADNet, IRCNN, and DnCNN, in the ratio of 2:3:6, respectively. The first model (ADNet) consists of Convolutional Neural Networks with attention along with median filter layers after every convolutional layer and a dilation rate of 8. In the case of the second model, it is a feed forward denoising CNN or DnCNN with median filter layers after half of the convolutional layers. For the third model, which is Deep CNN Denoiser Prior or IRCNN, the model contains dilated convolutional layers and median filter layers up to the dilated convolutional layers with a dilation rate of 6. By quantitative analysis, we note that our model performs significantly well when tested on the BSD500 and Set12 datasets.

Overview

CoChat Summary

## TL;DR

The paper proposes an ensemble denoising model for additive white Gaussian noise that combines three customized CNN denoisers: ADNet, IRCNN, and DnCNN, using a weighted average. On BSD500 and Set12, the ensemble slightly outperforms the individual CNN variants and clearly beats the non-deep learning baselines reported in the paper.

## Key findings

  • The proposed ensemble uses weights in the ratio 2:3:6 for ADNet : IRCNN : DnCNN.
  • On BSD500, the ensemble achieves average PSNRs of 31.73 dB, 29.20 dB, and 26.20 dB at noise levels 15, 25, and 50, respectively; these are the best reported values in Table 1.
  • On Set12, the ensemble also gives the strongest overall results, e.g. at noise level 15 the best non-deep-learning model reaches 23.54 dB, while the lowest CNN-based result is 32.36 dB, a difference of 8.82 dB.
  • The paper reports that each model takes about 2.5 hours to train, so the full ensemble takes about 7.5 hours.

## Methods

  • The study customizes three existing CNN denoisers: ADNet with median filter layers and dilation rate 8, DnCNN with median filter layers in the first half of layers, and IRCNN with median filter layers and dilation up to 6.
  • Training uses 200 grayscale BSD500 images, resized to 200 × 200, then split into 40 × 40 patches; Gaussian noise is added at standard deviations 15, 25, and 50.
  • At test time, the outputs of the three models are combined by a weighted average; performance is evaluated with PSNR against TV-Chambolle, Wavelet-VisuShrink, and Wavelet-BayesShrink baselines.

## Limitations

  • The paper states that future work should extend the approach to other noise types such as salt-and-pepper noise and Poisson noise.
  • The experiments are limited to AWGN on BSD500 and Set12; this is evident from the reported datasets and noise settings.
  • The paper does not report a full ablation on the effect of each added median-filter component beyond the tabled model comparisons; this is evident from the text.

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

Integrity signals

Registry

Crossref · 10.3390/a14040109

Cites

36 of 37 references verified0 citation errors · 1 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

3 Valid · 0 Not valid · 0 Not checked · 14 Own results and interpretations

3 more claims

Create a free account to see all 17 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 6
  • It may state two separate claims as one. 5

Create a free account to see these claims.

Citations

Cites

Show all 37 references

Cited by

Metadata

Authors
Dey, S., Bhattacharya, R., Schwenker, F., & Sarkar, R.
Affiliations
  • Jadavpur University (IN)
  • Universität Ulm (DE)
Corresponding author
Schwenker, Friedhelm
Venue
Algorithms
Publisher
MDPI AG
Published
2021-03-28
Volume
14
Issue
4
Pages
109
ISSN
1999-4893 (ISSN-L)
OpenAlex
W3140387395
Cited by
22
References
33
Citations by year
2026: 2 · 2025: 4 · 2024: 7 · 2023: 4 · 2022: 4 · 2021: 1
FWCI
1.44
Type
journal article
Language
en
Open access
Locations
  • Algorithms (publishedVersion, cc-by)
  • OPen Access Repositorium der Universität Ulm (OPARU) (Ulm University) (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Universität Ulm, Kommunikations- und Informationszentrum (kiz) (cc-by)
Funders
Jadavpur University
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Image and Signal Denoising Methods
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Image Enhancement Techniques
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Image Processing Techniques
Keywords
image denoising, median filter, ensemble learning, additive white Gaussian noise, convolutional neural networks, ADNet, DnCNN, attention mechanism, dilated convolution

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 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (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.