Image dehazing network based on improved CycleGAN

In view of the blur and color distortion in traditional CycleGAN image dehazing, animproved CycleGANImage dehazing network is proposed. The generator of present CycleGAN consists three parts: feature extraction sub-network, feature fusion sub-network and image restoration sub-network. The image feature extraction sub network is used to extract the content features and style features of the image.The feature fusion sub network uses two different attention mechanisms to fuse the extracted content features and style features. The image complex atom network restores the fused image features to a haze free image.Compared with traditional CycleGAN and existing defogging networks, the proposed network can achieve ideal dehazing results for both synthetic images and real images, and effectively solves the problems of blurring and color distortion in traditional CycleGAN images after dehazing.