Edge Guided Loss in Image Inpainting

Image inpainting, also known as image completion, is the process of filling in the missing region of an incomplete image to make the image is visually plausible. Recently, the learning-based inpainting approach has revolutionized image inpainting. A variety of deep learning models have been applied to image inpainting and have achieved amazing results. This study concentrated on the loss function rather than the structure of the model in learning-based approaches. In this paper, a novel loss function containing the combination of edge guided loss term and weighted perceptual loss term was proposed. The proposed method was tested on the test set of ICME 2019 image inpainting challenge, which PSNR score is 1.3354 dB higher than the method without edge guided loss term, SSIM is 0.0014 higher, the qualitative result produced by the proposed method in objects removal task also is visually plausible and pleasing which won the objects removal track in ICME 2019 challenge.