New ADR network enhances underwater images via joint dehazing and Retinex

Underwater images suffer from severe degradation due to wavelength-dependent light absorption, scattering, and turbidity from suspended particles. Classical image formation models (IFM) fail to capture nonlinear underwater light behavior, while purely data-driven methods lack physical interpretability. To address this, a team led by Sahana Ray has developed ADR, a novel three-stage network that extends the underwater image formation model with additional terms to perform joint dehazing and Retinex-based enhancement. The pipeline first dehazes the image using an improved physics-based model, then applies Retinex theory to correct illumination and color, and finally refines the output through an attention-enhanced U-Net++ architecture.

Experiments on the UIEB and UFO-120 benchmark datasets show that ADR yields competitive or superior results compared to existing state-of-the-art methods, both in terms of quantitative metrics and visual quality. The paper, accepted for the IEEE ICIP 2026 conference, provides a balanced approach that marries physical modeling with deep learning. For professionals working with autonomous underwater vehicles (AUVs), marine biology, archaeology, or offshore infrastructure inspection, ADR offers a practical tool for recovering clear, color-accurate images from challenging underwater environments.