Stokes-S₀ Prior-Guided Dual-Branch Network for Polarized Image Enhancement

Polarization imaging can simultaneously capture both intensity and polarization information of targets. However, under low-light conditions, polarized images commonly suffer from low contrast and high noise, leading to significant degradation in the Degree of Polarization (DoP) images. Conventional intensity image enhancement methods often fail to effectively restore degraded polarization information as they disregard the physical characteristics of polarization data. To overcome this limitation, we propose a dual-branch polarization image enhancement network guided by Stokes- S₀ feature priors. Leveraging the high signal-to-noise ratio (SNR) characteristic of S₀ images, our method pioneers the use of S₀ images and their edge information for DoP image restoration. The network employs spatial and channel attention mechanisms for feature selection and fusion, enabling joint enhancement of both S₀ and DoP images. Furthermore, existing polarization image enhancement methods still face the critical limitation of lacking paired normal/low-light polarization datasets. To address this fundamental challenge, we propose a low-light image generation method based on spatial and frequency domain feature analysis. We construct a paired polarization image dataset by simulating photon noise, readout noise, and illumination degradation. Both quantitative and qualitative experimental results demonstrate that the proposed method achieves significant improvements over existing enhancement approaches.