F-GAN: Real-Time Color Correction Model of Underwater Images

In recent years, color correction of underwater images is becoming a hot research field due to the wide application of Autonomous Underwater Vehicles and Remotely Operated Vehicles in monitoring marine environments. Many kinds of deep learning models have been presented by researchers to correct the color of the underwater images to facilitate the work of both people and machines. However, these models emphasize the correction effect more, and they can't be applied to real-time scenarios, especially for high-resolution images (480∗640). Therefore, we propose a more concise generative adversarial network-based model to achieve real-time color correction of underwater images. To supervise the training process better, we design a multiterm loss function including adversarial loss, content loss, and structural similarity loss to estimate the generated image quality. Experiments on the EUVP (Enhancing Underwater Visual Perception) dataset show that our approach produces visually satisfactory results for high-resolution images under the real-time requirement. Besides, our method can even outperform some state-of-the-art models.