Image Steganalysis of Low Embedding Rate Based on the Attention Mechanism and Transfer Learning
Round 1
Reviewer 1 Report
Reviewer #1: Comment 1: Authors work on problems of difficulty in extracting steganographic features from images with and 9 unsatisfactory detection performance of steganalysis. The method constructs a network model based on convolutional neural network, This method should provide justifiication snf comparision with another existing methods Comment2: "The proposed image steganalysis method based on attention mechanism and transfer learning achieves significant improvement" should be described Comment3: The proposed method (3. Proposed Method) should include references and comparision of parameters and results of experimental/simulation implementation should be given in numerical form in the abstract and conclusion. Comment 4: The work is having novel implementation review improve the steganalysis performance scope and used method should be compared and implemented suitably. Comment 5: The paper can be drafted in journal format. Comment 6: Improve the quality of figures and explain those properly. Comment 7: There are many English and grammatical issues in the paper which needs to be rectified.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
In this paper, the authors propose an image steganalysis method based on attention mechanism and transfer learning. The convolutional neural network model includes preprocessing, transposed, ordinary, and fully connected layers. After the standard convolutional layer, the authors add the efficient channel attention module to concentrate on the steganographic area of the picture, gather local cross-channel interaction information, realize adaptive feature weights, and improve feature extraction. The authors employ transfer learning to use the training model parameters of high embedding rate pictures as the initialization parameters of the low embedding rate training model to accomplish feature migration and enhance steganalysis results. Here are some comments for the further improvements
1. The compared models should have their relevant references in the experimental section
2. In Fig. 3, how does the left part convert to the right part? and how does the source domain affects the model?
3. How does the target domain work here? as there is not any arrow showing the direction
4. how do you setup the value for high- and low- embedding rate? any threshold value here?
5. what EC stands for?
6. Some image-based learning models should be studied
- A comprehensive review of video steganalysis
- Digital image feature recognition method of mobile platform based on machine learning
- A target recognition algorithm of multi-source remote sensing image based on visual Internet of Things
- An Intelligent Collaborative Image-Sensing System for Disease Detection
-Image steganalysis based on attention augmented convolution
7. The designed model should be compared with SOTA models published in recent 2 years
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
In this paper, the authors have presented an attention mechanism and transfer learning-based method for image steganography. The CNN based network model is built. It comprises of preprocessing, transposition, regular, and fully connected convolutional layers. A channel attention stage is incorporated after the standard convolutional layer. It permits to focus on the steganographic region of images, record local cross-channel interaction data, implement adaptive feature weight modification, and improve the capacity to extract steganographic features.
For authors of this paper, I have following remarks.
1. It is required to highlight the research gap and contribution of this paper.
2. The dataset used in this study should be clearly presented.
3. A comparison of performance with state-of-the-art concurrent approaches should be made, preferably in a Tabular form, while highlighting the key methods and findings.
4. Carefully proofread the manuscript to avoid any typos and grammatical mistakes.
5. The authors are invited to cover interesting aspects related to the Multiple Sclerosis Lesion Segmentation in Brain MRI Using Inception Modules Embedded in a Convolutional Neural Network.
This idea is proposed in the following reference:
https://doi.org/10.1155/2021/4138137
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Autho response to the reviewers queries are good and you can continue the future work on the research gap.
Comments: minor grammar and spelling check
Author Response
We sincerely appreciate the valuable comments. We tried our best to improve the manuscript and checked out some grammar and spelling problems. These problems were marked in yellow in the revised paper. In Addition, we also revised the introduction and conclusion.
Reviewer 2 Report
fine with me
Author Response
Thank you very much for your approval of the paper. In this revision, we have further improved the quality of the paper, checked the grammar and spelling problems and marked them in yellow, and also revised the introduction and conclusion.
Reviewer 3 Report
The authors have incorporated most of the recommended changes. However, it is recommended that the authors refer the following paper and its interesting aspects in future directions: https://doi.org/10.1155/2021/4138137
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
