Underwater Image Enhancement Algorithm Based on Transformer and Convolutional Network

Underwater images suffer from color distortion, key information loss and blurred details, and it is difficult to reconstruct high-quality images. Therefore, we propose a U-shaped codec network that integrates Transformer and convolution for underwater image enhancement. First, we scale the degradation image to the size of 256$\times$ 256$\times$ 3 and use it as input. Next, we propose the improved encoder and decoder, the feature extraction encoding module that mixes local-ViT and global-ViT is constructed and the convolutional decoding module incorporating channel attention is designed. Finally, extensive experiments reveal that our algorithm can effectively achieve color recovery and detail enhancement of underwater images.