Infrared Stripe Correction Algorithm Based on Wavelet Analysis and Gradient Equalization
Round 1
Reviewer 1 Report
In this manuscript, the authors propose a stripe correction algorithm based on wavelet analysis and gradient equalization. Please address the following comments:
1- Why did the authors use the image roughness ρ and the AVGE metrics with the suitcase, leaves and people datasets only? Why they did not apply them to the other datasets (cameraman, and vases)?
2- After Eq. 18, the authors should introduce the values of k1 and k2 used in the calculation of the SSIM metric.
3- Why did the authors add a yellow arrow on the image of the second column of Fig. 5? Why did the authors add it to this subfigure only?
4- Why did the authors add red ellipse on Figs. 12-14? Why did they add it on subfigure (b) only?
5- Line 94, please replace ''a deep CNN model'' by ''a deep convolutional neural network (CNN) model''.
6- The authors should change the title of subsection 5.4 (More instructions). I suggest the following title ''Limitations of the proposed method''.
7- The test image of Fig. 8 is known as ''cameraman image'' in the literature--not ''photographer''. Please mention the cameraman image in line 252. Please replace ''photographer' by ''cameraman image'' in the whole manuscript.
8- In line 111, please replace ''denoising algorithms, The wavelet'' by ''denoising algorithms, the wavelet'', line 114 ''this paper chooses'' by ''we choose'', line 251 ''We take a lot of experiments'' by ''We conducted several experiments'', line 425 ''to talk about the'' by ''to discuss the'', and line 411 ''the results obtained by CNN are not ideal.'' by ''CNN gives bad results.''
9- Remove Ref. [27] from the title of subsection 3.3 because you already added it in the text.
10- Please add highlighted results (best SSIM, best PSNR) to the abstract and conclusion sections.
11- Please add some lines of future works to the conclusion section
12- The quality of Fig. 15 should be improved.
Author Response
Thank you for your valuable comments which helped us to improve our manuscript as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in red in the manuscript.
The point-by-point response to the comments is given below.
Point 1: Why did the authors use the image roughness ρ and the AVGE metrics with the suitcase, leaves and people datasets only? Why they did not apply them to the other datasets (cameraman, and vases)?
Response: The experiments and indicators in this paper are mainly focused on the original infrared image(suitcase, leaves and people), and the original infrared image is used to carry out various algorithm experiments and comparisons. In the simulation image experiment(cameraman, and vases), we compare the visual effects and use the PSNR and SSIM indicators to evaluate the algorithm, which can explain the superior performance of the proposed algorithm. Therefore, the ρ and AVGE indicators are not used in the simulated image.
Point 2: After Eq. 18, the authors should introduce the values of k1 and k2 used in the calculation of the SSIM metric.
Response: k1=(M1×L)2 k2=(M2×L)2 The general M1=0.01、M2=0.03、L=255. So we set k1 and k2 as 6.502500 and 58.522500 in this paper to avoid system instability. We have added relevant content in the draft. The most original reference of SSIM index is：
Wang, Z., Bovik, A. C., Sheikh, H. R., & Simoncelli, E. P. (2004). Image Quality Assessment: From Error Visibility to Structural Similarity. IEEE Transactions on Image Processing, 13(4), 600–612. doi:10.1109/tip.2003.819861
Point 3: Why did the authors add a yellow arrow on the image of the second column of Fig. 5? Why did the authors add it to this subfigure only?
Response: In order to keep the image simple and clear , we only marked the second image. It can be clearly seen from the figure that the entire second image does not appear to be blurred, and the portion marked by the yellow arrow does not appear to lose the vertical edge detail. The first image shows residual stripe noise, and the third and fourth images appear blurry as a whole.
Point 4: Why did the authors add red ellipse on Figs. 12-14? Why did they add it on subfigure (b) only?
Response: Figure 12(b)-14(b) are all correction results of the TV algorithm, and a large amount of stripe noise remains in the correction results. In this paper, the red ellipse is uniformly used to mark the residual noise of the TV algorithm to correct the image. Related brief instructions have been highlighted in red in the article.
