《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (12): 3708-3714.DOI: 10.11772/j.issn.1001-9081.2021101723

• 人工智能 • 上一篇    

基于区域分块和轻量级网络的人脸反欺骗方法

贺丹1, 何希平1,2(), 李悦1, 袁锐1, 牛园园1   

  1. 1.重庆工商大学 人工智能学院,重庆 400067
    2.检测控制集成系统重庆市工程实验室(重庆工商大学),重庆 400067
  • 收稿日期:2021-10-08 修回日期:2021-12-14 接受日期:2021-12-23 发布日期:2021-12-31 出版日期:2022-12-10
  • 通讯作者: 何希平
  • 作者简介:贺丹(1997—),女,湖南涟源人,硕士研究生,主要研究方向:计算机视觉、深度学习、活体检测
    李悦(1998—),女,河南南阳人,硕士研究生,主要研究方向:计算机视觉、深度学习、人脸属性编辑
    袁锐(1999—),男,四川巴中人,硕士研究生,主要研究方向:计算机视觉、深度学习、人脸属性编辑
    牛园园(1998—),女,河南许昌人,硕士研究生,主要研究方向:深度学习、膜计算。
  • 基金资助:
    重庆市研究生科研创新项目(CYS21398);重庆工商大学研究生科研创新项目(yjscxx2021-112-99)

Face anti-spoofing method based on regional blocking and lightweight network

Dan HE1, Xiping HE1,2(), Yue LI1, Rui YUAN1, Yuanyuan NIU1   

  1. 1.School of Artificial Intelligence,Chongqing Technology and Business University,Chongqing 400067,China
    2.Chongqing Engineering Laboratory for Detection,Control and Integrated System (Chongqing Technology and Business University),Chongqing 400067,China
  • Received:2021-10-08 Revised:2021-12-14 Accepted:2021-12-23 Online:2021-12-31 Published:2022-12-10
  • Contact: Xiping HE
  • About author:HE Dan,born in 1997, M. S. candidate. Her research interests include computer vision, deep learning, liveness detection.
    LI Yue, born in 1998, M. S. candidate. Her research interests include computer vision, deep learning, face attribute editing.
    YUAN Rui, born in 1999, M. S. candidate. His research interests include computer vision, deep learning, face attribute editing.
    NIU Yuanyuan, born in 1998, M. S. candidate. Her research interests include deep learning, membrane computing.
  • Supported by:
    Graduate Scientific Research and Innovation Project of Chongqing(CYS21398);Graduate Innovation Project of Chongqing Technology and Business University(yjscxx2021-112-99)

摘要:

如何高效地辨别各种被攻击的人脸是人脸识别过程中迫切需要解决的问题。基于深度学习的人脸反欺骗方法在有着高性能的同时,也带来了庞大的参数量和计算量,使其无法部署在移动或嵌入式设备中。针对以上问题,提出了一种基于区域分块和轻量级网络的人脸反欺骗方法。首先,对训练样本进行随机区域分块;然后,设计了一种基于注意力机制的轻量级网络用于特征提取和图像分类;最后,为了提高测试准确率,对测试样本进行基于区域分块的数据扩增。实验结果表明,所提模型在CASIA-FASD和REPLAY-ATTACK数据集上达到了100%的准确率;在CASIA-SURF数据集的Depth模态上获得了99.49%的准确率和0.458 0%的平均分类错误率(ACER),远优于ResNet、ShuffleNet等卷积神经网络,且该模型的参数量也仅有0.258 2 MB。在实际应用中,端到端的轻量级网络结构使所提模型更方便部署在移动设备上来进行实时的人脸反欺骗检测。

关键词: 人脸反欺骗, 区域分块, 中心差分卷积, 注意力机制, 轻量级网络, 卷积神经网络

Abstract:

How to effectively identify all kinds of attacked faces is an urgent problem to be solved in the process of face recognition. The face anti-spoofing methods based on deep learning have high performance, but also bring a large number of parameters and calculation, so they cannot be deployed in mobile or embedded devices. To solve the above problems, a face anti-spoofing method based on regional blocking and lightweight network was proposed. Firstly, the training samples were randomly blocked. Then, a lightweight network based on attention mechanism was designed for feature extraction and image classification. Finally, in order to improve the detection accuracy, data augmentation was conducted on the test samples based on regional blocking. Experimental results show that the proposed model reaches 100% accuracy on REPLAY-ATTACK and CASIA-FASD datasets. At the same time, the proposed model obtains 99.49% accuracy and 0.458 0% Average Classification Error Rate (ACER) on the Depth modal of CASIA-SURF dataset, which are much better than those obtained by convolutional neural networks such as ResNet and ShuffleNet. And the parameter amount of the model is only 0.258 2 MB. In practical applications, the end-to-end lightweight network structure makes the proposed model easier to be deployed on mobile devices for real-time face anti-spoofing detection.

Key words: face anti-spoofing, regional blocking, Central Difference Convolution (CDC), attention mechanism, lightweight network, Convolutional Neural Network (CNN)

中图分类号: