Training-free multi-bit image watermark resists removal and forgery
Original title: Attack-Resilient Multi-Bit Image Watermarking Framework via Latent Coordinate Sign Clamping
Authors report a scalable 32–128-bit payload, a 94.6% decode rate against CtrlRegen+, and 98.8% forgery detection.
A cs.CR paper proposes a training-free watermark that embeds multi-bit payloads in the latent space of diffusion models for existing natural and AI-generated images. Decoding combines magnitude-weighted majority voting with BCH and Chase decoding. Defenses include a steganalysis-resistant noise camouflage layer and a dual-layer forgery check using perceptual hashing plus a Reed-Solomon fuzzy commitment over CLIP features.
Unconfirmed: Decode and detection rates are the authors' own empirical results.
- The stated payload capacity scales from 32 to 128 bits.
- Decode rate is 94.6% against CtrlRegen+ regeneration attacks.
- Joint CLIP and Reed-Solomon checking reaches 98.8% forgery-detection accuracy.
- The method is training-free and applies to images that already exist.
From the original article
1 min read at arxiv.orgWatermarking is essential for image copyright protection and attribution, yet existing methods suffer from fundamental trade-offs among post-hoc applicability to pre-existing images, multi-bit capacity, training-free deployment, and adversarial resilience.
By Marjan Beheshti · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.