《计算机应用》唯一官方网站 ›› 2023, Vol. 43 ›› Issue (3): 943-948.DOI: 10.11772/j.issn.1001-9081.2022020218

• 多媒体计算与计算机仿真 • 上一篇    

面向民国档案印章分割的改进U-Net

杨有1,2, 张汝荟2(), 许鹏程2, 康慷2, 翟浩2   

  1. 1.重庆国家应用数学中心(重庆师范大学),重庆 401331
    2.重庆师范大学 计算机与信息科学学院,重庆 401331
  • 收稿日期:2022-02-28 修回日期:2022-05-20 接受日期:2022-05-20 发布日期:2022-08-16 出版日期:2023-03-10
  • 通讯作者: 张汝荟
  • 作者简介:杨有(1965—),男,重庆人,教授,博士,主要研究方向:数字图像处理、计算机视觉
    张汝荟(1998—),女,江苏南京人,硕士研究生,主要研究方向:数字图像处理
    许鹏程(1997—),男,江苏南京人,硕士研究生,主要研究方向:数字图像处理
    康慷(1998—),男,浙江绍兴人,硕士研究生,主要研究方向:计算机视觉
    翟浩(1987—),男,山西大同人,博士,主要研究方向:数字图像处理、计算机视觉。
  • 基金资助:
    重庆市研究生联合培养基地项目(2019-45);重庆师范大学(博士启动/人才引进)基金资助项目(21XLB032)

Improved U-Net for seal segmentation of Republican archives

You YANG1,2, Ruhui ZHANG2(), Pengcheng XU2, Kang KANG2, Hao ZHAI2   

  1. 1.National Center for Applied Mathematics in Chongqing (Chongqing Normal University),Chongqing 401331,China
    2.College of Computer and Information Science,Chongqing Normal University,Chongqing 401331,China
  • Received:2022-02-28 Revised:2022-05-20 Accepted:2022-05-20 Online:2022-08-16 Published:2023-03-10
  • Contact: Ruhui ZHANG
  • About author:YANG You, born in 1965, Ph. D., professor. His research interests include digital image processing, computer vision.
    XU Pengcheng, born in 1997, M. S. candidate. His research interests include digital image processing.
    KANG Kang, born in 1998, M. S. candidate. His research interests include computer vision.
    ZHAI Hao, born in 1987, Ph. D. His research interests include digital image processing, computer vision.
  • Supported by:
    Chongqing Postgraduate Joint Training Base Project(2019-45);Doctoral Start-Up /Talent Introduction Funded Project of Chongqing Normal University(21XLB032)

摘要:

精准分割民国档案图像中的印章,有助于该类档案的智慧应用。针对民国档案印侵严重和过多噪声的问题,提出用于印章分割的网络UNet-S。该网络在保留U-Net的编解码器结构和跳跃连接的基础上从三个方面进行改进:一是使用多尺度残差模块替代U-Net原有的卷积层,使UNet-S既能有效提取多尺度特征,又能避免网络退化和梯度爆炸等问题;二是在多尺度残差模块中将普通卷积替换为深度可分离卷积(DSConv),大幅减少网络的参数量;三是使用BCEDiceLoss并根据仿真实验结果优选权重因子,以解决民国档案数据不平衡的问题。实验结果表明,相较于U-Net、DeepLab v2等网络,UNet-S的Dice相似系数(DSC)、平均交并比(mIoU)、平均像素准确率(MPA)取得了最优结果,最多提高了17.38%、32.68%和0.6%,参数量最多下降了76.64%。可见,UNet-S在民国档案数据集中分割效果更佳。

关键词: 深度可分离卷积, U-Net, 多尺度特征提取, 民国档案, 印章分割

Abstract:

Achieving seal segmentation precisely, it is benefit to intelligent application of the Republican archives. Concerning the problems of serious printing invasion and excessive noise, a network for seal segmentation was proposed, namely U-Net for Seal (UNet-S). Based on the encoder-decoder framework and skip connections of U-Net, this proposed network was improved from three aspects. Firstly, multi-scale residual module was employed to replace the original convolution layer of U-Net. In this way, the problems such as network degradation and gradient explosion were avoided, while multi-scale features were extracted effectively by UNet-S. Next improvement was using Depthwise Separable Convolution (DSConv) to replace the ordinary convolution in the multi-scale residual module, thereby greatly reducing the number of network parameters. Thirdly, Binary Cross Entropy Dice Loss (BCEDiceLoss) was used and weight factors were determined by experimental results to solve the data imbalance problem of archives of the Republic of China. Experimental results show that compared with U-Net, DeepLab v2 and other networks, the Dice Similarity Coefficient (DSC), mean Intersection over Union (mIoU) and Mean Pixel Accuracy (MPA) of UNet-S have achieved the best results, which have increased by 17.38%, 32.68% and 0.6% at most, and the number of parameters have decreased by 76.64% at most. It can be seen that UNet-S has good segmentation effect in the dataset of Republican archives.

Key words: Depthwise Separable Convolution (DSConv), U-Net, multi-scale feature extraction, archives of the Republic of China, seal segmentation

中图分类号: