As an important carrier and presentation form of underwater information; underwater images play an irreplaceable role in underwater environment detection. However; due to the impact of the ocean; underwater images often have quality degradation such as low resolution; blurred details; color distortion; and poor clarity. In response to the above problems; we designed a model based on Generative Adversarial Networks; which fuse underwater white balance images to achieve color correction and super-resolution of underwater images. We have conducted qualitative and quantitative experiments compared with other methods and proved that the method proposed in this article is superior to other methods in terms of visual effects.


The total ocean area is about 360 million square kilometers; accounting for about 71% of the earth's surface area; and it contains rich mineral and biological resources. With the continuous development of technology; the exploration of underwater resources and the construction of underwater facilities have become more frequent. Underwater robots use the "vision" system to obtain underwater information; and make final decisions through comprehensive analysis and judgment of the information. As one of the information acquired by the robot; underwater image plays a key guiding role. The quality of the images has a significant impact on underwater operations. However; harsh underwater environment often leads to poor image quality. First of all; sea water absorbs natural light. Red light disappears at about 5 meters underwater. As the depth increases; orange light and yellow light are successively absorbed; which will cause the underwater image to appear green. When the green light disappears after diving for about 30 meters; underwater image appears blue. Secondly; underwater suspended particles will also scatter the light; making the image captured by the camera appear foggy. The seabed terrain is also a key factor affecting the image quality. When the underwater robot is difficult to shoot the target at close range; the resolution of the image region of interest will be low; making it difficult for the robot to make accurate judgments. Therefore; in order to solve the above problems; we need to enhance and super-resolution underwater images. The real underwater image is shown in Fig. 1.

This content is only available via PDF.
You can access this article if you purchase or spend a download.