Round 3: Reviewer(s)' Comments to Author: Referee: 1 Comments to the Author (There are no comments.) Referee: 2 Comments to the Author Thank the authors for seriously considering the comments and revisions. Additional experimental results have been added, including some important ablation experiments. Many statements have been further clarified, and there is also a further review and comparison of existing works. This revised manuscript is now in much better shape and can be accepted. Round 2: Reviewer(s)' Comments to Author: Referee: 1 Recommendation: Needs Minor Revision Comments: The title of the reviewed paper is "FaceSigns: Semi-Fragile Neural Watermarks for Media Authentication." It has come to my attention that the term "Neural Watermarks" is not commonly used in the academic community and is not defined within the paper. I recommend either modifying the term or providing an appropriate definition within the paper. Additional Questions: Review's recommendation for paper type: Full length technical paper Should this paper be considered for a best paper award?: No Does this paper present innovative ideas or material?: Yes In what ways does this paper advance the field?: Is the information in the paper sound, factual, and accurate?: Yes If not, please explain why.: What are the major contributions of the paper?: The first deep learning based semi-fragile watermarking system that can certify the authenticity of digital media by introducing a deep learning based semi-fragile watermarking technique that allows media authentication by verifying an invisible secret message embedded in the image pixels. Rate how well the ideas are presented (very difficult to understand=1 very easy to understand =5: 5 Rate the overall quality of the writing (very poor=1, excellent=5: 4 Does this paper cite and use appropriate references?: Yes If not, what important references are missing?: Should anything be deleted from or condensed in the paper?: No If so, please explain.: Is the treatment of the subject complete?: Yes If not, What important details / ideas/ analyses are missing?: Please help ACM create a more efficient time-to-publication process: Using your best judgment, what amount of copy editing do you think this paper needs?: Light Most ACM journal papers are researcher-oriented. Is this paper of potential interest to developers and engineers?: Yes Referee: 2 Recommendation: Accept Comments: The author has addressed my concerns in the revised manuscript, and I have no further issues. Additional Questions: Review's recommendation for paper type: Full length technical paper Should this paper be considered for a best paper award?: No Does this paper present innovative ideas or material?: Yes In what ways does this paper advance the field?: This paper introduces a semi-fragile image watermarking framework, which aims to embed recoverable watermark into images and videos for media authentication. The framework is specifically designed to be semi-fragile, allowing for the recovery of watermark data under benign transformations, while making the recovery ineffective when faced with malicious transformations. Experimental results prove the effectiveness and efficiency of the proposed framework. Is the information in the paper sound, factual, and accurate?: Yes If not, please explain why.: What are the major contributions of the paper?: - A semi-fragile image watermarking framework for media authentication. - A differentiable tampering procedure simulation method. Rate how well the ideas are presented (very difficult to understand=1 very easy to understand =5: 5 Rate the overall quality of the writing (very poor=1, excellent=5: 4 Does this paper cite and use appropriate references?: Yes If not, what important references are missing?: Should anything be deleted from or condensed in the paper?: No If so, please explain.: Is the treatment of the subject complete?: Yes If not, What important details / ideas/ analyses are missing?: Please help ACM create a more efficient time-to-publication process: Using your best judgment, what amount of copy editing do you think this paper needs?: Light Most ACM journal papers are researcher-oriented. Is this paper of potential interest to developers and engineers?: Yes Round 1: Reviewer's comments Referee: 1 Recommendation: Needs Minor Revision Comments: The paper introduces a deep learning-based semi-fragile watermarking method FaceSigns that aims to proactively embed a semi-fragile watermark into a real image or video for future media authentication. This paper provides an in-depth analysis of the challenges of detecting synthetic or tampered media in a proactive manner. The experimental evaluation demonstrates that FaceSigns can embed a 128-bit secret as an imperceptible image watermark that can be recovered at different compression levels, while being non-recoverable when unseen malicious manipulations are applied. The proposed method is novel enough and the results are promising. Overall, the