An FCM-Based Image De-Noising with Spatial Statistics Pilot Study
Round 1
Reviewer 1 Report
The article is devoted to an interesting scientific direction, but it is poorly written
1. The abstract contains abbreviations that have not been decoded anywhere before FCM, PSNR, SSIM ....
2. paragraph 2.3. should be significantly expanded.
3. Give a step-by-step algorithm for the proposed method.
4. The quality of the diagrams given in the article is disgusting.
Do you have an explanation why different methods give different results on different images? is it possible to choose the best method for the image in advance?
5. Give average PSNR, SSIM values for several images. It will be more objective.
Author Response
September 3, 2023
Jenny Chen
Assistant Editor
Applied Science
Dear Jenny Chen,
Thank you for your email of Aug. 24, 2023, in which you noticed that my manuscript entitled “An FCM-based image de-noising with spatial statistics pilot study (Manuscript ID: applsci-2568874)” required revision for publication in Applied Science.
The comments regarding my manuscript were extremely helpful in preparing a clearer version. I have rewritten many paragraphs according to the Referees’ recommendations.
Thank you very much for your advice. Enclosed is a copy of the revised version of the manuscript and a list of the revisions.
Your acknowledgement will be greatly appreciated. Thank you again.
Sincerely yours,
Tzong-Jer Chen, Ph.D.
Associate Professor
Department of Mathematics & Computer Science, Wuyi University, Wuyishan, Fujian, 354300 China.
Phone: +86-1809-4158256
E-mail: [email protected]
LIST OF REVISIONS
Author: Tzong-Jer Chen
Title: An FCM-based image de-noising with spatial statistics pilot study.
Manuscript ID: applsci-2568874.
The Reviewers comments regarding my manuscript were extremely helpful to me in preparing a clearer version. I have revised this manuscript according to the Reviewers’ suggestions. The revised parts are briefly described as follows:
Review Report (Reviewer 1)
The article is devoted to an interesting scientific direction, but it is
poorly written.
Response: This manuscript was reedited carefully for grammar and proof-
read by a native English speaker.
- The abstract contains abbreviations that have not been decoded anywhere before FCM, PSNR, SSIM ....
Response: Per your suggestion, all abbreviations were redefined when it
first appears in this article.
- paragraph 2.3. should be significantly expanded.
Response: As suggested, I expanded this section properly. A latest paper
about FCM was added in Reference section.
- Seyed EH, Fatemeh GJ, Mostafa HK. A fuzzy C-means algorithm for
optimizing data clustering. Expert Systems with Applications. Volume
227, 2023. 120377.
- Give a step-by-step algorithm for the proposed method.
Response: I apologize for the confusion caused by my poor expression. A
brief flow chart for proposed image denoise scheme showed in Figure 4.
- The quality of the diagrams given in the article is disgusting.
Response: The Figures are the same as a Reference paper (#32) of
Applied Science.
- Arnal J and Súcar L. Hybrid Filter Based on Fuzzy Techniques for Mixed Noise Reduction in Color Images Appl. Sci. 2020, 10, 243; doi:10.3390/app10010243
For clearness, I made some re-arrangement for Figures.
Do you have an explanation why different methods give different results on different images? is it possible to choose the best method for the image in advance?
- Why different methods give different results on different images?
Response: Image denoising is one of the main tasks in image procession.
This method can be divided into spatial domain filtering and transform
domain filtering, such as Average filter, Median filter for spatial domain
and Wavelet or Fourier Transformation filters for transform domain. In
addition, the de-noising filter can be divided as linear or non-linear filters.
Different denoising methods have different effects on images and will
produce different results it is also for different images.
Digital images are corrupted may by Impose noise (IN) or by Gaussian
white noise. Unlike Gaussian white noise, the characteristic of IN is that,
for images contaminated by IN, not all the pixel values are changed, but
only a portion of the pixels are contaminated by noise. Dealing with
different types of noise require suitable filtering schemes. This can have a
different result on the image.
- is it possible to choose the best method for the image in advance?
Response: Yes, it is possible. Actually, people cannot read noise fully
because the noise is a random signal. So, the judgement of filtering
process is good or not which is based on some metrics such as PSNR or
SSIM. If we have a best image quality metrics then we can have a best
judgement scheme for image noise filtering.
- Give average PSNR, SSIM values for several images. It will be more objective.
Response: As suggested, a section entitled “4.3 Comparison of schemes
by metrics” was added and four Tables were added in this section. These
Tables are;
Table I. PSNR and SSIM for Barbara using various denoising schemes. The random padded Gaussian white noise μ=100, σ= 15, on 20% of the entire image. Ave3=3×3 Ave5=5×5 average filter, G05= Gaussian filter (σ=0.5), Bi=Bilateral filter, NLM=Non-local means.
