INVISIBLE WATERMARKING FOR AI-GENERATED IMAGES: A COMPARATIVE STUDY OF SPATIAL AND TRANSFORM DOMAIN TECHNIQUES

Invisible watermarking provides an effective mechanism for identifying and authenticating AI-generated images without introducing visible artifacts. Spatial-domain techniques such as LSB are simple to implement and offer high payload capacity, but they are generally vulnerable to compression, noise, and image processing attacks. In contrast, transform-domain techniques such as DCT and DWT offer greater robustness and are better suited for real-world deployment where images may be modified, compressed, or redistributed. This comparative study shows that transform-domain watermarking is a stronger choice when robustness and reliability are priorities, while spatial-domain methods remain useful for lightweight or low-complexity applications. As AI-generated media continues to improve in realism, robust watermarking will remain an important tool for authenticity verification, intellectual property protection, and responsible AI content governance [4,3].

[1]    Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio, “Generative adversarial networks,” Communications of the ACM, vol. 63, no. 11, pp. 139–144, 2020. [2]    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High-resolution image synthesis with latent diffusion models,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 10684–10695, 2022. [3]    S. J. Nightingale and H. Farid, “AI-synthesized faces are indistinguishable from real faces and more trustworthy,” Proceedings of the National Academy of Sciences, vol. 119, no. 8, e2120481119, 2022. [4]    L. Verdoliva, “Media forensics and deepfakes: An overview,” IEEE Journal of Selected Topics in Signal Processing, vol. 14, no. 5, pp. 910–932, 2020. [5]    J. Zhu, R. Kaplan, J. Johnson, and L. Fei-Fei, “HiDDeN: Hiding data with deep networks,” in Proceedings of the European Conference on Computer Vision (ECCV), pp. 657–672, 2018. [6]    Cox, M. Miller, J. Bloom, J. Fridrich, and T. Kalker, Digital Watermarking and Steganography, 2nd ed. Morgan Kaufmann, 2007. [7]    S. Katzenbeisser and F. A. P. Petitcolas, Information Hiding: Techniques for Steganography and Digital Watermarking. Artech House, 2000. [8]    R. Panigrahi and N. Padhy, “An effective steganographic technique for hiding the image data using the LSB technique,” Cyber Security and Applications, vol. 3, Art. no. 100069, 2025. [9]    N. Saikrishna and M. G. Resmipriya, “An Invisible Logo Watermarking using Arnold Transform,” Procedia Computer Science, vol. 93, pp. 808–815, 2016. [10]                             Y. Wang, Y. Luo, Z. Wang, and H. Pan, “A Hidden DCT-Based Invisible Watermarking Method for Low-Cost Hardware Implementations,” Electronics, vol. 10, no. 12, 2021. [11]                             M. Chandra, “A DFT-DWT Domain Invisible Blind Watermarking Technique for Copyright Protection of Digital Images,” in Advances in Wavelet Theory and Their Applications in Engineering, Physics and Technology, D. Baleanu, Ed. InTech, 2012. [12]                             P. C. Mandal, I. Mukherjee, G. Paul, and B. N. Chatterji, “Digital image steganography: A literature survey,” Information Sciences, vol. 609, pp. 1451–1488, Sep. 2022. [13]                             H.-W. Tseng and H.-S. Leng, “A reversible modified least significant bit (LSB) matching revisited method,” Journal of Information Hiding and Multimedia Signal Processing, vol. 5, no. 1, pp. 1–8, 2009. a.       Ramesh, P. Dhariwal, A. Nichol, C. Chu, and M. Chen, “Hierarchical text-conditional image generation with CLIP latents,” arXiv preprint arXiv:2204.06125, 2022. [14]                             K. Ghazanfari, S. Ghaemmaghami, and S. R. Khosravi, “LSB++: An improvement to LSB+ steganography,” in TENCON 2011—2011 IEEE Region 10 Conference, IEEE, 2011.