Small Sample Palmprint Recognition Based on Image Augmentation and Dynamic Model-Agnostic Meta-Learning

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

2.2.3 DMAML-based training method for small-sample learning networks
section explained well with proper equation.

 

Figure 2 can be improve more.

 

3.1 Datasets and data preprocessing

in line 348
---> what methodlogy you applied for crooping ROI 128x128

 

It will be better to add one discussion part.

Overall paper is well written. No further revision need from my side.

Reference are given proper way.

COnclusion written in very well presented.
Table 1. Comparison of the results of the expansion experiments with different models:
did not understand. As the figure has no labeling. What kind of comparison actually showing.

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The paper proposes a method to enhance palmprint recognition. Here are some considerations.
1. In the abstract, the authors should provide a numerical comparison against the state of the art (i.e., highlight the improvements, not only the achieved quantitative values).
2. Please check the manuscript for the references, as it is an uncommon format.
3. Please check the text for typos.
4. Figures 3 and 4 are small-sized, hard to read, and should be improved, as they depict important parts of the proposed method. Furthermore, the description of the neural network should somehow be improved. For example, why do authors use imperative tenses, e.g., in line 170? Please provide a qualitative description, explaining the rationale behind the selection of the layers, and a quantitative, formal description, in Section 2.1.1. In other words, provide an improved, better version of the Section. For example, the authors provided adequate details in section 2.1.2; therefore, section 2.1.1 can be improved accordingly.
5. The module described in Section 2.2.1 is very similar to the standard Inception module used in GoogLeNet, with average pooling instead of max pooling. Please provide further details about the provided innovation. Furthermore, from the figure, it is not clear the use of residual connections.
6. The authors considered only channel attention. Did they also evaluate spatial attention?
7. Sections 2.2.2 and 2.2.3 provide bullet points describing algorithms, resulting in a cluttered and unclear narrative. The authors should focus on a more qualitative description, with a formal version of the algorithm proposed as a flow chart or written in a proper box, for readability.
8. The authors should briefly describe the evaluation metrics for the general audience.
9. The authors should provide further details on the actual training process (e.g., did they use hyperparameter optimisation? Did  they split the dataset? And so on.)
10. The algorithms used for comparison in 3.3 are outdated, ranging from proposals provided from 2014 to 2021. Did the authors provide an updated research on state-of-the-art methods? A quick search on Google Scholar highlights several contributions starting from 2024. The authors can select these for proper and fair comparison. The same goes in Section 3.6, with the exception maybe of FETA, which is significantly outperformed. All in all, I suggest the authors search nd consider more recent algorithms for comparison to provide more credibility to their thesis.

Overall, the work is interesting and well written. However, it requires some major improvements to its scientific soundness to be considered for publication.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Contributions:

This study presents a Dynamic Model-Agnostic Meta-Learning (DMAML) for recognizing a small-sample palmprint. My comments are as follows:

1. The proposed method is not novel.
2. This paper is not well presented.
3. (Line 104 on page 3) The sub-section title is wrong.
4. (Page 3) The output of the probability block is incorrect.
5. (Page 4) Figure 2 is incorrect. The size of the input image is 128*128; however, the size becomes 128*64 in the second layer of the encoding. It is wrong.
6. (Page 4) The two sub-blocks do not appear in the overall architecture in Fig. 2.
7. (Page 5) Is att a vector in eq. (1)? If so, it should be bolded.
8. (Page 6) How can you obtain D(x^hat)? Please present it in an equation.
9. (Line 205 on page 6) p(x^hat) did not appear in the equation. (2).
10. (Lines 210 to 223 on page 6)Please create a table to present the hyperparameters.
11. (Page 7) The circle_add operator was not defined in eq. (3).
12. (Page 7) The input and output are missing in Fig. 5.
13. (Page 8) Equation (5) is incorrect. Freqⁱ should contain parameters u and v.
14. (Page 8) Equations (5) and (6) should be merged.
15. (Page 8) Equation (7) is incorrect.
16. (Page 8) Is matt a vector in eq. (8)? If so, it should be bolded.
17. (Page 8) y_ic was not defined in eq. (9).
18. (Line 288 on page 8) Theta did not appear in the equation. (9).
19. (Page 8) Equation (10) is incorrect. The iteration parameter was missing.
20. (Pages 8 and 9) Inner and outer loops are unclear. Please provide a pseudo-code to present to them.
21. (Page 9) The derivation for w in eq. (11) was missing.
22. (Page 9) The two parameters “current_step” and “warmup_steps” should be presented by symbols in eqs. (13) and (14).
23. (Page 11) Table 1 should be a figure rather than a table.
24. (Page 11) The size of the original and generated images should be consistent.
25. (Lines 91 to 93 on page 3) This is a general statement. Please remove it.
26. There are many abbreviations in this paper. Please create an abbreviation table in the appendix.
27. (Line 112 on page 3) x_i was not defined.
28. (Line 155 on page 4) “lth” should be revised as “l^th”.
29. (Lines 141 and 142) The expression is not adequate. Page 5 also has the same problem.
30. (Line 162 on page 5)"l" was not defined.
31. (Line 165 on page 5) S was not defined.
32. (Page 5) The sub-caption is not well presented in Fig. 4. Please revise it according to the author’s guide.
33. (Pages 11 to 14) The sub-grid lines should be removed in Tables 2 to 6.
34. (Pages 14 and 15) The title should be removed in Figs. 6 and 7. In addition, the caption and figure should be on the same page.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors fixed all the highlighted issues, including figure quality, comparison with the state of the art, innovation description, and clarity. Therefore, I suggest the paper to be considered for publication.

Author Response

Thank you for your recognition and valuable suggestions. We will continue to work hard in the future!

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have improved the quality of this paper. I think it can be accepted for publication after a minor revision.
Minor comments:
In my previous comment 9: p(x^hat) did not appear in the equation. (2). The authors have provided eq. (3). However, the expression is not adequate. As well known, the symbol of random weight ∂ represents the partial derivative operator. Please change the symbol. In addition, there is a typo in the expression that is denoted as a rectangle.

In my previous comment 10: Please create a table to present the hyperparameters. However, I cannot find the table.

Comments on the Quality of English Language

The quality of the English language is acceptable.

Author Response

 Please see the attachment

Author Response File: Author Response.pdf