proposed method has sufficient contribution to the field and the paper is well-written and easy to follow. Pro: - The proposed FaceSigns is a deep learning-based semi-fragile watermarking technique that enables media authentication by verifying an invisible secret message embedded in the media. - The paper provides a comprehensive introduction to media authentication, which makes it easy to understand the challenges and importance. - The proposed method meets the desirable properties of watermark recovery, semi-fragility, and visual imperceptibility. - Figure 8 and Figure 9 show the excellent results of this paper for both benign transforms and malicious transforms. Cons: - In this paper, the author chose 128-bit watermark, but further research can be conducted on the watermark capacity, such 256, 512 or even more bits. - The authors introduce both reconstruction loss and discriminator loss to ensure image quality. What is the impact of the discriminator on image quality? The author can provide some results of ablation experiments. - Why the author didn't choose bit-wise BCE loss but l1 loss for the watermark recovery. - What about the results when the image/video undergo transformations that are unseen during training? Additional Questions: Review's recommendation for paper type: Full length technical paper Should this paper be considered for a best paper award?: No Does this paper present innovative ideas or material?: Yes In what ways does this paper advance the field?: The paper introduces a deep learning-based semi-fragile watermarking method FaceSigns that aims to proactively embed a semi-fragile watermark into a real image or video for future media authentication. Is the information in the paper sound, factual, and accurate?: Yes If not, please explain why.: What are the major contributions of the paper?: The proposed FaceSigns is a deep learning-based semi-fragile watermarking technique that enables media authentication by verifying an invisible secret message embedded in the media. Rate how well the ideas are presented (very difficult to understand=1 very easy to understand =5: 5 Rate the overall quality of the writing (very poor=1, excellent=5: 4 Does this paper cite and use appropriate references?: Yes If not, what important references are missing?: Should anything be deleted from or condensed in the paper?: No If so, please explain.: Is the treatment of the subject complete?: Yes If not, What important details / ideas/ analyses are missing?: Please help ACM create a more efficient time-to-publication process: Using your best judgment, what amount of copy editing do you think this paper needs?: Light Most ACM journal papers are researcher-oriented. Is this paper of potential interest to developers and engineers?: Yes Referee: 2 Recommendation: Needs Minor Revision Comments: Summary: This paper proposes a semi-fragile image watermarking framework to embed recoverable watermark data into images and videos for media authentication. The watermarking framework is designed to be semi-fragile: the watermark data is recoverable if the image/video undergoes benign transforms and the recovery breaks if the media has been maliciously tampered. Strengths: 1. The topic of the manuscript is highly relevant. 2. The manuscript is written in a clear manner. 3. The experimental results are quite positive. Weaknesses: 1. In the introduction section, the authors claim their first contribution is “… our work is the first deep learning based semi-fragile watermarking system that can certify the authenticity of digital media.” However, it has been brought to my attention that literature [1] has also studied this topic. Therefore, it is not appropriate to declare the proposed method as “the first... system.” The authors should describe their contributions more accurately and comprehensively. 2. There is no content in the manuscript that explains the meaning of Fig. 1 and Fig. 2. In the revised version, the authors should provide an explanation of the significance of these figures to help readers understand the ideas presented in the manuscript. 3. In the experimental section, the authors should explain the reasons for choosing the comparison algorithms. Additionally, it would be beneficial to include more comparison algorithms in this section, especially recent related work such as [1]. 4. The experimental results of "FaceSigns SemiFragile" in Fig. 5 appear slightly different from those in Fig. 12. The authors should provide an explanation for this discrepancy. Furthermore, it is unclear why SemiFragile DCT is not included in Fig. 12. 5. This paper should be carefully proofread (e.g., explanations for Fig. 1 and Fig. 2; the inconsistent use of "semi-fragile" and "semifragile" throughout the manuscript). [1] Hussain, Shehzeen, et al. "FastStamp: Accelerating Neural Steganography and Digital Watermarking of Images on FPGAs." Proceedings of the 41st IEEE/ACM International Conference on Computer-Aided Design. 