Point 5: Line 94, please replace ''a deep CNN model'' by ''a deep convolutional neural network (CNN) model''.
Response: we are very sorry for our incorrection writing. According to your opinion, we have made modifications in the article.
Point 6: The authors should change the title of subsection 5.4 (More instructions). I suggest the following title ''Limitations of the proposed method''.
Response: Thank you very much for your valuable advice. According to your opinion, we have made modifications in the article.
Point 7: The test image of Fig. 8 is known as ''cameraman image'' in the literature--not ''photographer''. Please mention the cameraman image in line 252. Please replace ''photographer' by ''cameraman image'' in the whole manuscript.
Response: we are very sorry for our incorrection writing. We have made correction according to your comments.
Point 8: In line 111, please replace ''denoising algorithms, The wavelet'' by ''denoising algorithms, the wavelet'', line 114 ''this paper chooses'' by ''we choose'', line 251 ''We take a lot of experiments'' by ''We conducted several experiments'', line 425 ''to talk about the'' by ''to discuss the'', and line 411 ''the results obtained by CNN are not ideal.'' by ''CNN gives bad results.''
Response: We have made correction according to your comments.
Point 9: Remove Ref. [27] from the title of subsection 3.3 because you already added it in the text.
Response: We have revised it in the article.
Point 10: Please add highlighted results (best SSIM, best PSNR) to the abstract and conclusion sections.
Response: We have added relevant content in the abstract and conclusion sections.
Point 11: Please add some lines of future works to the conclusion section.
Response: We have added relevant content in the conclusion part.
Point 12: The quality of Fig. 15 should be improved.
Response: Fig.15. is the column mean curve of the original image calculated and the corrected image by matlab. According to your opinion, we have improved the image in the new draft.
Special thanks to you for your good comments.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This is an interesting research paper. There are some suggestions for revision.
1. Please highlight your contributions in introduction.
2. Please discuss the current issues of infrared image mentioned in the following paper.
Z. Zhu, G. Qi, Y. Chai, H. Yin, and J. Sun, A novel visible-infrared image fusion framework for smart city. IJSPM 13(2): 144-155 (2018)
3. Please discuss the following image decomposition and denoising solution.
Z. Zhu, H. Yin, Y. Chai, Y. Li, and G. Qi, A novel multi-modality image fusion method based on image decomposition and sparse representation. Inf. Sci. 432: 516-529 (2018)
4. In section 5, please list where ceramic photographer image comes from, and where Tendero’s dataset can be downloaded.
5. For one-dimensional random Gaussian process, is there any guidance to select a suitable standard deviation η?
6. Except AVGE, is there any other objective evaluation metric?
Author Response
Thank you for your valuable comments which helped us to improve our manuscript as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in red in the manuscript.
The point-by-point response to the comments is given below.
Point 1: Please highlight your contributions in introduction.
Response: We have made correction according to your comments. Relevant content has been added in the introduction.
Point 2: Please discuss the current issues of infrared image mentioned in the following paper.
Z. Zhu, G. Qi, Y. Chai, H. Yin, and J. Sun, A novel visible-infrared image fusion framework for smart city. IJSPM 13(2): 144-155 (2018)
Response: The infrared image is imaged based on the temperature difference between the target and the background. The infrared image has a strong penetrating ability, and the infrared image can afford good radiometric resolution information, but the infrared image also has many limitations. Infrared images are difficult to reflect the details of the target. For instance, in the field of fire warning processing, infrared image can detect a fire immediately, but may not provide an accurate location.
Point 3: Please discuss the following image decomposition and denoising solution.
