基于反向掩码和WGAN-GP的破损老照片修复算法

Damaged Old Photo Inpainting Algorithm Based on Reverse Mask and WGAN-GP

针对当前老照片修复算法修复的老照片存在模糊、伪影等问题,提出结合反向掩码机制和生成对抗网络的破损老照片修复算法。该方法使用反向掩码技术聚焦于待修复目标区域,采用联合感知驱动的损失函数和WGAN-GP鉴别器网络训练,增强修复图像整体和局部一致性。此外,通过变换RGB图像色彩空间和添加高斯滤波,扩充老照片图像数据集。经过实验,结果表明该算法相较于对比算法能够精准修复老照片中缺失部分。已修复照片的结构和纹理与原照片保持一致,在感官视觉上更自然。

Aiming at the problems of blur and artifacts in old photo inpainted by current old photo repairing algorithms,a dam-aged old photo inpainting algorithm that combines the reverse mask mechanism and generative adversarial network is proposed.This method utilizes reverse mask techniques focused on the target area to be inpainted,and proposes a joint perception-driven loss func-tion and WGAN-GP discriminator network training to enhance the global and local consistency of the inpainted image.In addition,by transforming the RGB image color space and adding Gaussian filtering,the old photo image data set is expanded.After experi-ments,the results show that the algorithm in this paper can accurately inpaint the missing parts in old photo compared to the compar-ison algorithm.The structure and texture of the inpainted photo are consistent with the original photo,which is more natural in the senses and vision.