RollingPol: Single-Shot Polarization Imaging for Underwater Sensing

Abstract

Underwater optical sensing is severely degraded by backscattered light from suspended particles. While polarization-based image enhancement methods can suppress this backscattering effect, existing methods are either costly, or require sequential image captures that do not handle dynamic scenes. Recently, learning-based enhancement methods have shown promising results, but they depend on large paired datasets that are difficult to obtain underwater and introduce noise or artifacts due to domain gaps. To address these limitations, we present RollingPol, a low-cost, portable hardware add-on for any commodity rolling-shutter camera to enable polarization imaging with a single image capture. It employs stacked liquid crystals to introduce rapid, continuous 0°-180° polarization sweep while pixels are sequentially scanned by a rolling-shutter camera. As such, it encodes multiple polarization states within a single image frame, enabling fast acquisition even in dynamic environments. A neural network is then built to reconstruct images of original resolution at any polarization states, which can be aggregated for polarization-based image enhancement. Experiments across diverse water conditions show that RollingPol improves image clarity by 58%. It reduces noise by 73% compared to raw images and by 86% relative to the best learning-based enhancement methods. Furthermore, using RollingPol output as a prior further boosts pre-trained learning-based enhancement models by up to 21%, demonstrating that lightweight polarization cues can significantly improve underwater imaging.