An Image Inpainting Method Considering Edge Connectivity of Defects

In this paper, we propose a new image inpainting method that improves the line distortion and unnatural coloration in the restoration area observed in the conventional methods. Many deep learning inpainting methods have been proposed in recent years. However, these conventional methods have problems with distorted lines and unnatural colors in the restoration area. We solve the problems of the reconstruction of distorted lines and unnatural color simultaneously by applying edge generator from EdgeConnect to DeepFill v2. Through evaluation experiments, we show that the proposed model is better than the conventional methods in PSNR and SSIM of images with clear color boundaries and complex edges.