Near-Infrared Image Colorization Using Asymmetric Codec and Pixel-Level Fusion

Round 1

Reviewer 1 Report (Previous Reviewer 3)

Thank you very much for your reply

The manuscript still have the same issues that in the first version

The first decision was reject and no changes were provided that solved the issues on the first reviewer round. 

This manuscript do not have the interest to the scientific community in order to continue with the publication process. 

The practical daily application of the finding do not have enough interest to the research community. 

Thanks for the effort

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report (Previous Reviewer 2)

Please, publish the manuscript as it is. 

Author Response

Thank you so much for your positive evaluation.

Reviewer 3 Report (New Reviewer)

This manuscript is about the colorization of NIR images. The authors suggested a new method that is more effective than existing methods. After a few minor revisions, there is no problem with publication.

1. In the Abstract section, describe what BFWLS stands for.

2. Describe what B stands for in Equation (4).

3. In the text that describes the superscripts of Equations, write the superscripts in italics.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report (New Reviewer)

This paper proposes an original NIR colorization method using Asymmetric codec and pixel-level fusion.

It can generate high quality RGB images in different natural scenes.

Authors compare classical colorization methods based on common dataset,

and the experimental results show that the proposed method well enhance color textures.

There are some minor points to correct (see joint PDF).

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report (Previous Reviewer 3)

The revised manuscript do not solved the original proposed comments

The research have the same methodological issues that have been adressed in the first round of revision

Thank you very much for the revised paper

 

Author Response

Please see the attachment

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 paper, a NIR image colorization algorithm is proposed. This work is based on the following stages: asymmetric codec (ACD) that is based on the ResNet to the encoder and the UNet to decoder, a global and local feature fusion network (GLFFNet) that is based on two convolutional blocks, is located between the encoder and the decoder to reduce the loss information in the pooling layer. Both ACD and GLFFNet are called ColoNet. Finally, bilateral filtering and weighted least squares filtering (BFWLS) is used to fuse the pixel-level information of the input NIR image with the luminance of the output image obtained from the ColorNet. To consider the paper for publication, authors must be considering the following concerns:

Strong English revision and punctuation is required.

Several format mistakes are in the affiliation format.

All the abbreviations must be defined previously for their use.

I suggest polishing the Introduction according to 1.-justification, 2.-background, 3.-description of the proposed work, and 4.-organization of the proposed work.

In 2.1 check the “I” of Image

In 2.2 an extra “]” is provided in ref [24], ref [24] is provided for Suarez et. al. [24] or for Both [24]. Reference citation must be updated.

In 3.1 asymmetric codec (ACD) term was previously defined in the introduction.

In 3.1.1, line 168, I suggest rewrite the mentioned activation function tanh() by tanh(x).

Could you further in deep into the description of equation 4?

In 3.2 I suggest justifying the use of YUV color space.

Into the text is not provided the definition of d and b of the equations 5 and 6, I suppose that d refers to a detailed image and b to a base image, I suggest including these definitions.

In Line 260, I suggest not using the adverb ‘obvious', because nothing in science or in a mathematical context is obvious.

In 4.1 a weak justification of the experimental results is provided, asseverate that the proposed algorithm shows better results according to the vegetation color (Authors opinion) is not a correct result, in this way, I suggest including a Mean Opinion Score (MOS) with 250 opinions as a minimum.

To evaluate numerically the colored image (RGB image) obtained from the proposed algorithm I recommend including the Normalized Color Difference (NCD). The results should be included in Table 1, Table 2, and other evaluations where it is required.

Consider if the use of the PSNR is a feasible numerical evaluation, due to the PSNR compares two signals the original image obtained from the dataset [32], and the one obtained by the proposed algorithm. I think that the use of the PSNR is not a correctly numerical evaluation due to ideally you do have not the original color image. If this comment is not appropriate, explain in detail the correct way to obtain this evaluation.

Consider including the equations and their descriptions of all the parameters used in the performance evaluation of the proposed work (PSNR, MSE, SSIM, and NCD).

Consider the author guide to citing Figure 8, figure 9, and Figure 10. In another hand, when the figures are divided into an n-by-m grid each subfigure must be numbered by an index such as a), b), c), ..., etc.

 

Could you include a numerical evaluation for the Ablation studies section? If it's not possible, include the NCD and a MOS of the performance evaluation. PSNR, MAE, and SSIM are not enough parameters to evaluate the performance of the proposed work, because they show poor values.

Reviewer 2 Report

In this paper a novel NIR colorization method is proposed using asymmetric

18 codec (ACD) and pixel-level fusion.

The paper is considered significant for the following reasons

a.       The theoretical basis used for developing the proposed procedure is fully justifiable.

b.      The mathematical background used is detailed and descriptive.

c.       The manuscript is fully supported with block diagram and figures.

d.      The manuscript is well organized.

e.      Related work is very well reviewed and the references are up to date.

f.        Comparisons with existing techniques are satisfactory and prove the significant capabilities of the proposed technique.

Reviewer 3 Report

The authors mainly study the colorization of near-infrared images. Image colorization methods cannot be extended to NIR image colorization since the wavelength band of the NIR image exceeds the visible light spectral range and it is often linearly independent od the luminance of the RGB image.

The found that an intensive comparison analysis based on common datasets is conducted to verify superiority over existing methods in qualitative and quantitative visual assessments.

The main issue in this manuscript is that the authors plagiarism the content from other sources and the text of the paper it is not original.

In the following, I am going to report the main plagiarism articles:

Regular article

An improved DualGAN for near-infrared image colorization

Author links open overlay panelWeiLiangDeruiDingGuoliangWei

LinkNet: Exploiting Encoder Representations forEfficient Semantic Segmentation

Colorizing Near Infrared Images Through a Cyclic Adversarial Approach of Unpaired Samples

The manuscript should be rejected due this anti ethical issue