Polarization-Aware Low-Light Image Enhancement

Polarization-based vision algorithms have found uses in various applications since polarization provides additional physical constraints. However, low-light conditions, their performance would be severely degenerated the captured polarized images could noisy, leading to noticeable degradation degree of (DoP) and angle (AoP). Existing image enhancement methods cannot handle well they operate intensity domain, without effectively exploiting information provided by polarization. In this paper, we propose a Stokes-domain pipeline along with dual-branch neural network problem polarization-aware manner. Two application scenarios (reflection removal shape from polarization) are presented show how our can improve results.