Z. Zhu, H. Yin, Y. Chai, Y. Li, and G. Qi, A novel multi-modality image fusion method based on image decomposition and sparse representation. Inf. Sci. 432: 516-529 (2018)
Response: A cartoon-texture decomposition and sparse representation algorithm is proposed in the literature to decompose the image into the cartoon part and the texture part. For the texture part of the image, the sparse representation method is adopted to obtain more accurate texture part through the dictionary acquired by learning and the algorithm of max-l1.However, the stripe noise of infrared image exists in the vertical part of the image. Therefore, we use wavelet decomposition to get the vertical part of the image. (A short explanation has been added in section 3.1 of the draft).
Point 4: In section 5, please list where ceramic photographer image comes from, and where Tendero’s dataset can be downloaded.
Response: The ceramic and cameraman image is available under the CC-BY license [27]
[27] He, K.; Sun, J.; Tang , X.; Guided image filtering. IEEE Trans. Pattern Anal. 2013,35,1397–1409.
Tendero’s dataset can be downloaded from the following link:
http://demo.ipol.im/demo/glmt_mire/ or http://www.ipol.im/pub/art/2012/glmt-mire/
(Relevant instructions have been supplemented in this paper)
Point 5: For one-dimensional random Gaussian process, is there any guidance to select a suitable standard deviation η?
Response: Owing the intensity of each column of stripe noise is roughly the same, and the columns are obviously different from each other, this paper first generates a one-dimensional random gaussian vector with an average value of 0 and a standard deviation of η, then expands the gaussian vector to the whole image and adds it to the original image, which is our simulated process of adding noise.As we increase the value of η, we add more and more striated noise in the image. But the value of η should be between 0 and 1.
Point 6: Except AVGE, is there any other objective evaluation metric?
Response: Some people define energy based on horizontal gradients as another measure of performance(The reference papers are at the end). the principle is that stripe noise only affects horizontal gradient energy, but has little effect on vertical gradient. The formula is defined as:
where (m,n) denote the size of testing image, f(i,j) denote image f at the coordinate (i,j).
There are similarities with the AGVE metrics used in this article.
Cao, Y.; He,Z.; Yang,J.; Ye,X.; Cao,Y. A multi-scale non-uniformity correction method based on wavelet decomposition and guided filtering for uncooled long wave infrared camera. Signal Processing. Image Communication. 2018 ,60, 13-21.
Special thanks to you for your good comments.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors have shown a lot of efforts to improve the manuscript and this should be well appreciated. They have addressed most of my comments carefully and in detail by adding more materials in the text. As a result, I now recommend the current manuscript can be accepted for publication after addressing the following minor amendments.
a) English usage in this manuscript must be substantially improved. There are many grammatical errors, vague descriptions, and typos (e.g., lines 35 and 40).
b) I believe that the image roughness ρ and the AVGE metrics could be used to measure the performance of the proposed method with cameraman, and vases datasets. For each row in Table 2 or Table 3, the authors can calculate ρ and the AVGE metrics from each corrected image that had a certain level of noise (here, the authors have 10 levels).
c) The authors should add a list of abbreviations and symbols ate the end of the manuscript.
Author Response
Response to Reviewer 1 Comments
I should like to express my appreciation to you for suggesting how to improve our paper .We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in red in the manuscript.
The point-by-point response to the comments is given below.
a) English usage in this manuscript must be substantially improved. There are many grammatical errors, vague descriptions, and typos (e.g., lines 35 and 40).
Response: we are very sorry for our written mistakes, We tried our best to improve the manuscript and made some changes in the manuscript. These changes will not influence the content and framework of the paper. And here we did not list the changes
but marked in red in revised paper.
b) I believe that the image roughness ρ and the AVGE metrics could be used to measure the performance of the proposed method with cameraman, and vases datasets. For each row in Table 2 or Table 3, the authors can calculate ρ and the AVGE metrics from each corrected image that had a certain level of noise (here, the authors have 10 levels).
Response: Thank you very much for your valuable comments. In order to maintain the integrity of the article, we add a description of the ρ and AVGE parameters in section 5.3.2 of the article.
c) The authors should add a list of abbreviations and symbols ate the end of the manuscript.