Table II. PSNR and SSIM for Lena using various denoising schemes. The random padded Gaussian white noise μ=0, σ=20, on 30% of the entire image. Ave3=3×3 Ave5=5×5 average filter, G05= Gaussian filter (σ=0.5), Bi=Bilateral filter, NLM=Non-local means.
Table III. Average PSNR and SSIM for Barbara using various denoising schemes. The random padded Gaussian white noise: μ=0, 0, 0, 30, 100; σ= 10, 15, 20, 30, 50, 60; on 20%, 50%, 75% of the entire image. There are 20 images in total. Ave3=3×3 Ave5=5×5 average filter, G05= Gaussian filter (σ=0.5), Bi=Bilateral filter, NLM=Non-local means.
Table IV. Average PSNR and SSIM for Lena by various denoise schemes. The random padded Gaussian white noise: μ=0, 0, 0, 30, 100; σ= 10, 15, 20, 30, 50, 60; on 20%, 50%, 75% of the entire image. There are 20 images in total. Ave3=3×3 Ave5=5×5 average filter, G05= Gaussian filter (σ=0.5), Bi=Bilateral filter, NLM=Non-local means.
Author Response File: Author Response.docx
Reviewer 2 Report
Dear Authors,
The article, in general, is related to the problem of image processing. In particular, the authors propose a scheme for de-noising images. The general layout of the article is correct; however, in my opinion, many elements need to be improved or extended:
1. List of references. I have two suggestions:
1.1. you provided an extensive list of 36 references. However, there is no publication from Applied Sciences. You could better justify your subject as suitable for the journal.
1.2 there is a reference to a web page (34.) with no explanation and no access date. The required information should be added here.
2. You use the acronym "FCM" multiple times (title, keywords, abstract, etc.) However, the full name (fuzzy c-means) doesn't appear until line 101 of the text. In my opinion, you should give the full name, at least in keywords, instead of the acronym.
3. The plan of experiments is outlined cursorily and not very clearly; it should be rewritten.
4. Comparative schemas
4.1. What were the criteria for selecting comparative schemes (SM3, SM5, Gauss, BL, NLM)?
4.2. How were benchmarking schemes implemented? Were ready-made matching Matlab libraries used, or did the authors implement them themselves?
5. Figures and their legends are mixed. There are two legends next to each other (Fig.4 and Fig.5). It seems that i.e. legends of Figs. 5 - 10 are above the figures.
6. The results presented in the figures are briefly commented on in the text. In my opinion, this should be described in more detail.
7. Discussion: you should more clearly present the advantages and maybe weaknesses of your scheme compared to others.
8. I suggest that a native speaker proofread the manuscript; however, I'm not qualified enough in English to make it obligatory.
Regards,
Author Response
September 3, 2023
Jenny Chen
Assistant Editor
Applied Science
Dear Jenny Chen,
Thank you for your email of Aug. 24, 2023, in which you noticed that my manuscript entitled “An FCM-based image de-noising with spatial statistics pilot study (Manuscript ID: applsci-2568874)” required revision for publication in Applied Science.
The comments regarding my manuscript were extremely helpful in preparing a clearer version. I have rewritten many paragraphs according to the Referees’ recommendations.
Thank you very much for your advice. Enclosed is a copy of the revised version of the manuscript and a list of the revisions.
Your acknowledgement will be greatly appreciated. Thank you again.
Sincerely yours,
Tzong-Jer Chen, Ph.D.
Associate Professor
Department of Mathematics & Computer Science, Wuyi University, Wuyishan, Fujian, 354300 China.
Phone: +86-1809-4158256
E-mail: [email protected]
LIST OF REVISIONS
Author: Tzong-Jer Chen
Title: An FCM-based image de-noising with spatial statistics pilot study.
Manuscript ID: applsci-2568874.
The Reviewers comments regarding my manuscript were extremely helpful to me in preparing a clearer version. I have revised this manuscript according to the Reviewers’ suggestions. The revised parts are briefly described as follows:
Review Report (Reviewer 2)
- List of references. I have two suggestions:
1.1. you provided an extensive list of 36 references. However, there is no publication from Applied Sciences. You could better justify your subject as suitable for the journal.
Response: I'm sorry for my carelessness. I searched on the Web and found
a very valuable paper from Applied Science. The paper used FCM to
decrease contamination from a mixed combination of impulse and
Gaussian noise on digital images. This paper was added in References
section as No.32.
- Arnal J and Súcar L. Hybrid Filter Based on Fuzzy Techniques for
Mixed Noise Reduction in Color Images Appl. Sci. 2020, 10, 243;
doi:10.3390/app10010243
1.2 there is a reference to a web page (34.) with no explanation and no access date. The required information should be added here.
Response: Per your suggestion, The References No. 36. was amended as
below (The Reference # 34 was change to #36.).
