A novel underwater color correction method based on underwater imaging model and generative adversarial network

Abstract

Underwater vision has played an increasingly important role in ocean research, ocean military, and underwater fishing. The raw underwater images suffer from color distortion and poor image quality, which make underwater vision more challengeable than open air vision. To address the mentioned problems, a novel model for underwater image color correction is proposed in this paper. Firstly, a synthetic underwater image generating network is presented to overcome the lack of effective underwater training data. Specifically, it combines a background color prior model to generate synthetic underwater images via in-air images data. By the benefit of the color prior and imaging algorithm, the generating model can be with fewer number of parameters and more effective. Meanwhile, a simple yet effective model is proposed to train on the in-air images and corresponding rendered synthetic underwater images for color correction. Different from other image restoration models with multiple substructure or complicated construction, the proposed method has only few convolution layers and combining with a dense-like and a res-like structure. Finally, the enhanced results demonstrate the superiority of the proposed method, which performs favorably against the existing state of the art methods in both effectiveness and model size aspects.