Underwater Image Enhancement Based on Feature Fusion Generative Adversaral Networks

Aiming at the problems of low contrast and color distortion of underwater images, an underwater image enhancement algorithm based on feature fusion generative adversarial networks is proposed. Firstly, color correction algorithm is applied to the underwater degraded image, and then the feature of color corrected image is extracted by convolution neural network. Secondly, the feature of underwater degraded image can be extracted by the feature extraction network which is based on U-Net, and then fuse it with the feature of color correction image. Finally, the convolution neural network is used to reconstruct the fusion feature to the enhanced image. The experimental results on Underwater-ImageNet dataset show that the algorithm can effectively improve the contrast and clarity of underwater degraded image, and the enhanced image color is more realistic. Compared with other algorithms on the Underwater-ImageNet dataset, the underwater image quality measure (UIQE) is increased by 0.308, the natural image quality evaluator (NIQE) is reduced by 0.638. The contrast and sharpness of the enhanced underwater image are improved and the colors are more realistic.