- The MATLAB implementation code of Bilateral Filter, downloaded
from http://people.csail.mit.edu/jiawen/, July, 2021.
- You use the acronym "FCM" multiple times (title, keywords, abstract, etc.) However, the full name (fuzzy c-means) doesn't appear until line 101 of the text. In my opinion, you should give the full name, at least in keywords, instead of the acronym.
Response: Per your suggestion, all abbreviations were redefined when it
first appears in this article.
- The plan of experiments is outlined cursorily and not very clearly; it should be rewritten.
Response: I apologize for the confusion caused by my poor expression. A
brief flow chart for proposed image denoise scheme showed in Figure 4.
- Comparative schemas
4.1. What were the criteria for selecting comparative schemes (SM3, SM5, Gauss, BL, NLM)?
Response: There are many image filters used in papers. These filters be
applied in spatial domain filtering or transform domain filtering. In
addition, the de-noising filter can be divided as linear or non-linear filters.
Three clusters pre-assumed for FCM in this work: they are heavy noisy, medium noisy and less noisy areas. Each pixel in image is partially belong to clusters and the rates were determined by a FCM membership function. Based on above, the pilot study needs three filters they are actually different degree in denoising. The sequence of denoising effects is SM5 > SM3 and > Gauss 05.
There are many papers mentioned that the BL and NLM they are state-of-the-art filters.
4.2. How were benchmarking schemes implemented? Were ready-made matching Matlab libraries used, or did the authors implement them themselves?
Response: The calculations of Moran statistics, Ave, Gaussian filters,
proposed denoising, PSNR, SSIM etc. implemented by myself. The FCM
used Matlab libraries and Bilateral NLM filter were download from web
at the courtesy of the owner.
- Figures and their legends are mixed. There are two legends next to each other (Fig.4 and Fig.5). It seems that i.e. legends of Figs. 5 - 10 are above the figures.
Response: As suggested, I made some re-arrangement for legends of
Figures.
- The results presented in the figures are briefly commented on in the text. In my opinion, this should be described in more detail.
Response: Response: As suggested, a section entitled “4.3 Comparison of
schemes by metrics” was added and four Tables were added in this
section. These Tables are;
Table I. PSNR and SSIM for Barbara using various denoising schemes. The random padded Gaussian white noise μ=100, σ= 15, on 20% of the entire image. Ave3=3×3 Ave5=5×5 average filter, G05= Gaussian filter (σ=0.5), Bi=Bilateral filter, NLM=Non-local means.
Table II. PSNR and SSIM for Lena using various denoising schemes. The random padded Gaussian white noise μ=0, σ=20, on 30% of the entire image. Ave3=3×3 Ave5=5×5 average filter, G05= Gaussian filter (σ=0.5), Bi=Bilateral filter, NLM=Non-local means.
Table III. Average PSNR and SSIM for Barbara using various denoising schemes. The random padded Gaussian white noise: μ=0, 0, 0, 30, 100; σ= 10, 15, 20, 30, 50, 60; on 20%, 50%, 75% of the entire image. There are 20 images in total. Ave3=3×3 Ave5=5×5 average filter, G05= Gaussian filter (σ=0.5), Bi=Bilateral filter, NLM=Non-local means.
Table IV. Average PSNR and SSIM for Lena by various denoise schemes. The random padded Gaussian white noise: μ=0, 0, 0, 30, 100; σ= 10, 15, 20, 30, 50, 60; on 20%, 50%, 75% of the entire image. There are 20 images in total. Ave3=3×3 Ave5=5×5 average filter, G05= Gaussian filter (σ=0.5), Bi=Bilateral filter, NLM=Non-local means.
- Discussion: you should more clearly present the advantages and maybe weaknesses of your scheme compared to others.
Response:
Advantages: The first paragraph in Conclusion section. There is “This work proposed estimation the variation of spatial information in RE images by Moran statistics. The Moran’s Z measurements were used as feature data in FCM. Three clusters were pre-assumed for FCM in this work: they are heavy, medium and less noisy or edge areas. The rates of each position partially belong to clusters were determined by a FCM membership function. The membership function uses as a weight to calculate the weighted sum of each position. “. The denoising process was transformed as a linear combination of product of three SMs with membership function in the same position.
Weaknesses: Based on above, the pilot study needs three filters they are actually different degree in denoising. Further researches on the optimized number of clusters and better smoothed or structure versions using in linear combination are needed to optimize this scheme, in future.
All above had suggested in Conclusion section.
- I suggest that a native speaker proofread the manuscript; however, I'm not qualified enough in English to make it obligatory.
Response: This manuscript was reedited carefully for grammar and proof-
read by a native English speaker.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The authors have corrected the comments. I recommend accepting the article.
Reviewer 2 Report
Dear Author,
In my opinion, the manuscript has been sufficiently improved to be published in Applied Sciences.
Best Regards,