🤖 AI Summary
This study addresses the ambiguity in color constancy estimation caused by chromatic cue collapse in monochromatic scenes. To overcome this, we propose the SAFE framework coupled with a learnable color space. The method achieves feature modulation and adaptive correction through a four-token structured illumination representation, scene-complexity-adaptive reweighting, and scene-dependent chromatic normalization. Experimental results demonstrate that this approach effectively resolves estimation challenges in monochromatic environments, reducing the mean angular error by 10%, the best 25% error by 20%, and the worst 25% error by 5.8%. These improvements signify a substantial enhancement in both the accuracy and robustness of color constancy estimation under challenging conditions.
📝 Abstract
Color constancy on pure-color scenes is challenging: when most pixels share a narrow band of hues, every chromaticity-based cue collapses to a single point and standard estimators become ambiguous. We propose a compact framework that couples two innovations: (i) SAFE, a Scene-Aware FeaturE modulation network that organizes illumination cues into a structured four-token representation, which is then selectively reweighted based on scene complexity features; (ii) the Learned Color Space (LCS), a scene-dependent chromaticity normalization that directly addresses the chromaticity collapse problem for pure-color scenes. Experiment results show that SAFE consistently improves performance in pure-color scenes. Compared to the best-performing baseline in each metric, it reduces the mean angular error by 10%, the best-25% error by 20%, and the worst-25% error by 5.8%.