Boosting Illuminant Estimation in Deep Color Constancy through Enhancing Brightness Robustness

TL;DR

This work identifies brightness vulnerability as a crucial robustness concern in DNN-driven color constancy (DNNCC) and introduces BRE, a hyperparameter-free plug-in that combines adaptive adversarial brightness augmentation with brightness-robustness-aware optimization. BRE uses a differentiable, parameterized brightness curve (with $L=32$) to generate high-risk brightness variations and trains models to extract brightness-invariant features via adversarial training and a brightness-aware contrastive loss. Evaluations on ColorChecker and Cube+ show BRE consistently improves illuminant estimation across multiple DNNCC baselines, with notable reductions in angular error and improved robustness to brightness changes, all without additional testing overhead. The results underscore the value of explicitly addressing brightness variation in color constancy and offer a practical pathway to more robust DNNCC systems.

Abstract

Color constancy estimates illuminant chromaticity to correct color-biased images. Recently, Deep Neural Network-driven Color Constancy (DNNCC) models have made substantial advancements. Nevertheless, the potential risks in DNNCC due to the vulnerability of deep neural networks have not yet been explored. In this paper, we conduct the first investigation into the impact of a key factor in color constancy-brightness-on DNNCC from a robustness perspective. Our evaluation reveals that several mainstream DNNCC models exhibit high sensitivity to brightness despite their focus on chromaticity estimation. This sheds light on a potential limitation of existing DNNCC models: their sensitivity to brightness may hinder performance given the widespread brightness variations in real-world datasets. From the insights of our analysis, we propose a simple yet effective brightness robustness enhancement strategy for DNNCC models, termed BRE. The core of BRE is built upon the adaptive step-size adversarial brightness augmentation technique, which identifies high-risk brightness variation and generates augmented images via explicit brightness adjustment. Subsequently, BRE develops a brightness-robustness-aware model optimization strategy that integrates adversarial brightness training and brightness contrastive loss, significantly bolstering the brightness robustness of DNNCC models. BRE is hyperparameter-free and can be integrated into existing DNNCC models, without incurring additional overhead during the testing phase. Experiments on two public color constancy datasets-ColorChecker and Cube+-demonstrate that the proposed BRE consistently enhances the illuminant estimation performance of existing DNNCC models, reducing the estimation error by an average of 5.04% across six mainstream DNNCC models, underscoring the critical role of enhancing brightness robustness in these models.

Figures (7)

• Figure 1: The DNNCC model is sensitive to brightness variation. On the left, two light sources (A and B) with identical chromaticity but different positions create varying brightness, leading to inconsistent chromaticity estimates. This suggests that brightness variation could be a potential cause of degradation in the DNNCC model. On the right, we propose a Brightness Robustness Enhancement (BRE) strategy, which mitigates these effects by adversarially augmenting high-risk brightness images and incorporating robustness-aware optimization during training, thereby significantly reducing illuminant estimation errors across multiple datasets.
• Figure 2: Visualization of the synthetic dataset. The colored bar on the left shows illuminant labels, with arrows and coordinates above the image indicating the illumination direction and world coordinate position. Red and blue represent the training and test sets, respectively.
• Figure 3: Visualization of brightness variation impact on the FC4 model. The estimated illuminant and error are shown on the right of the fourth and fifth column images, respectively.
• Figure 4: An overview of the proposed brightness robustness improvement framework
• Figure 5: Visual comparison of DNNCC models with angular errors in the lower right corner.