ML Security PapersUncolorable Examples: Preventing Unauthorized AI Colorization via Perception-Aware Chroma-Restrictive Perturbation
Yuki Nii 1, Futa Waseda 1, Ching-Chun Chang 2, Isao Echizen 1,2
Published on arXiv
2510.08979
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
PAChroma significantly degrades colorization quality across multiple state-of-the-art models (DeOldify, DDColor, ACDO, MC-V2) while maintaining imperceptible visual changes to the original grayscale image.
PAChroma (Perception-Aware Chroma-Restrictive Perturbation)
Novel technique introduced
AI-based colorization has shown remarkable capability in generating realistic color images from grayscale inputs. However, it poses risks of copyright infringement -- for example, the unauthorized colorization and resale of monochrome manga and films. Despite these concerns, no effective method currently exists to prevent such misuse. To address this, we introduce the first defensive paradigm, Uncolorable Examples, which embed imperceptible perturbations into grayscale images to invalidate unauthorized colorization. To ensure real-world applicability, we establish four criteria: effectiveness, imperceptibility, transferability, and robustness. Our method, Perception-Aware Chroma-Restrictive Perturbation (PAChroma), generates Uncolorable Examples that meet these four criteria by optimizing imperceptible perturbations with a Laplacian filter to preserve perceptual quality, and applying diverse input transformations during optimization to enhance transferability across models and robustness against common post-processing (e.g., compression). Experiments on ImageNet and Danbooru datasets demonstrate that PAChroma effectively degrades colorization quality while maintaining the visual appearance. This work marks the first step toward protecting visual content from illegitimate AI colorization, paving the way for copyright-aware defenses in generative media.
Key Contributions
- First defensive paradigm ('Uncolorable Examples') against unauthorized AI colorization via imperceptible adversarial perturbations embedded in grayscale images
- PAChroma method using Laplacian filtering to concentrate perturbations on high-frequency regions and diverse input transformations during optimization to improve cross-model transferability and compression robustness
- Four-criteria framework (effectiveness, imperceptibility, transferability, robustness) for evaluating practical colorization defenses, validated on ImageNet and Danbooru datasets
🛡️ Threat Analysis
PAChroma creates imperceptible perturbations that prevent AI colorization models from producing valid color outputs — this is a content protection/output integrity defense analogous to anti-deepfake perturbations and style-transfer protections explicitly listed under ML09. The goal is protecting content from unauthorized AI-generated output, not adversarial misclassification for its own sake.