Forgery Detection for Anti-Counterfeiting Patterns Using Deep Single Classifier

Round 1

Reviewer 1 Report

The paper proposed a deep learning single classifier for Forgery Detection of Anti-counterfeiting Patterns. The idea is valid, and it could be beneficial in many applications. However, the paper has the following points to be considered:

-        The abstract did not show the results gained.

-         Deep Learning has been used in similar papers for the same problem. The authors should clarify the difference between their approach and other papers. Here are some of the recent work on the same problem:

o   Lee, S. H., & Lee, H. Y. (2018). Counterfeit bill detection algorithm using deep Learning. Int. J. Appl. Eng. Res, 13, 304-310.

o   Sharma, P., Kumar, M., & Sharma, H. (2023). Comprehensive analyses of image forgery detection methods from traditional to deep learning approaches: an evaluation. Multimedia Tools and Applications, 82(12), 18117-18150.

o   Ali, S. S., Ganapathi, I. I., Vu, N. S., Ali, S. D., Saxena, N., & Werghi, N. (2022). Image forgery detection using deep Learning by recompressing images. Electronics, 11(3), 403.

o   Sharma, P., Kumar, M., & Sharma, H. (2023). Comprehensive analyses of image forgery detection methods from traditional to deep learning approaches: an evaluation. Multimedia Tools and Applications, 82(12), 18117-18150.

I am not sure why the authors compared their work with the traditional methods where there is similar work done on the same problem

Okay

Author Response

Point 1: The abstract did not show the results gained.

 

Response 1: Much thanks for your observation. We have revised the abstract and now the result is added to the abstract.

 

Point 2: Deep Learning has been used in similar papers for the same problem. The authors should clarify the difference between their approach and other papers.

 

Response 2: Thank you very much for your rigorous views, we have add a comparison with some deep learning based methods in section 4. Also we have revised the dicussion section and added some expansion, compared with multi-classes classification approachs, there are few methods for deep one class classification due to the difficulty in training with only one normal class.

 

Point 3: I am not sure why the authors compared their work with the traditional methods where there is similar work done on the same problem

 

Response 3: Thanks for your observation. We have add a comparison with some recent methods in section 4.

Reviewer 2 Report

In this paper the authors have proposed a deep learning based authentication scheme for anti-counterfeiting pattern captured by smartphones. In the hybrid method,  they have used u-Net for feature extraction and OCSVM for classification. The content is very interesting; however, the following points need to be addressed for improving the quality of the manuscript.

Major concern:

·         It will better to illustrate how the hamming distance and jaccard similarity is calculated through example with two sample binary images in section 3.1.

·         Explain the significance of all the symbols used in Eq.(9) ,(10), (11) and (12).

·         Clearly specify the splitting of training and test set of the original dataset and type of validation used in the experimentation.

·         Explain clearly, what is represented by C1, C2, C3 and C4 in section 4.3. Same way give explanation for C5, C6, C7.

·         Compare the model with some SOTA techniques and other classification criteria.

·         Put a separate discussion section (containing aim, result, finding of the work and limitation) before conclusion.

 

Minor points:

·         In line 371, instead of writing the statement “Generate 90 anti-counterfeit code digital images using the anti-counterfeit code generation system” write as “90 anti-counterfeit code digital images are generated using the anti-counterfeit code generation system”

Author Response

Point 1: It will better to illustrate how the hamming distance and jaccard similarity is calculated through example with two sample binary images in section 3.1.

 

Response 1: Thanks for your views, we have revised the metrices calculation process in section 3.1 and the construction process of feature vector is illustrated.

 

Point 2: Explain the significance of all the symbols used in Eq.(9) ,(10), (11) and (12).

Response 2: Much thanks for your observation, we have added explansion of the significance of symbols used in related formulations.

 

Point 3: Clearly specify the splitting of training and test set of the original dataset and type of validation used in the experimentation.

Response 3: Thanks for your views, we have revised section 4.2 and the spliting of training and testing set in both two stages is illustrated.

 

Point 4: Explain clearly, what is represented by C1, C2, C3 and C4 in section 4.3. Same way give explanation for C5, C6, C7.

Response 4: Thank you very much for your rigorous views. We have added footnotes to the table 2 and 3. We hope the notes can help illustrated meanings of the contents.

 

Point 5: Compare the model with some SOTA techniques and other classification criteria.

Response 5: Thanks for your observation. We have add a comparison with some recent methods in section 4. And the performance criterias used in experiment are illustrated in section 4.2.

 

Point 6: Put a separate discussion section (containing aim, result, finding of the work and limitation) before conclusion.

Response 6: Thanks for your views, we have replaced the conclusion section with dicussion section and the aim, result, finding of the work and limitation are added.

 

Point 7: In line 371, instead of writing the statement “Generate 90 anticounterfeit code digital images using the anti-counterfeit code generation system” write as “90 anti-counterfeit code digital images are generated using the anti-counterfeit code generation system.”

Response 7: Thanks for your observation. We have revised the statement.

 

 

Round 2

Reviewer 1 Report

I have no further comments. 

NA

Author Response

I have no further revision.

Reviewer 2 Report

All of my concerns are addressed. It will be better give conclusion after discussion section.

Author Response

Point 1: All of my concerns are addressed. It will be better give conclusion after dicussion section.

 

Response 1: Much thanks for for your evaluation and affirmation of our work, we rewrite the conclusion section as supplement to the discussion section.

Author Response File: Author Response.docx