Abstract
When smartphone cameras are used to take photos of digital screens, usuallymoire patterns result, severely degrading photo quality. In this paper, wedesign a wavelet-based dual-branch network (WDNet) with a spatial attentionmechanism for image demoireing. Existing image restoration methods working inthe RGB domain have difficulty in distinguishing moire patterns from true scenetexture. Unlike these methods, our network removes moire patterns in thewavelet domain to separate the frequencies of moire patterns from the imagecontent. The network combines dense convolution modules and dilated convolutionmodules supporting large receptive fields. Extensive experiments demonstratethe effectiveness of our method, and we further show that WDNet generalizes toremoving moire artifacts on non-screen images. Although designed for imagedemoireing, WDNet has been applied to two other low-levelvision tasks,outperforming state-of-the-art image deraining and derain-drop methods on theRain100h and Raindrop800 data sets, respectively.