NNST-based Image Outpainting via SinGAN
Ryuto Sugahara, Weiwei Du · 2024
The restoration process can be further categorized into image inpainting and outpainting depending on the extent of the damage. Image inpainting is utilized when only a few parts require repair, while image outpainting is employed for more extensive restoration efforts. This study focuses on extensively damaged ancient fabric characterized by the presence of a repeated pattern. In order to complement the extensively damaged ancient fabric, this paper introduces a novel image outpainting algorithm based on Neural Neighbor Style Transfer (NNST) using SinGAN. SinGAN is employed to initially complement the damaged region, followed by the utilization of NNST to reconstruct the complemented region. The purpose is to enhance the clarity of the previously blurry region and adjust colors to align with the colors of the known region. The effectiveness of the proposed method is verified through both objective and subjective experiments, demonstrating its superiority over SinGAN.