Median Filter Aided CNN Based Image Denoising: An Ensemble Approach
Open accessThe 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
Cites
Statistics
Verification
Claims the paper makes
3 Valid · 0 Not valid · 0 Not checked · 14 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“which is a traditional non-linear filter that replaces the pixel centred in a window with the median of the window, and helps in the efficient removal of impulse noise as mentioned in [26] by Liang et al.”
What the sources say
- Supportedcites [26] Liang, L., Deng, S., Gueguen, L., Wei, M., Wu, X., and Qin, J. (2019). Convolutional Neural Network with Median Layers for Denoising Salt-and-Pepper Contaminations. arXiv. source
The Methodology section explicitly describes the median filter as especially efficient for removing impulse noise and says it replaces the center pixel of a window with that window’s median. The Conclusion also reports that incorporating median filtering into a neural network helps denoise salt-and-pepper noise. (LLM, conf=0.95)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For the Set12 dataset, the trend is similar, as shown in Table 2.”
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“in the case of the Set12 dataset, our model outperforms most of the existing denoising models.”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“| Ensemble-model | 31.73 | 29.20 | 26.20 |”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“When compared to non-deep learning based models, it completely outperforms them.”
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
3 more claims
Create a free account to see all 17 claims and every source verdict.
- Probably not a substantial claim. 6
- It may state two separate claims as one. 5
Citations
Cites
Very Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148 · cited 1×
“However, these drawbacks were overcome by AlexNet [19], and then other architectures, like VGG [20] and GoogleNet [21].”
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155
Gradient-based learning applied to document recognition
Lecun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998) · Proceedings of the IEEE · doi:10.1109/5.726791 · Cited by 58,805 · cited 1×
“Conventional CNNs, as well as the LeNet [18], have real-world application in handwritten digit recognition, but they have certain drawbacks.”
BackgroundGoing deeper with convolutions
Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Vanhoucke, V., & Rabinovich, A. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298594 · Cited by 48,562 · cited 1×
“However, these drawbacks were overcome by AlexNet [19], and then other architectures, like VGG [20] and GoogleNet [21].”
Image denoising by sparse 3-D transform-domain collaborative filtering
Dabov, K., Foi, A., Katkovnik, V., & Egiazarian, K. (2007) · IEEE Transactions on Image Processing · doi:10.1109/tip.2007.901238 · Cited by 9,283
Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising
Zhang, K., Zuo, W., Chen, Y., Meng, D., & Zhang, L. (2017) · IEEE Transactions on Image Processing · doi:10.1109/tip.2017.2662206 · Cited by 8,983
Ideal spatial adaptation by wavelet shrinkage
Donoho, D. L., & Johnstone, I. M. (1994) · Biometrika · doi:10.1093/biomet/81.3.425 · Cited by 8,760 · cited 2×
“For Wavelet BayesShrink [10] and Wavelet VisuShrink [9] models, we use soft denoising and estimate the standard deviation of the noise using the method in [9].”
MethodMartin, D., Fowlkes, C., Tal, D., & Malik, J. (2002) · Proceedings Eighth IEEE International Conference on Computer Vision. ICCV 2001 · doi:10.1109/iccv.2001.937655 · Cited by 8,216 · cited 1×
“in [29] and it can be downloaded from https://www2.”
Backgroundscikit-image: image processing in Python
van der Walt, S., Schönberger, J. L., Nunez-Iglesias, J., Boulogne, F., Warner, J. D., Yager, N., Gouillart, E., & Yu, T. (2014) · PeerJ · doi:10.7717/peerj.453 · Cited by 6,794
chambolle (2004) · doi:10.1023/b:jmiv.0000011321.19549.88 · Cited by 3,227
Adaptive wavelet thresholding for image denoising and compression
Chang, S. G., Yu, B., & Vetterli, M. (2000) · IEEE Transactions on Image Processing · doi:10.1109/83.862633 · Cited by 2,936 · cited 2×
“For Wavelet BayesShrink [10] and Wavelet VisuShrink [9] models, we use soft denoising and estimate the standard deviation of the noise using the method in [9].”
MethodFFDNet: Toward a Fast and Flexible Solution for CNN-Based Image Denoising
Zhang, K., Zuo, W., & Zhang, L. (2018) · IEEE Transactions on Image Processing · doi:10.1109/tip.2018.2839891 · Cited by 2,678
Learning Deep CNN Denoiser Prior for Image Restoration
Zhang, K., Zuo, W., Gu, S., & Zhang, L. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.300 · Cited by 2,116 · cited 8×
“For the DnCNN + median layer model, we take 611 th part of its output, for the IRCNN (dilation rate up to 6) + median layer model, we take 311 th part of its output and for the ADNet (dilation rate = 8) + median layer model, we take 211 th part of its output and then calculate the final output by adding all of them.”
