Active Color-Aware Demosaicing for Color Polarization Images

Color division-of-focal-plane (DoFP) polarization cameras can capture polarization images that reveal both scene texture and physical properties in a single shot, driving numerous recent advancements in computer vision. However, due to the bilayer array, the system sacrifices resolution in exchange for higher data capture efficiency, leading to severe degradation of polarization-related parameters. Existing methods lack explicit consideration of the correlations between different colors under the same polarization angle, resulting in the ineffective utilization of color information. In this paper, we propose a one-stage color-aware polarization demosaicing framework, along with a neural network that adaptively fuses features from different color channels in a prioritized, complementary manner. Experimental results show that our framework not only improves the quality of polarized intensity images but also achieves state-of-the-art accuracy in reconstructing polarization-related parameters.