RGB-UV Multi-Feature Fusion Color Constancy Algorithm

Considering that image color significantly affects the accuracy of object detection in the field of machine vision, this paper proposes a branched convolutional neural network model that integrates RGB and UV dual-color domain features. First, the most effective features highly correlated with the light source are extracted separately in the RGB and UV domains, and single-color domain light source estimation is performed. Second, to address the issue of incomplete color feature representation in a single color domain, features from different color domains are fused for light source estimation. The three obtained light source estimates are then weighted and fused to achieve more stable light source prediction performance. Finally, experiments are conducted on the reprocessed ColorChecker and NUS-8 datasets. The results demonstrate that the RGB-UV multi-feature fusion color constancy algorithm achieves higher accuracy and is better suited for complex lighting scenarios.