The unique nature of polarization information can provide a reliable physical prior for underwater multimodal image fusion. Existing methods mainly employ the integration of linearly polarized images from multiple directions, which is essentially an intensity fusion process of images. To solve this problem, we propose an underwater polarization imaging technology based on multi-polarization modality fusion. This method employs the total intensity S0 to provide the basic scene brightness, uses the degree of linear polarization (DoLP) as a physical prior, fully exploits the rich texture features in DoLP to compensate for S0, and integrates color information channels to better preserve the color characteristics of the scene. In addition, we develop a Polarization Feature Enhancement Module (PFEM) tailored for polarization data, which embeds a customized gating mechanism to select features and adaptively fuse feature vectors from different channels. Finally, we construct and publicly release an underwater polarization image dataset with multiple turbidity levels and materials, and systematically verify the robustness of the proposed method.
Qian et al. (Sun,) studied this question.