MethodInfluentialDeep learning in environmental remote sensing: Achievements and challenges
Yuan, Q., Shen, H., Li, T., Li, Z., Li, S., Jiang, Y., Xu, H., Tan, W., Yang, Q., Wang, J., et al. (2020) · Remote Sensing of Environment · doi:10.1016/j.rse.2020.111716 · Cited by 1,697 · cited 1×
“The presence of noise in images can act as a deterrent in the course of various tasks in image processing, like facial recognition [1], form processing [2], environmental remote sensing [3], etc.”
BackgroundGoing Deeper with Convolutions
Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Vanhoucke, V., & Rabinovich, A. (2014) · arXiv · doi:10.48550/arxiv.1409.4842 · Cited by 1,461
Toward Convolutional Blind Denoising of Real Photographs
Guo, S., Yan, Z., Zhang, K., Zuo, W., & Zhang, L. (2019) · 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2019.00181 · Cited by 1,221 · cited 1×
“Another blind denoising model (CBDNet) [25] proposed by Guo et al.”
MethodFFDNet: Toward a fast and flexible solution for CNN-Based image denoising
Zhang, K., Zuo, W., & Zhang, L. (2018) · TUScholarShare (Temple University) · Cited by 916
Residual Dense Network for Image Restoration
Zhang, Y., Tian, Y., Kong, Y., Zhong, B., & Fu, Y. (2021) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2020.2968521 · Cited by 897 · cited 1×
“Hence, CNNs were introduced for image denoising [16,17].”
MethodAn Algorithm for Total Variation Minimization and Applications
Chambolle, A. (2004) · Journal of Mathematical Imaging and Vision · doi:10.1023/b:jmiv.0000011325.36760.1e · Cited by 832
Attention-guided CNN for image denoising
Tian, C., Xu, Y., Li, Z., Zuo, W., Fei, L., & Liu, H. (2020) · Neural Networks · doi:10.1016/j.neunet.2019.12.024 · Cited by 688 · cited 2×
“Median filter layers are added to ADNet [4] up to the Sparse Block or SB along with a dilation rate of 8.”
MethodFace Recognition Systems: A Survey
Kortli, Y., Jridi, M., Al Falou, A., & Atri, M. (2020) · Sensors · doi:10.3390/s20020342 · Cited by 623 · cited 1×
“The presence of noise in images can act as a deterrent in the course of various tasks in image processing, like facial recognition [1], form processing [2], environmental remote sensing [3], etc.”
BackgroundImage Blind Denoising with Generative Adversarial Network Based Noise Modeling
Chen, J., Chen, J., Chao, H., & Yang, M. (2018) · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2018.00333 · Cited by 599 · cited 1×
“[24] used a combination of Generative Adversarial Network (GAN) and CNN blind denoiser (GCBD) to first generate the noise samples and then use the noise patches to create a paired training”
MethodShrinkage Fields for Effective Image Restoration
Schmidt, U., & Roth, S. (2014) · 2014 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2014.349 · Cited by 543 · cited 1×
“Markov Random Field (MRF) based methods are used in [13], which provide competitive results when compared to the methods prior to it.”
Image restoration by sparse 3D transform-domain collaborative filtering
Dabov, K., Foi, A., Katkovnik, V., & Egiazarian, K. (2008) · SPIE Proceedings · doi:10.1117/12.766355 · Cited by 461
Image restoration using a neural network
Zhou, Y.-T., Chellappa, R., Vaid, A., & Jenkins, B. K. (1988) · IEEE Transactions on Acoustics, Speech, and Signal Processing · doi:10.1109/29.1641 · Cited by 412
Non-local Color Image Denoising with Convolutional Neural Networks
Lefkimmiatis, S. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.623 · Cited by 363 · cited 1×
“Lefkimmiatis proposed a Color Non-Local Network (CNLNet) [22], which uses the inherent non-local self-similarity on natural images to efficiently perform denoising.”
BackgroundImage Restoration Using Convolutional Auto-encoders with Symmetric Skip Connections
Mao, X.-J., Shen, C., & Yang, Y.-B. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1606.08921 · Cited by 323 · cited 1×
“Hence, CNNs were introduced for image denoising [16,17].”
MethodNonlinear mean filters in image processing
Pitas, I., & Venetsanopoulos, A. (1986) · IEEE Transactions on Acoustics, Speech, and Signal Processing · doi:10.1109/tassp.1986.1164857 · Cited by 194 · cited 1×
“These include non-linear [7] and non-adaptive filtering methods [8], which help to preserve edge information as well as signal to noise ratio information to estimate the noise.”
