DPF-Net: Physical Imaging Model Embedded Data-Driven Underwater Image Enhancement

TL;DR

Underwater images suffer from color distortion and blur due to absorption and scattering. The authors propose DPF-Net, a two-stage framework that fuses a physical underwater imaging model with data-driven enhancement by embedding predicted physical parameters into a neural embedding space and enforcing physics-based supervision. A pretrained Degraded Parameters Estimation Module (DPEM) provides physically consistent parameters, while a CT-UNet with a Transformer-based LGFI and a PFGM leverages both data-driven features and physical cues, guided by degradation consistency, regionally weighted reference, and weakly supervised Lab losses. Experiments on UIEB, SUIM-EQi2022SGUIE, and RUIE demonstrate state-of-the-art performance and strong generalization, with ablations confirming the effectiveness of the two-stage training and loss design. The work highlights a practical path toward reliable, physics-informed UIE with robust color and detail restoration for real-world underwater imaging tasks.

Abstract

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 first train a physical parameter estimate module using synthetic datasets to guarantee the trustworthiness of the physical parameters, rather than solely learning the fitting relationship between raw and reference images by the application of the imaging equation, as is common in prior studies. 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. To maintain the uniformity of the restoration process amid underwater imaging degradation, we propose a physics-based degradation consistency loss. Additionally, we suggest an innovative weak reference loss term utilizing the entire dataset, which alleviates our model's reliance 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.

Figures (16)

• Figure 1: Schematic diagram of underwater imaging process.
• Figure 2: The framework of the proposed DPF-Net. The two branches, CNN-Transformer UNet (CT-UNet) and Physical Feature Generation Module (PFGM), receiving physical parameters input from the pre-trained Degraded Parameters Estimation Module (DPEM) into embedding space. The final output image is obtained after the output features from the two branches undergo the Feature Fusion module. The network loss includes reference loss, degradation consistency loss, and weak reference loss based on the Lab color space.
• Figure 3: The architecture and training methodology of the proposed DPEM. Underwater images are synthesized from terrestrial RGB-D images by configuring various combinations of degradation parameters within the underwater physical imaging model. Then, DPEM is trained to predict these degradation parameters using the synthesized images as ground truth.
• Figure 4: Raw image (left) and corresponding relative depth map (right).
• Figure 5: Different synthetic degradation images of one in NYU-Depth-V2 silberman2012indoor.