Semantic Color Naturalness Breaker: Preventing Illegitimate Colorization via Content-Aware Color Priors

Abstract

We address unauthorized colorization of grayscale media by adding imperceptible perturbations to released grayscale images to degrade unauthorized colorization while preserving visual quality. We introduce a semantic-level approach guided by Content-aware Color Distributional Distance (CaCDD), a metric that measures color plausibility from content-aware color priors. Experiments on ImageNet show the method remains effective under modest perturbation constraints and common post-processing.

Futa Waseda

Project Assistant Professor