UMGAN: Underwater Image Enhancement Network for Unpaired Image-to-Image Translation

Due to light absorption and scattering underwater images suffer from low contrast, color distortion, blurred details, uneven illumination, which affect vision tasks research. Therefore, image enhancement is of great significance in applications. In contrast existing methods for specific environments or reliance on paired datasets, this study proposes an multiscene generative adversarial network (UMGAN) enhance images. The implements unpaired image-to-image translation between the turbid domain clear domain. It has a impact several types. Feedback mechanisms noise reduction are designed optimize generator address issue artifacts GAN-produced Furthermore, global–local discriminator employed improve overall while adaptively modifying local region effect. resolves over- underenhancement regions. training data eliminated through cycle consistency structure. UMGAN performs satisfactorily various types when compared quantitatively qualitatively other state-of-the-art algorithms. strong robustness can be applied different scenes.