Deep Learning Edge Detection in Image Inpainting
Zheng Zheng, Haiyan Qiao · 2022
Deep learning-based image inpainting algorithm EdgeConnect was recently developed by K. Nazeri et al. Unlike traditional image inpainting methods, EdgeConnect uses the innovative edge-first and color-next approach. In this paper, EdgeConnect is improved to become a completely deep learning-based image inpainting method by using deep learning-based edge detection. The EdgeConnect and the modified EdgeConnect are both trained on the same datasets and the results are scored with the image quality assessment metrics including PSNR, SSIM, MAE and FID. The modified EdgeConnect in this paper demonstrates better learning efficiency and inpainting performance.