BackgroundLearning Deep CNN Denoiser Prior for Image RestorationPreprint
Zhang, K., Zuo, W., Gu, S., et al. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1704.03264 · Cited by 127 · cited 8×
“For the DnCNN + median layer model, we take 611 th part of its output, for the IRCNN (dilation rate up to 6) + median layer model, we take 311 th part of its output and for the ADNet (dilation rate = 8) + median layer model, we take 211 th part of its output and then calculate the final output by adding all of them.”
MethodInfluentialMixed Gaussian-impulse noise reduction from images using convolutional neural network
Islam, M. T., Rahman, S. M. M., Ahmad, M. O., & Swamy, M. N. S. (2018) · Signal Processing Image Communication · doi:10.1016/j.image.2018.06.016 · Cited by 78
Convolutional Neural Network with Median Layers for Denoising Salt-and-Pepper Contaminations
Liang, L., Deng, S., Gueguen, L., Wei, M., Wu, X., & Qin, J. (2019) · arXiv · doi:10.48550/arxiv.1908.06452 · Cited by 44 · cited 2×
“…filter layer, which is a traditional non-linear filter that replaces the pixel centred in a window with the median of the window, and helps in the efficient removal of impulse noise as mentioned in [26] by Liang et al. It is applied to each element of a feature channel in a moving window fashion.”
MethodConvolutional neural network with median layers for denoising salt-and-pepper contaminations
Liang, L., Deng, S., Gueguen, L., Wei, M., Wu, X., & Qin, J. (2021) · Neurocomputing · doi:10.1016/j.neucom.2021.02.010 · Cited by 44
Poisson noise reduction from X-ray images by region classification and response median filtering
Kirti, T., Jitendra, K., & Ashok, S. (2017) · Sadhana · doi:10.1007/s12046-017-0654-4 · Cited by 37 · cited 1×
“In the work [27], responsive median filters and the modified Harris corner point detector are used for reduction of Poisson noise in X-ray images.”
MethodMulti-frame image restoration using a neural network
Chiang, Y.-W., & Sullivan, B. J. (2003) · IEEE International Midwest Symposium on Circuits and Systems (MWSCAS) · doi:10.1109/mwscas.1989.101962 · Cited by 29 · cited 1×
“With the advent of deep learning, neural network based methods [14,15] were originally proposed.”
MethodAn edge-preserving subband coding model based on non-adaptive and adaptive regularization
Hong, S.-W., & Bao, P. G. (2000) · Image and Vision Computing · doi:10.1016/s0262-8856(99)00020-7 · Cited by 18 · cited 1×
“These include non-linear [7] and non-adaptive filtering methods [8], which help to preserve edge information as well as signal to noise ratio information to estimate the noise.”
BackgroundUnderstanding contents of filled-in Bangla form images
Bhattacharya, R., Malakar, S., Ghosh, S., Bhowmik, S., & Sarkar, R. (2021) · Multimedia Tools and Applications · doi:10.1007/s11042-020-09751-3 · Cited by 13 · cited 1×
“The presence of noise in images can act as a deterrent in the course of various tasks in image processing, like facial recognition [1], form processing [2], environmental remote sensing [3], etc.”
BackgroundSelf-citationImage denoising via lossy compression and wavelet thresholding
Chang, S. G., Yu, B., & Vetterli, M. (n.d.) · Proceedings of International Conference on Image Processing · doi:10.1109/icip.1997.647985
Cited by
Ensemble of Deep Learning Models for Sleep Apnea Detection: An Experimental Study
Mukherjee, D., Dhar, K., Schwenker, F., & Sarkar, R. (2021) · Sensors · doi:10.3390/s21165425 · Cited by 43
Self-citationEfficient image denoising with heterogeneous kernel-based CNN
Hu, Y., Tian, C., Zhang, J., & Zhang, S. (2024) · Neurocomputing · doi:10.1016/j.neucom.2024.127799 · Cited by 38
A parallel and serial denoising network
Zhang, Q., Xiao, J., Tian, C., Xu, J., Zhang, S., & Lin, C. (2023) · Expert Systems with Applications · doi:10.1016/j.eswa.2023.120628 · Cited by 30
Enhancing sonar image quality for underwater object recognition in marine environments: A review
Zhang, C., Chai, Y., Wang, X., & Chen, Y. (2025) · Ocean Engineering · doi:10.1016/j.oceaneng.2025.121862 · Cited by 26
Quantifying Salt and Pepper Noise Using Deep Convolutional Neural Network
Kumain, S. C., & Kumar, K. (2022) · Journal of The Institution of Engineers (India) Series B · doi:10.1007/s40031-022-00729-3 · Cited by 9
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
- gold
- doi.org
- Licence
- 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.