Physics-inspired Learning for Structure-Aware Texture-Sensitive Underwater Image Enhancement

Xinwei Xue (Dalian University of Technology)*; Zexuan Li (Dalian University of Technology ); Long Ma (Dalian University of Technology); Risheng Liu (Dalian University of Technology); Xin Fan (Dalian University of Technology)

---

PMLR Page

Abstract

Recently, improving the visual quality of an underwater image by deep learning-based methods has drawn much attention. Unfortunately, in real-world environments, diverse environmental factors (e.g., blue/green color distortion) heavily limit their performance. In other words, strengthening the robustness of the underwater image enhancement method is the critical point. In this paper, we devote ourselves to develop a new architecture with strong robustness and adaptation. Concretely, inspired by the underwater imaging principle, we establish a new physics-inspired learning model that easy to realize. A Structure-Aware Texture-Sensitive Network (SATS-Net) is further developed to portray the model. In which, the structure-aware module is responsible for structural information, and the texture-sensitive module is in charge of textural information. In this way, SATS-Net successfully incorporates robust characterization absorbed from the physical principle to achieve strong robustness and adaptation. We conduct extensive experiments to verify that SATS-Net is superior to existing advanced techniques in various real-world underwater environments.