Multi-Purpose Oriented Real-World Underwater Image Enhancement

Images captured underwater usually suffer from weak illumination, color cast, fuzz and noise, which severely degrade the visibility. Numerous methods have been proposed to improve the quality of underwater images, but rarely of them can give a comprehensive consideration to all these problems, which makes them hard to adapt for various and complex real-world underwater scenes. Herein, a novel multi-purpose oriented approach for real-world underwater image enhancement is proposed. To manipulate different information on the corresponding layers, we firstly decompose the input image into illumination layer and reflectance layer. Subsequently, compensation of the brightness is carried out on the illumination layer, while color correction and contrast enhancement are implemented on the reflectance layer through a multi-scale processing strategy. Benefiting from this strategy, the proposed approach is provided with high control flexibility, which can significantly improve the visibility of underwater images while efficiently suppress the amplification of noise. Both qualitative and quantitative evaluations demonstrate that the proposed method has superior robustness, accuracy and effectiveness for complex marine circumstance.