SfP-underwater: Attention-based shape from polarization for underwater scattering environments

Abstract

The development of underwater optical 3D imaging technology faces significant challenges due to the degradation of image quality caused by water scattering effects. Polarization imaging offers a promising solution for underwater optical 3D imaging by suppressing water scattering and estimating surface normals. We propose an attention-mechanism-based method for shape from polarization in underwater scattering environments (SfP-Underwater). Our approach first leverages a dual-pooling Swin transformer (DPSFormer) encoder to efficiently extract features, followed by a parallel spatial and channel attention fusion module for deep feature fusion. Furthermore, we created an SfP-Underwater dataset with multiple targets in different underwater scattering environments. Experimental results demonstrate that our method outperforms state-of-the-art SfP approaches in both texture detail reconstruction and surface normal estimation accuracy. Notably, our method exhibits robust performance across different scattering media, providing a novel and efficient solution for underwater 3D reconstruction.