Retinex-Based Relighting for Night Photography

Round 1

Reviewer 1 Report

The authors present in this paper a method for improving the quality of images taken at night.

Although, the topic seemed interesting, the paper is not well written and therefore it is difficult to read. It has a lot of issues with the English language and style, i.e. incomprehensible phrases, grammar errors, typos, etc. I cannot list all the problems, only a some of them for reference:
- line 66: computer graphic -> computer graphics
- line 231: "Therefore, when light is added to the original image to the original image,..."

Also, the authors sometimes use not the professional terms (for example in line 55: "the only estimating thing").

As a suggestion, the word 'mean' is usually used instead of 'average' in scientific papers.

On page 7 there is a pseudocode of an algorithm, but I am not sure where the authors mention it in the text. Although, the pseudocode itself is not correct. For example:
- In line 3 there is  'random select a sample from A', but there is no A in that line.
-  Also, 'rng' function is not defined (I think it should be a random generator function).
- In line 6 (as a comment), there should be D' instead of D; in line 8 S instead of D'.
Also, it would be easier to rewrite if they were using a format for the pseudocode which has a line number. Thus, the pseudocode must be corrected.

In the equations, not all the variables are clearly defined, mainly the vaiables in the lower and upper indices. In addition, in line 103 (in the definition of Eq. (2) ) the authors use the character 'R' in three different style which make significantly more difficult to understand. There are a lot of other characters, why not use different ones?

Moving on to the results, I am not sure, that the visual results are satisfying. The problem is that difficult to compare the original images to the result ones, since they are in different figures, on different pages. It would be more transparent if the images could be examined in pairwise, say in a 'big' three-column figure (e.g. the images in the first row should be the current Fig.10(a), Fig.13(a), Fig.14(a), respectively), as they are in Figure 7. Furthermore, it seems as if the values of the green channel are sometimes overstated.

To conclude all, an overall correction must be prepared.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This is a very nice paper which process the night photography by using retinex-based model and a series of algorithms. The authors designed a sequence of  experiments and model comparisons to validate their approach. 

I have several comments as below:

1. I suggest the authors to add a flowchart, in order to systematically represent the algorithm with a global picture.

2. The English representation might be improved in several parts. For an example,

"The biggest difference" in line 46 at paragraph two of page 2, might be rewritten as "the sigficant difference".

3. I suggest to represent "process that is entirely independent of the convolution radius" in line 77. To my understanding, this radius should be restricted in a reasonable scope.

4. For the color constancy, I suggest to supply references to support equation (2).

5. In equation (8), the geodetic distance and mix-Euclidean-geodetic distance is not clearly defined.

6. I suggest to add an illustration for the function  Q_p in equation (9), to help the readers to understand the shape of the set.

I strongly recommend  to ACCEPT it for publication, with minor modifications.

Author Response

 Please see the attachment, starting from Page 4.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The paper has improved significantly. Thank you for considering all my suggestions! I have left only a few minor recommendations as follows.

According to the pseudocode of Algorithm 1:

-        there is an extra ) at the end of Line 5

-        I think, that in Line 18, there should be D instead of ||D||

-        as well as in line 27 A instead of ||A||

-        in Line 24, should be |A| instead of ||A||

-  maybe the same in Lines 13 and 14 for ||T|| and |T|

According to the flowchart (which I found very useful):

-        there is an error at the point, where the comparison of N_l > |A| is made. After that, N_l = |A| is examined. If the inequality is correct, it cannot be equal to the one from which is greater.

-        I think that in the sentence ‘Random select a sample p_next from D′ and set p_next to A’ should be ‘add’ instead of ‘set’

I also find the Symbols table very useful.

Great work!

Author Response

Please see the attachment

Author Response File: Author Response.pdf