东北大学学报(自然科学版) ›› 2025, Vol. 46 ›› Issue (11): 48-57.DOI: 10.12068/j.issn.1005-3026.2025.20250072

• 信息与控制 • 上一篇    下一篇

基于双阶段深度学习的图像修复方法

王天琪1, 张迪2(), 张鑫宇2, 张一鸣3   

  1. 1.东北大学 档案馆,辽宁 沈阳 110819
    2.东北大学 信息科学与工程学院,辽宁 沈阳 110819
    3.东北大学 机械工程与自动化学院,辽宁 沈阳 110819
  • 收稿日期:2025-06-24 出版日期:2025-11-15 发布日期:2026-02-07
  • 通讯作者: 张迪
  • 作者简介:王天琪(1991—),女,辽宁沈阳人,东北大学档案馆中级馆员.
  • 基金资助:
    辽宁省档案局科技项目(2024-B-17);辽宁省教育科学规划重点项目(JG25DA009)

Dual-Stage Deep Learning-Based Image Inpainting Method

Tian-qi WANG1, Di ZHANG2(), Xin-yu ZHANG2, Yi-ming ZHANG3   

  1. 1.Archives,Northeastern University,Shenyang 110819,China
    2.School of Information Science & Engineering,Northeastern University,Shenyang 110819,China
    3.School of Mechanical Engineering & Automation,Northeastern University,Shenyang 110819,China.
  • Received:2025-06-24 Online:2025-11-15 Published:2026-02-07
  • Contact: Di ZHANG

摘要:

针对现有图像修复方法在处理复杂损伤和依赖成对训练样本方面的不足,提出一种端到端的双阶段自监督图像修复方法.该方法包括退化模拟、边缘修复和色彩重建3个环节,并通过协同优化实现结构与色彩的协同还原.实验在档案馆历史图像集上进行,采用峰值信噪比(peak signal-to-noise ratio,PSNR)、感知相似度(learned perceptual image patch similarity,LPIPS)、弗雷歇起始距离(Fréchet inception distance,FID)和色彩丰富度得分(Colorfulness Score)4项指标进行评估.结果表明,该方法在图像还原精度、感知质量和色彩表现方面均优于主流方法,具备良好的实用价值和鲁棒性.

关键词: 图像修复, 自监督学习, 图像退化模拟, 协同优化, 属性解耦, 深度学习

Abstract:

In view of the shortcomings of existing image inpainting methods in handling complex damage and relying on paired training samples, an end-to-end dual-stage self-supervised image inpainting method was proposed. The method included degradation simulation, edge restoration, and color reconstruction; synergistic restoration of structure and color was achieved through cooperative optimization. Experiments were conducted on a historical image dataset from archives, and evaluation was performed using four metrics: peak signal-to-noise ratio(PSNR), learned perceptual image patch similarity(LPIPS), Fréchet inception distarce(FID)and Colorfulness Scare. Experimental results demonstrate that the proposed method outperforms existing mainstream methods in terms of image restoration accuracy, perceptual quality, and color performance, exhibiting good practical value and robustness.

Key words: image inpainting, self-supervised learning, image degradation simulation, cooperative optimization, attribute disentanglement, deep learning

中图分类号: