LiBRA: Detection-Aware Image Watermark Removal via Bidirectional Latent Optimization

2026-10-02

AI Summary

Existing AI-based image watermark removal methods are prone to watermark inversion residuals or excessive degradation of visual quality. To address these limitations, this work proposes LiBRA, a method that performs bidirectional optimization guidance in the latent space to drive decoder confidence toward random guessing, thereby mitigating inversion risks. Additionally, frequency masking is introduced to constrain perturbation regions and preserve generated image quality. Built upon an autoencoder architecture, LiBRA employs binomial testing for rigorous statistical validation. Experimental results demonstrate that LiBRA achieves covert and undetectable watermark removal while maintaining high visual fidelity, offering a new paradigm for evaluating watermark robustness.

Abstract

Digital watermarking supports source attribution for AI-generated images, but its reliability depends on resistance to removal attacks. Some attacks attempt to remove watermarks by forcing the decoded watermark to differ from the original. However, this can produce an inverted watermark that remains detectable, causing removal to fail, while further attempts to alter the watermark may unnecessarily degrade image quality. To address these limitations, we present LiBRA (Latent In-band Bidirectional Removal Attack), which aims to make watermarks undetectable while preserving image quality. Instead of continually pushing the watermark toward inversion, LiBRA adjusts the image to conceal the watermark without encouraging further changes that could degrade image quality. Some attacks keep pushing decoded bits away from the original watermark, even when further changes preserve detectability and damage image quality. With access to the watermark key and decoder, LiBRA makes bounded changes in a public autoencoder's latent space. Unlike inversion-driven objectives that cannot correct excessive inversion, LiBRA guides average decoding confidence toward random guessing from either direction. This helps avoid an inverted but detectable watermark. Leaving individual bits flexible allows image-quality constraints to favor less damaging changes, while an optional frequency-guided mask limits their location. We verify removal using an exact two-sided binomial test rather than assuming the confidence target guarantees success.

Problem

• watermark removal
• digital watermarking
• image quality
• detection evasion
• AI-generated images

Innovation

• Watermark Removal
• Bidirectional Latent Optimization
• Detection-Aware Attack
• Autoencoder Latent Space
• Binomial Test