DPF-Net: Physical imaging model embedded data-driven underwater image enhancement

Due to the complex interplay of light absorption and scattering in the underwater environment, underwater images experience significant degradation. This research presents a two-stage underwater image enhancement network called the Data-Driven and Physical Parameters Fusion Network (DPF-Net), which harnesses the robustness of physical imaging models alongside the generality and efficiency of data-driven methods. We train the Degraded Parameters Estimation Module (DPEM) on synthetic datasets with preset physical parameters as ground truth. This approach learns more authentic underwater imaging model, contrasting with prior works that directly fit raw-to-reference image mappings through the imaging equation. This module is subsequently trained in conjunction with an enhancement network, where the estimated physical parameters are integrated into a data-driven model within the embedding space. During model training, in addition to traditional reference-based losses, we use a degradation consistency loss to ensure physical consistency. Furthermore, we propose a new weak reference loss term that leverages the color distribution of the entire training set, thereby alleviating the reliance of our model on the quality of individual reference images. Our proposed DPF-Net demonstrates superior performance compared to other benchmark methods across multiple test sets, achieving state-of-the-art results. The source code and pre-trained models are available on the project home page: https://github.com/OUCVisionGroup/DPF-Net.