A LUNet based on large kernel attention mechanism for image denoising
Yonggang Yao · International Conference on Electronic Information Technology (EIT 2022) · 2022
U-shaped networks are widely used in the field of image denoising with their multiscale and jump connection structures in recent years. The feature extraction structures mainly used in previous works are convolutional neural networks (CNNs), but their ability to extract information at a distance is poor. In this paper, we propose a U-shaped network structure combined with large kernel attention structure for image denoising, in which the CNNs structure can effectively extract local information while the large kernel attention structure has better extraction of global information and lower computational cost compared with Transformer, through which local-global features are learned on the feature maps of the input noisy images at different scales, which can effectively enhance the performance of the network and improve the image denoising task. The performance of the network can be enhanced to improve the performance of the image denoising task.