Underwater polarization imaging via Stokes anisotropic diffusion

Underwater image restoration is limited by scattering and absorption, which attenuate light intensity and alter its polarization state during propagation. Stokes vectors are widely used in underwater polarization imaging to characterize the intensity and polarization properties of the light. However, existing underwater polarization image restoration methods commonly neglect the spatial variability of the Stokes vector, which compromises the accuracy of polarization information correction and backscattered light estimation. In this paper, we propose a Stokes-vectors-based anisotropic diffusion method for underwater polarized image restoration. Specifically, Perona-Malik anisotropic diffusion is applied to the Stokes vectors. It adaptively modulates the Stokes vectors across its spatially varying regions, using gradient magnitudes from local neighborhoods. Subsequently, the diffused Stokes vectors are fused with the minimum-polarization image to estimate the backscattered light. Furthermore, to mitigate small-scale bright artifacts and localized anomalous scattering effects in the backscattered light, a morphological opening operation is also applied. The modulated backscattered light is then integrated with the Jaffe-McGlamery underwater imaging model to reconstruct high-quality restored images. Experimental results demonstrate that the proposed method effectively suppresses backscattered light under a wide range of turbidity conditions. Compared with conventional methods, this method achieves significant improvement in preserving target edge features.