Response: Thank you very much for your valuable advice. According to your opinion, we have add a list of abbreviations at the end of the article
Special thanks to you for your good comments.
Author Response File: Author Response.pdf
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
In this manuscript, the authors propose an infrared stripe correction algorithm based on wavelet decomposition and gradient equalization. I have the following comments:
1- The authors did not present and discuss the related studies comprehensively. They mentioned 6 studies only and ignored very interesting studies, such as:
https://www.mdpi.com/1424-8220/18/12/4299
https://www.sciencedirect.com/science/article/pii/S1350449515300293
https://www.mdpi.com/2072-4292/9/6/559/htm
The authors should add a separate section for related work, including recently published papers, and discuss the advantages and disadvantages of each method.
2-The authors did not explain why they used the wavelet decomposition method, specifically. Why they did not use DCT or FFT, for example? What is their motivation?
3-Deep learning has been exploited in several fields, of which stripe noise removal. For instance, the authors of the following article have used deep convolutional networks to remove stripe noise:
Removing Stripe Noise From Infrared Cloud Images via Deep Convolutional Networks https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=8419235
The authors should discuss such methods and compare them with their method.
4- The PSNR and SSIM metrics have been used in most of the related work. However, the authors did not use them in this manuscript. It is difficult to see the advantage of the proposed method compared to the related methods.
5- In table 1, the authors reported the MSE values of their method and three other methods. These values are incorrect. How does the proposed method have the highest MSE value and the authors mention that their method is the best? The best method should have the lowest MSE value. For instance, in the following paper, Shengwei Zhang et al. reported the RMSE of their method and it was lower than the other methods.
Stripe Noise Removal for Infrared Images Using Guided Filter https://www.spiedigitallibrary.org/conference-proceedings-of-spie/10157/101572R/Stripe-noise-removal-for-infrared-images-using-guided-filter/10.1117/12.2247044.short
6-Wavelet decomposition has been employed in several stripe noise removal methods, however, the authors did not mention or discuss them. For example:
https://ieeexplore-ieee-org.sabidi.urv.cat/document/6805934
Stripe and ring artifact removal with combined wavelet — Fourier filtering https://www.osapublishing.org/DirectPDFAccess/B171842B-CBA8-F140-B1102A9B5A490FE6_179485/oe-17-10-8567.pdf?da=1&id=179485&seq=0&mobile=no
7-The authors should explain the performance of their method with stripe with discontinuity.
8-The authors did not present the PSNR, SSIM or MSE results with different stripe levels.
9-The guided filter was used in the several paper, however, the authors did not compares their method with it. For example, the authors of the following paper used the guided filter:
Shengwei Zhang et al. , stripe Noise Removal for Infrared Images Using Guided Filter https://www.spiedigitallibrary.org/conference-proceedings-of-spie/10157/101572R/Stripe-noise-removal-for-infrared-images-using-guided-filter/10.1117/12.2247044.short
The authors should explain the steps and the parameters of the guided filters in the text in details.
10- The description of the wavelet decomposition is inaccurate. The authors should define all parameters, Ts, m, n, Z, etc.
11-The description of the stripe noise model is has few information. More details are required. The authors should use symbols in different places for different roles or different symbols for the same role. For instance, Xn and X in Eqs. 5 and 6.
12- What did the authors mean by ''binary calculation of the difference'', line 142?
13-The optimization process of Eq. 9 is not clear. More details are required.
14- The authors should provide the full names of all abbreviations, such as IRFPA and and NUC.
15-The quality of all figures should be improved.
16- The authors should add simulated experiments, where the stripes with periodic and non-periodic noise are mainly determined by “Intensity”. The authors may use publicly available images, such as the ones available at the available on the websites:
DigitalGlobe with http://www.digitalglobe.com/product-samples
https://earthexplorer.usgs.gov/
MODIS data https://ladsweb.nascom.nasa.gov/
17-The authors should list the limitations of their method.
