Deep Generative Model for Image Inpainting With Local Binary Pattern Learning and Spatial Attention

Deep learning (DL) has demonstrated its powerful capabilities in the field of image inpainting. The DL-based inpainting approaches can produce visually plausible results, but often generate various unpleasant artifacts, especially boundary and highly textured regions. To tackle this challenge, work, we propose a new end-to-end, two-stage (coarse-to-fine) generative model through combining local binary pattern (LBP) network with an actual network. Specifically, first LBP using U-Net architecture is designed to accurately predict structural information missing region, which subsequently guides second for better filling pixels. Furthermore, improved spatial attention mechanism integrated into network, by considering consistency not only between known region generated one, also within itself. Extensive experiments on public datasets including CelebA-HQ, Places Paris StreetView demonstrate that our generates results than state-of-the-art competing algorithms, both quantitatively qualitatively. source code trained models are available at https://github.com/HighwayWu/ImageInpainting.