Moiré Artifacts Removal in Screen-shot Images via Multiple Domain Learning

We propose a deep learning-based moiré artifacts removal algorithm for screen-shot images using multiple domain learning. First, we develop the pixel and discrete cosine transform (DCT) networks to estimate clean preliminary images by exploiting complementary information of the moiré artifacts in different domains. Next, we develop a clean edge predictor to estimate a clean edge map for the input moiré image. Then, we propose the refinement network to further improve the quality of the pixel and DCT outputs using the estimated edge map as the guide information and to merge the two refined results to provide the final result. Experimental results on a public dataset show that the proposed algorithm outperforms conventional algorithms in quantitative and qualitative comparison.