江汉大学学报(自然科学版) ›› 2024, Vol. 52 ›› Issue (2): 68-77.doi: 10.16389/j.cnki.cn42-1737/n.2024.02.008

• 计算机图形图像学 • 上一篇    

基于改进 Criminisi 算法的中国古画修补系统

陈 莹,贾 茜*,漆为民,孙一鸣,黄心怡   

  1. 江汉大学 人工智能学院,湖北 武汉 430056
  • 发布日期:2024-04-11
  • 通讯作者: 贾茜
  • 作者简介:陈 莹(1996— ),女,硕士生,研究方向:计算机视觉。
  • 基金资助:
    江汉大学高水平学术成果培育项目(KYCXJJ202317);江汉大学校级科研项目(2021yb052)

Chinese Ancient Painting Repair System Based on Improved Criminisi Algorithm

CHEN Ying,JIA Qian*,QI Weimin,SUN Yiming,HUANG Xinyi   

  1. School of Artificial Intelligence,Jianghan University,Wuhan 430056,Hubei,China
  • Published:2024-04-11
  • Contact: JIA Qian

摘要: Criminisi 算法广泛用于文物修补中,由于 Criminisi 算法采用全局搜索的方式寻找匹配 块,导致修补速度慢,因此提出一种基于改进 Criminisi 算法的中国古画修补系统。首先,改进优 先权函数,使用加权求和的形式,解决了修补后期如果优先权为 0 则导致修补效果差的问题;然 后,在计算样本块间的相似度时,引入几何距离判断更优的匹配块,避免只考虑颜色差导致的纹 理匹配错误;最后,使用步长为 2 的搜索方式,减少了冗余搜索,提高了修补速率。另外,使用 MATLAB R2021a 设计了古画修补系统,方便文物保护人员进行操作。实验结果表明,该算法提 升了修补质量,提高了修补速度。

关键词: Criminisi 算法, 中国古画修补, 图像修补, 欧氏距离

Abstract: The Criminisi algorithm is widely used in cultural relics repair,which finds matching blocks by global search,resulting in slow repair speed. This paper proposed the Chinese ancient painting repair system based on improved Criminisi algotithm. Firstly,the priority function was improved,and the problem that the priority was 0,which led to a poor repair effect,was solved by the weighted sum. Then,when calculating the similarity between sample blocks,the better matching blocks were judged by the geometric distance to avoid the error of texture matching caused by only considering the color difference. Finally, the redundant search was reduced and the repair rate was improved by the search method with a step size of 2. Additionally,a system of restoring ancient paintings was designed using MATLAB R2021a facilitating easy operation for cultural relic protection personnel.The experimental results showed that the improved Criminisi algorithm improved repair quality and operation speed.

Key words: Criminisi algorithm, Chinese ancient painting, image repair, Euclidean distance

中图分类号: