Problem
Research questions and friction points this paper is trying to address.
frequency-domain watermarking
Latent Diffusion Models
copyright infringement
message capacity
geometric patterns
DeepFreqMark: End-To-End Learnable Frequency-Domain Watermarking with Spherical Attack Simulation for Latent Diffusion Models
🤖 AI Summary
This work addresses the challenges of copyright protection and forgery detection in images generated by latent diffusion models, where existing frequency-domain watermarking methods suffer from limited payload capacity and inflexible pattern design. To overcome these limitations, we propose the first end-to-end learnable frequency-domain watermarking framework, which introduces neural encoders and decoders into the latent space to replace handcrafted watermark designs. By incorporating spherical linear interpolation (Slerp) to simulate realistic attacks, our approach preserves the characteristics of Gaussian perturbations while circumventing the computational bottleneck associated with DDIM inversion. The proposed method supports message payloads of up to 256 bits and achieves significantly lower bit error rates under real-world attacks, outperforming current state-of-the-art baselines.
📝 Abstract
The proliferation of AI-generated images produced by Latent Diffusion Models (LDMs) has raised critical concerns regarding copyright infringement and misinformation. Although existing frequency-domain watermarking methods embed handcrafted geometric patterns into the initial latent noise prior to generation, they suffer from limited capacity and rigid pattern designs. We propose DeepFreqMark, an end-to-end learnable frequency-domain watermarking framework that replaces manual pattern engineering with a neural message encoder and decoder. To circumvent the computational bottleneck caused by Denoising Diffusion Implicit Model (DDIM) inversion during training, we introduce a Spherical Linear Interpolation (Slerp)-based attack simulation. This approach operates directly on the noise latent while strictly preserving the Gaussian variance. Extensive experiments demonstrate that DeepFreqMark achieves significantly lower Bit Error Rates (BER) than baseline methods under real-world attacks and scales to 256 bits message capacity. Our source code is available at https://github.com/chenhsiu48/DeepFreqMark.
Innovation
Methods, ideas, or system contributions that make the work stand out.
frequency-domain watermarking
latent diffusion models
end-to-end learnable
spherical attack simulation
neural message encoding
🔎 Similar Papers
No similar papers found.
