Underwater image enhancement method works on low power devices
Physics-Guided Spectral Distillation for Underwater Image Enhancement on Resource-Constrained Devices
Computer Vision and Pattern Recognition
Summary
Seeing clearly underwater is hard because water changes how things look in photos. The authors developed a new method that teaches smaller, faster computer models to improve underwater pictures by learning from bigger models and physics knowledge about light in water. This method keeps the important color and detail while being quick enough to run on small underwater robots. They showed their method works well on multiple test sets and helps underwater robots spot objects better in real life.
What this means in practice
- •For marine robotics teams: Enhance underwater camera images in real time on robots with limited computing power to improve navigation and perception.
- •For autonomous vehicle developers: Improve visual input quality for underwater object detection systems where computational resources are constrained.
Authors
Yifan Chen, Kai He, Ye Zheng, Jijun Lu, Zhe Sun, Tao Chen
Abstract
Underwater image enhancement is crucial for improving visual perception in marine applications. Existing underwater image enhancement studies mainly focus on enhancement quality and visual fidelity, while rarely considering real-time deployment capability, which is essential for resource-constrained underwater robots. To this end, we introduce a physics-guided spectral distillation (PSD) method, which reduces model capacity for real-time applications while maintaining the high performance of underwater image enhancement models. To decompose the outputs of teacher and student models, PSD adopts a multilevel Haar discrete wavelet transform. It transfers low-frequency color and illumination information as well as high-frequency structural details through band-specific objectives. Moreover, the distillation process of PSD is degradation-aware. We estimate degradation-aware weights through a physical head and combine them with ground-truth-guided reliability masks to selectively retain valuable teacher guidance. Experiments on the UIEB, LSUI, and EUVP datasets validate the effectiveness of the proposed method. Furthermore, we demonstrate the benefits of enhanced images for downstream perception tasks, including object detection. Deployment on a self-developed ROV further demonstrates its practical applicability in real-world underwater scenarios.