Weiwei Du, Urano Yuto, Irawati Nurmala Sari
Image inpainting is widely used in image reconstruction, image matting, 3D generation from 2D images [1], and many other tasks. Existing inpainting methods can be broadly classified into two categories: deep learning-based approaches [3] and non-deep learning-based approaches. Deep learning models require high-performance computing resources, and some patch-based inpainting methods [2] do not restore damaged regions well due to interference between the line structure region and the texture region.This paper proposes a patch-based image inpainting method based on mask optimization with auxiliary lines. The proposed method separates the line structure region from the texture region with mask optimization, thereby preventing mutual interference and maintaining consistency between the repaired and undamaged regions. In addition, damaged regions with strong gradients can be effectively restored. The effectiveness of the proposed method has been verified using the evaluation metrics PSNR, SSIM, and MSSSIM. © 2025 IEEE.
Kyoto Institute of Technology, Information and Human Science, Kyoto, Japan; Kyoto Institute of Technology, Information Science, Kyoto, Japan; Brawijaya University, Informatics Engineering, Malang, Indonesia
