Performance Improvement of Image-Reconstruction-Based Defense against Adversarial Attack

Round 1

Reviewer 1 Report

The revised version of the paper 1833533 "Performance Improvement of Image Reconstruction Based Defense against Adversarial Attack" has been improved with respect to its original version. Authors can also improve the paper further by including some related references, such as the followings:

- Zichong Chen, Xianwen Luo, "Fuzzy Control Method for Synchronous Acquisition of High Resolution Image based on Machine Learning," International Journal of Circuits, Systems and Signal Processing, vol. 16, pp. 367-373, 2022.

- N. Shylashree, M Anil Naik, A. S. Mamatha, V. Sridhar, "Design and Implementation of Image Edge Detection Algorithm on FPGA," International Journal of Circuits, Systems and Signal Processing, vol. 16, pp. 628-636, 2022.

 

- Stella Vetova, "A Comparative Study of Image Classification Models using NN and Similarity Distance," WSEAS Transactions on International Journal of Electrical Engineering and Computer Science, vol. 3, pp. 109-113, 2021.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

General Comments:

The paper deals with image classification using deep neural networks (DNNs) and presents an approach to enhance the time-accuracy tradeoff in reconstruction-based defense methods that defend DNNs against adversarial attacks. Good reduction in time-complexity is obtained at the price of little drop in accuracy for DIPDefend method. The topic is important in image classification and the proposed approach is interesting.

 

Specific Comments:

1.    Section 2:

a)   Figure 4: The instant {\t_o} should be indicated. Also, clarify what is “A” and “B”.

b)   Equations (1) and (2): Please clarify {\p_t}, {\s_t}.

 

2.    Section 3: The structure of the LPF should be clearly explained (or referenced). Also, the kernel of the LPF should be defined.

 

3.    Section 4: Does the monotonicity of the curve in Figure 6 ensure that a peak in SES-PSNR always exists?

 

Typos:

Please use “iterations” instead of “iteration” in the x-axis of Figure 5.

In Figure 6: “Epsilon” instead of “Epslion”.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf