Generative Adversarial Network for Image Super-Resolution Combining Texture Loss

Round 1

Reviewer 1 Report

The paper described a new algorithm to image super-resolution reconstruction.

The paper is well written, organized and interesting to the reader.
Some suggestions for improving the paper:

Define some acronyms in the text such as CNN, SRCNN, SRGAN, LR, ReLU, LeakyReLU, MSE, VGG, ... The first reference mus be defined as [1] (line 38) Line 43, change to "... Generative Adversarial Network (GAN) and [8]..." Line 45, change to "... Batch Normalization (BN) ..." Line 61, add [8] after ".... by Ian Goodfellow et al. ..." Equation (1) just put the equation on a single line, separate the words "min" and "max". All the equation in text should appear in mathematical format. Line 76 add the word "where ..." in the begining Line 117 replace the word "Here ..." to "where ..." Line 119 change ".... RB (Residual Block) ..." Line 135 change ".... Global Average Pooling (GAP) ..." Line 148, 156, 173, 188,  replace the word "Here ..." to "where ..." The number of the equations must be placed at the same distance from the right margin in all equations. Line 186 and 187 the word "gram" should be change to "Gram". Defined "gen" in the equation (7). Sometimes appears L2 (line 186) and others L₁ (line 200). Line 207 and 208 change "... Peak signal to Noise Ratio (PSNR) and Structural Similarity Index (SSIM)..." Increase the equation (8) and (9). It is difficult to see. Define the acronymous Set5 and Set14, line 216. Page 240 put the word "Fig." before number 6. The sequence of images in figure 5 and 6 must be all in one line or divide 3 images per line. References  Remove [J], [C] in the references. Put a final point before the year in references [9,11,12, 14,17,23,27] Put a space before the year in references [8,13,21,26]

Author Response

Thank you for your review. We have made corresponding amendments to the manuscript based on your comments. All changes are highlighted in red in the manuscript.

Author Response File: Author Response.docx

Reviewer 2 Report

I have minor concerns about this work:

Related work and Methodology should be separate sections In the related work, authors should describe their work, how it is different from other recently published studies.

Author Response

Thank you for your review. We have made corresponding amendments to the manuscript based on your comments. All changes are highlighted in red in the manuscript.

Author Response File: Author Response.docx

Reviewer 3 Report

First of all, I want to congratulate to authors for a well written manuscript.

In my opinion, one topic  is the most important. Wht that model?? It is necessary to compare different models, and justify the structure.

Kind Regards.

Author Response

Thank you for your review. We have made corresponding amendments to the manuscript based on your comments. All changes are highlighted in red in the manuscript.

Author Response File: Author Response.docx

Reviewer 4 Report

The manuscript presents a new GAN-based model for solving an image super-resolution problem. The proposed model, referred to as TSRGAN, is some architectural modification of the SRGAN model which was proposed in [3]. All the suggested modifications are well justified, and the results seem to be sufficiently strong. However, the writing style and the introduction to the topic should be considerably revised. First, the manuscript in its current form is not suitable for potential readers of a multi-disciplinary journal, such as Applied Sciences. The introduction has a limited scope and wrongly restricts the methodology for the single image super-resolution (SISR) problem to interpolation methods or deep learning architectures. It is not true and many other methods exist to tackle an exactly the same problem. Moreover, the neural network-based super-resolution methods usually need a large set of training images. In this respect, the term ‘’single’’ in the context of the discussed SISR is somehow misleading and should be clarified. Hence, the introduction needs to be revised and alternative methods for SISR should be discussed. For example, the SISR problem can be formulated in terms of sparse representation (Yang et al, Image Super-Resolution Via Sparse Representation, IEEE Trans. Image Proc., 19(11) 2010 ), low-rank matrix approximation  (S.Z. Tang et al, Single Image Super-Resolution Method via Refined Local Learning. 2015), or tensor decomposition models (J. Hatvani et al,  A Tensor Factorization Method for 3-D Super Resolution With Application to Dental CT, IEEE Trans. Med. Imaging, 2019; R. Zdunek et al, Image Completion with Hybrid Interpolation in Tensor Representation, Applied Sciences, 2020). There are also other interesting approaches. It would be good to compare the proposed method with other related methods or if not possible due to a short time for revision, the competitive methods should be shortly discussed or at least cited. If some Python or Matlab implementations are taken from GitHub, the right citations should be added. The revised version of the manuscript should also consider the following comments:

1. Any abbreviations, especially in the abstract, should be explained. The English needs a moderate revision.
2. The mathematical symbols are not always well defined, e.g. in (7) and (9). Please clarify them.
3. Section 2.1: the first sentence. The reference to the work by Ian Goodfellow is missing.

Author Response

Thank you for your review. We have made corresponding amendments to the manuscript based on your comments. All changes are highlighted in red in the manuscript.

Author Response File: Author Response.docx

Round 2

Reviewer 4 Report

Thank you for revising the manuscript accordingly. The minor revision of the English is recommended but this process can be done by an editorial office after accepting this manuscript for publication.