A Single Image Deraining Network Based on Global Feature Perception
Bo Fu, Hongguang Wang · 2022
Rainy weather can degrade the quality of image captured outdoors, which in turn reduces the effectiveness of the subsequent computer vision algorithm. Therefore, as a way to improve the performance of subsequent visual tasks, single image deraining has become an important research topic. In this paper, a single image deraining method based on global feature perception is proposed, which can effectively locate the rain streak and estimate corresponding background content via Swin Transformer mechanism. Different from convolutional neural network and traditional Transformer, the proposed network can extract the global features of an image while reducing the amount of computation. Combing advantages of Swin Transformer and convolutional network, a three-stage deraining network is proposed. First, the coarse features of an image are extracted through the convolutional layer. Then, using the feature refining module based on Swin Transformer, global features are extracted and fused. Finally, through the image reconstruction module, the clean image is generated by using residual learning. The experimental results show the superiority of the proposed network.