2022. Additional Questions: Review's recommendation for paper type: Full length technical paper Should this paper be considered for a best paper award?: No Does this paper present innovative ideas or material?: Yes In what ways does this paper advance the field?: Is the information in the paper sound, factual, and accurate?: Yes If not, please explain why.: What are the major contributions of the paper?: This paper proposes a semi-fragile image watermarking framework to embed recoverable watermark data into images and videos for media authentication. The watermarking framework is designed to be semi-fragile: the watermark data is recoverable if the image/video undergoes benign transforms and the recovery breaks if the media has been maliciously tampered. Rate how well the ideas are presented (very difficult to understand=1 very easy to understand =5: 4 Rate the overall quality of the writing (very poor=1, excellent=5: 4 Does this paper cite and use appropriate references?: No If not, what important references are missing?: [1] Hussain, Shehzeen, et al. "FastStamp: Accelerating Neural Steganography and Digital Watermarking of Images on FPGAs." Proceedings of the 41st IEEE/ACM International Conference on Computer-Aided Design. 2022. Should anything be deleted from or condensed in the paper?: No If so, please explain.: Is the treatment of the subject complete?: Yes If not, What important details / ideas/ analyses are missing?: Please help ACM create a more efficient time-to-publication process: Using your best judgment, what amount of copy editing do you think this paper needs?: Moderate Most ACM journal papers are researcher-oriented. Is this paper of potential interest to developers and engineers?: Yes Referee: 3 Recommendation: Needs Major Revision Comments: This manuscript presents a new deep learning-based semi-fragile watermarking system. It aims to ensure the integrity of digital images and videos by reliably detecting tampering. The proposed framework outperforms existing methods by generating imperceptible watermarks while maintaining the desired semi-fragile characteristics. By incorporating a carefully designed set of benign and malicious transformations during training, the system demonstrates generalizability to real-world image and video transformations. It also excels in detecting Deepfake facial and image compositing manipulations, surpassing previous image watermarking techniques. The manuscript highlights the significance of FaceSigns for media authenticators in social media platforms, news agencies, and legal offices, as it contributes to creating more trustworthy platforms and fostering consumer trust in digital media. Reviewer's Comments: 1. The manuscript presents certain novel aspects and practical contributions. However, as an academic or research article, it would greatly benefit from incorporating more empirical and theoretical analysis. 2. The quality of writing and organization in the manuscript falls below the expected level for publication in this journal. It is recommended that the authors revise the manuscript to improve clarity and coherence. 3. It would be valuable to include more details and examples of the experiments and evaluations conducted to demonstrate the effectiveness of the proposed FaceSigns watermarking system. Providing a thorough analysis of the results would strengthen the credibility of the findings and support the claims made by the authors. 4. The theoretical foundations of the semi-fragile watermarking approach should be further elaborated. Including a more in-depth discussion of the underlying principles and concepts would enhance the scholarly contribution of the manuscript. 5. Finally, the authors should consider providing a more comprehensive discussion on the limitations and potential future directions of their work. Additional Questions: Review's recommendation for paper type: Full length technical paper Should this paper be considered for a best paper award?: No Does this paper present innovative ideas or material?: Yes In what ways does this paper advance the field?: Is the information in the paper sound, factual, and accurate?: Yes If not, please explain why.: What are the major contributions of the paper?: Rate how well the ideas are presented (very difficult to understand=1 very easy to understand =5: 3 Rate the overall quality of the writing (very poor=1, excellent=5: 3 Does this paper cite and use appropriate references?: Yes If not, what important references are missing?: Should anything be deleted from or condensed in the paper?: No If so, please explain.: Is the treatment of the subject complete?: No If not, What important details / ideas/ analyses are missing?: as commented below Please help ACM create a more efficient time-to-publication process: Using your best judgment, what amount of copy editing do you think this paper needs?: Moderate Most ACM journal papers are researcher-oriented. Is this paper of potential interest to developers and engineers?: Yes