Single-Core Multiscale Residual Network for the Super Resolution of Liquid Metal Specimen Images

Round 1

Reviewer 1 Report

The aim of the paper is to introduce a CNN based on the VGG architecture to increase the resolution of images. The proposed method is used to analyze liquid metal specimen images. The results have been evaluated by SSIM and PSNR similarity measures. Also, the computational complexity evaluation has been performed. A comparative analysis of the proposed method against some of the most commonly used ANN super-resolution algorithms is provided in the final part of the paper.

This paper deals with an interesting subject. The methodology is clearly described. The provided experiments seem to prove good performances of the proposed approach.

However, some minor corrections should be made, as follows:

-           Line 122 – the variable used in Equation 2 is denoted by k, not K 

-           The authors should provide further explanations regarding Figure 1

-           Figure 2 is unnecessary 

-           The authors should briefly describe their contribution to the research filed in the introductory part of the paper.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The manuscript introduces a new Neural Network for superresolution of images having as an application liquid metal specimen images. The proposed method is well based on the consolidated literature, and indeed generalizes them with a multiscale analysis of the images, by a suitable multiscale residual network.

The subject is of paramount interest. The manuscript is well written and organized (but with several misprints and some unclear sentences, see later). The methods are technically sound, and the results convincing. Therefore the manuscript deserves consideration for the publication in the journal.

I woulkd ask the authors to address the following issues:

• The description of the data set is deficitary in that it is not clear how many images authors use for the analysis.
• The procedure for defining the training and test data sets is not clear. From one hand, authors report (Section 3.2) that to avoid oversmoothing, some augmented images are generated from the original ones (e.g., by rotation, zoom). Therefore it seems that they do not distinguish between a training and a test data set. Later on (Section 4.3.1), while dealing with Sub-module Analysis, authors seem to use a simple Hold-Out analysis (split of the data set into a training and test data set), without specifying the proportion of the two data sets, nor if augmented images are also considered. Authors are requested to clarify in only one part (better Section 3.2) how training/test is made. Moreover the most consolidated procedure for a training/test policy is K-fold Cross Validation (with K=10 if data allow), that solves problems due to the unique sampling of Hold-Out. It would be also preferable to have double policies: the first one where 17x17 subimages are randomly split between training and test data sets, the second one (as apparently made in Section 4.3.1) where whole images are split between the two training and test data sets.
• Figures 7-9 should be rearranged: Figure 8 is somewhat redundant, so that I think it would be better to show only the zoomed detail of the images (originally showed as an inset) removing the entire image, already shown in Fig. 7. Figure 9 is obscure, besides the fact that the contrast chart of image contours has never been defined. In the other side, Figure 9 does not seem to have been recalled in the text, an it does not seem to include additional information. Authors should consider to remove it.
• Authors seem to have problems with numbering of Figures and Tables (possibly they did not use a referring system in the original Word document). Therefore they are requested to pay much attention in recalling Figures and Tables with the right numbers.
• Authors also are not careful in addressing variables with the same letters in equations and in the text. In particular, small and capital letters DO matter, so that they have to be used absolutely consistently
• One English word "boldened" is not in the common use and should be replaced
• A couple of sentences are obscure and should be rephrased

All editing issues are reported as highlighted in the attached file

 

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf