BA-GAN: Block Attention GAN model for Underwater Image Enhancement
Yue Wang, Huijie Fan, Shiben Liu, Jiaxin Liu, Sun Li, Yandong Tang · 2021 IEEE International Conference on Unmanned Systems (ICUS) · 2021
Since underwater images are an important route to underwater information, underwater image enhancement becomes an important fundamental study to underwater vision tasks. Regarding problems with underwater degraded images, such as color distortion causing image rendering blue or green, blur and low contrast leading to image loss of detail information, we propose a Generative Adversarial Network (GAN) based model that adds an attention mechanism to enhance underwater images. The model finally achieves the purpose of generating clear underwater images by learning the relationship between underwater degraded images and ground truth values, removing color interference in the image and enhancing the details of the image. We compared the proposed model with the the state-of-the-art methods on both EUVP and Underwater Image Enhancement Benchmark Dataset, experimental results on both datasets prove that our method performs more stable than state-of-the-art methods.