Transformer-based underwater polarization descattering in natural illumination

Xueyan Ding, Jiaxin Wang, Yafei Wang, Bingbing Zhang

Abstract: Light propagating through water is scattered and absorbed by particles, resulting in low contrast and color distortion. Polarization imaging utilizes the polarization property of light, which can effectively reduce the effect of scattered light and thus improve the quality of underwater images. In this paper, we propose a transformer-based underwater polarization descattering method. This method introduces a multi-branch transformer block to enhance the texture information in images through efficient feature extraction and fusion, effectively suppressing the background scattered light and generating clear underwater images. In addition, we constructed a polarimetric imaging dataset comprising 1580 polarimetric image pairs captured at four polarization angles ($0^\circ$, $45^\circ$, $90^\circ$, and $135^\circ$) under various water turbidity conditions. A variety of objects of different materials are used for data acquisition to ensure the diversity and representativeness of the dataset. The data collection considers different water quality conditions from clear to highly turbid, which helps evaluate the performance of the proposed method in various complex environments. Extensive experiments demonstrate that our method achieves superior performance in terms of both visual quality and quantitative metrics compared to existing methods.