Low-Rank and Total Variation Regularization with ℓ0 Data Fidelity Constraint for Image Deblurring under Impulse Noise
Round 1
Reviewer 1 Report
This paper proposes a new method for removing impulse noise that combines the nuclear norm and the detection ℓ0TV model while considering the low-rank structure commonly found in visual images. The nuclear norm maintains this structure, while the detection ℓ0TV criterion promotes sparsity in the gradient domain, effectively removing impulse noise while preserving edges and other vital features. The paper is interesting and well-written but needs minor revision:
1. It'll be better to present an algorithm as a flowchart.
2. The presentation of Table 3 should be improved.
3. The paper needs part Discussion.
4. It'll be better to present a computational complexity of the proposed approach and compare it with other methods.
Minor editing of English language required. The authors should fix typos in the text.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Summary:
This paper proposes a new method for removing impulse noise in visual images. The proposed method combines the nuclear norm and the detection model while considering the low-rank structure commonly found in visual images. To solve the nonconvex and nonsmooth optimization problem, the authors use a mathematical process with equilibrium constraints (MPEC) to transform it. They then use the proximal alternating direction multiplication algorithm to solve the transformed problem. The convergence of the algorithm is proven under mild conditions. The experimental results show the effectiveness of the proposed methods.
Concerns:
1. The utilization of Low Rank Prior and Total Variation Regularization in image deblurring is widespread, thereby limiting the novelty of this paper.
2. To demonstrate the efficacy of the proposed method, it is recommended that the authors evaluate it on real-world rainy images captured by various devices, such as those found in the RWBI dataset.
3. Although the authors conducted experiments, they did not take into account several state-of-the-art methods. To enhance the comprehensiveness of the paper, it is suggested that they review and cite the latest survey papers on deep deblurring.
4. Comparing the complexity of different methods in terms of parameters, running time, or FLOPs would be beneficial.
5. The code for the proposed method is currently not publicly available.
6. To generate superior deblurred images, it is suggested that the authors of the proposed method engage in a discussion with regard to current GAN-based deep deblurring techniques such as DBLRNet and DeblurGAN.
.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The figure captions state only the obvious. Making the figure captions more complete and comprehensive would be very useful, as many times readers will be interested just in them and the captions, without necessarily giving full attention to the main text. What one is really curious about when looking at a figure is what is the main message; the most interesting feature that one should look out for, not just what is presented.
Transfer learning is a popular way to generate image representations, and has been used in medical image analysis. Please discuss the methods in ' NAGNN: classification of COVID‐19 based on neighboring aware representation from deep graph neural network' and 'Detection of abnormal brain in MRI via improved AlexNet and ELM optimized by chaotic bat algorithm'.
The authors need to analyze the potential reasons why the proposed method achieved better results than state-of-the-art approaches.
Please provide some drawbacks of your method and some future research directions in Conclusion.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
The authors have addressed several of the concerns raised. However, to further demonstrate the effectiveness of the proposed method, it is recommended that the authors conduct experiments on the RWBI dataset, which consists of real-world blurry images captured by a variety of devices. This additional evaluation would enhance the validation of the method's efficacy in handling real-world scenarios with diverse image quality.
n/a
Author Response
Thank you again for reviewing our research paper. We have applied our method
to real-world blurry images for image processing. Figure 1 and Figure 2 show actual blurry images and the restored images at different blur kernels. Figure 3 shows the real defocused blurred image and the restored images at different blur radius parameters r. Based on the images shown in Figure 1-3, it can be seen that our method has produced images that are slightly sharper than the original for some blur kernels. But we found that the effect is not particularly significant. However, our algorithm performs well in processing images with real
impulse noise. We think one possible reason is that our algorithm needs to know the blur kernel to complete the deblurring process. However, estimating the accurate blur kernel may be more difficult due to the more complex and imperfect blur process in the real world than the theoretical model. Therefore, in future research work, we will continue to explore more accurate algorithms for measuring the blur kernel and incorporate them into our algorithm to
expect better results. Meanwhile, we also believe that it is possible to explore more complex image processing algorithms, such as combining deep learning methods with our algorithm, to better process real-world images.
Author Response File: Author Response.pdf
Reviewer 3 Report
The paper is acceptable.
Author Response
Thank you very much! Thank you for your kind consideration!
