Moiré Pattern Removal with a Generative Adversarial Network

Abstract

Moiré patterns can be seen in camera-captured digital screen photos due to the interference between the pixel grids of the camera sensor and the pixel grids of the digital screen. It severely degrades the quality of the photos. With the rapid development of personal devices, people are using digital camera to take photos more and more often. Among them, it's very common to see camera-captured screen photos, so the work of Moiré pattern removal is very meaningful for improving user experience. In this paper, we introduce a novel method of Moiré pattern removal based on the Generative Adversarial Network (GAN). To train our model, we built a dataset of paired Ground and Moiré images, which has 16,500 images totally. Experiments showed that, given Moiré images as the input, the trained generator of our GAN nets can produce Moiré-free images of high quality.