CMFNet: An End-to-End Ultra-Lightweight Underwater Image Enhancement Network
Songxuan Li, Xiaolong Yang, Huajin Han, Qingbo Meng · 2024
As the main carrier of underwater information, underwater video images play a vital role in human exploration and development of the ocean. However, due to the optical properties of water and the complex and variable underwater environmental factors, the quality of observed underwater images is severely degraded, affecting the execution of underwater equipment tasks. Moreover, most current underwater image enhancement methods have high computational costs, which limits their application in real-time large-scale underwater image processing. In order to solve this problem, we propose the CMFNet, a new lightweight network can efficiently enhance underwater images in real time. This network processes through a color recovery module, a multi-scale feature extraction module, and a feature fusion module to improve image color representation while maintaining content style and spatial texture. To the best of our knowledge, CMFNet is the only network that can enhance 2K resolution images in real time. Extensive experiments and visual comparisons have shown that CMFNet can achieve optimal or suboptimal results with fewer parameters and computational complexity compared to other advanced underwater image enhancement methods, making it valuable for practical applications.