Deepfake Detection By using Watermark

Authors: B. Laxmi Narayana, Sk. SaiBabu, P. Chandrasekhar, V. Yohan, R. Vijaykumar, R. William Carey

DOI Link: https://doi.org/10.22214/ijraset.2026.78123

Abstract

This paper presents a comprehensive semi-fragile watermarking framework for deepfake detection based on Discrete Wavelet Transform (DWT) and Singular Value Decomposition (SVD). The proposed system embeds an imperceptible watermark in the frequency domain of digital images and verifies integrity through Bit Error Rate (BER) analysis. Unlike deep learning-based approaches, the system is deterministic, computationally efficient, and does not require training datasets. Experimental evaluation on 2,041 images demonstrates high visual quality (PSNR ? 51 dB, SSIM ? 0.998) and strong detection accuracy (~96%). The framework provides a proactive authentication mechanism suitable for digital forensics and media verification.

Introduction

The rapid advancement of deepfake technologies, including GANs and diffusion models, has made detecting fake images increasingly difficult, raising concerns about misinformation, identity theft, and tampered digital evidence. Conventional machine learning–based detection methods require large datasets and retraining, and often fail to generalize to new deepfake methods. This work proposes a proactive, deterministic solution using semi-fragile watermarking based on hybrid Discrete Wavelet Transform (DWT) and Singular Value Decomposition (SVD). The watermark is embedded imperceptibly in the image’s frequency domain, robust to normal processing but disrupted by malicious deepfake modifications. The system embeds, extracts, and evaluates watermarks using Bit Error Rate analysis to classify images as real or fake, achieving 96% detection accuracy while maintaining high visual quality (PSNR ≈ 51?dB, SSIM ≈ 0.998). Unlike AI-based methods, it is training-free, computationally efficient, and provides an explainable approach for authenticating digital images.

Conclusion

This paper has discussed a semi-fragile watermarking technique for deepfake image detection using the DWT-SVD transformation model. The proposed method embeds a semi-fragile watermark in the frequency domain of an image and relies on Bit Error Rate (BER) analysis for image tampering detection. Unlike machine learning-based deepfake image detection techniques, the proposed method does not require any training data, neural networks, or GPU support. Experimental results showed that the embedding of the watermark satisfies the requirements of high visual quality with an average PSNR of 51 dB and SSIM of 0.998, thus proving the imperceptibility of the watermark. The results of the robustness test showed a clear separation of the BER values between the original and manipulated images, thus facilitating accurate classification. The proposed system attained a detection accuracy of about 96% on a total of 2,041 images with low FP and FN rates. Despite the fact that the framework is dependent on the embedding of the watermark in advance and is vulnerable to some geometric and compression attacks, the framework is a lightweight and scalable solution for digital content authentication. The paper also emphasizes the importance of signal processing methods in dealing with the manipulation of AI-generated media. The system can develop into a comprehensive deepfake detection framework with future improvements such as adaptive embedding and security hardening.

References

[1] I. Daubechies, Ten Lectures on Wavelets. Philadelphia, PA, USA: SIAM, 1992. [2] R. C. Gonzalez and R. E. Woods, Digital Image Processing, 4th ed. Pearson, 2018. [3] G. H. Golub and C. F. Van Loan, Matrix Computations, 4th ed. Johns Hopkins Univ. Press, 2013. [4] Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: From error visibility to structural similarity,” IEEE Trans. Image Process., vol. 13, no. 4, pp. 600–612, Apr. 2004. [5] C.-T. Hsu and J.-L. Wu, “Hidden digital watermarks in images,” IEEE Trans. Image Process., vol. 8, no. 1, pp. 58–68, Jan. 1999. [6] A. Balafrej et al., “Enhancing practicality and efficiency of deepfake detection,” Sci. Rep., vol. 14, 2024, pp. 1–10. [7] P. Liu, Q. Tao, and J. T. Zhou, “Evolving from single-modal to multi-modal facial deepfake detection: A survey,” arXiv, Jun. 2024. [8] N. A. Chandra et al., “Deepfake-Eval-2024: A multi-modal in-the-wild benchmark of deepfakes circulated in 2024,” arXiv, 2025. [9] R. Ramanaharan, “Deepfake video detection: Insights into model generalisation,” J. Vis. Commun. Image R., vol. 83, 2025, pp. 102–117. [10] M. Alrashoud, “Deepfake video detection methods and approaches,” J. Comput. Sci. Technol., 2025. [11] A. Heidari, “Deepfake detection using deep learning methods: A review,” WIREs Data Mining Knowl. Discov., 2024. [12] B. C. Soundarya, “Deepfake detection: Critical review of state-of-the-art techniques,” SN Appl. Sci., 2026. [13] D. L. T. Bale, L. C. Ochei, and C. Ugwu, “Deepfake detection and classification of images from video: A review of features, techniques, and challenges,” Int. J. Intell. Inf. Syst., vol. 13, no. 2, pp. 20–28, 2024. [14] OpenCV, “Open Source Computer Vision Library,” 2023. [Online].

Copyright

Copyright © 2026 B. Laxmi Narayana, Sk. SaiBabu, P. Chandrasekhar, V. Yohan, R. Vijaykumar, R. William Carey. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Download Paper

Paper Id : IJRASET78123

Publish Date : 2026-03-10

ISSN : 2321-9653

Publisher Name : IJRASET

DOI Link : Click Here