《计算机应用》唯一官方网站 ›› 2023, Vol. 43 ›› Issue (2): 595-600.DOI: 10.11772/j.issn.1001-9081.2021122214

• 多媒体计算与计算机仿真 • 上一篇    

姿态干扰下的深度人脸验证

王奇(), 雷航, 王旭鹏   

  1. 电子科技大学 信息与软件工程学院,成都 610054
  • 收稿日期:2022-01-04 修回日期:2022-03-02 接受日期:2022-03-21 发布日期:2022-05-16 出版日期:2023-02-10
  • 通讯作者: 王奇
  • 作者简介:雷航(1960—),男,四川自贡人,教授,博士,CCF会员,主要研究方向:嵌入式操作系统
    王旭鹏(1986—),男,山东烟台人,助理研究员,博士,CCF会员,主要研究方向:计算机视觉、模式识别。
  • 基金资助:
    国家自然科学基金资助项目(61802052)

Deep face verification under pose interference

Qi WANG(), Hang LEI, Xupeng WANG   

  1. School of Information and Software Engineering,University of Electronic Science and Technology of China,Chengdu Sichuan 610054,China
  • Received:2022-01-04 Revised:2022-03-02 Accepted:2022-03-21 Online:2022-05-16 Published:2023-02-10
  • Contact: Qi WANG
  • About author:LEI Hang, born in 1960, Ph. D., professor. His research interests include embedded operating system.
    WANG Xupeng, born in 1986, Ph. D,research assistant. His research interests include computer vision, pattern recognition.
  • Supported by:
    National Natural Science Foundation of China(61802052)

摘要:

人脸验证广泛应用于生活中各种场景,而普通RGB图像的获取依赖于光照条件。为解决光照和头部姿态对任务的干扰,提出了一个基于卷积神经网络的孪生网络L2-Siamese。首先,直接将成对的深度图作为输入;然后,用两个共享权重的卷积神经网络分别提取面部特征后,引入L2范数将不同姿态的人脸特征约束在一个半径固定的超球上;最后,通过全连接层将特征之间的差异映射为(0,1)区间的概率值来判断该组图像是否属于同一对象。为了验证L2-Siamese的有效性,在公共数据集Pandora上进行了测试。实验结果显示,L2-Siamese整体性能良好。将Pandora根据头部姿态干扰大小进行分组后的测试结果表明,在头部最大姿态干扰下,与当前最好的算法全卷积孪生网络相比,该网络预测准确率提高了4个百分点,有明显提升。

关键词: 人脸验证, 深度图, 深度学习, 孪生网络, 神经网络

Abstract:

Face verification is widely used in various scenes in life, and the acquisition of ordinary RGB images is extremely dependent on illumination conditions. In order to solve the interference of illumination and head pose, a convolutional neural network based Siamese network L2-Siamese was proposed. Firstly, the paired depth images were taken as input. Then, after using two convolutional neural networks that share weights to extract facial features respectively, L2 norm was introduced to constrain the facial features with different poses on a hypersphere with a fixed radius. Finally, the fully connected layer was used to map the difference between the features to the probability value in (0,1) to determine whether the group of images belonged to the same object. In order to verify the effectiveness of L2-Siamese, a test was conducted on the public dataset Pandora. Experimental results show that L2-Siamese has good overall performance. After the dataset was grouped according to the size of head pose interference, the test results show that the prediction accuracy of L2-Siamese is 4 percentage points higher than that of the state-of-the-art algorithm fully-convolutional Siamese network under the maximum head pose interference, illustrating that the accuracy of prediction has been significantly improved.

Key words: face verification, depth image, deep learning, Siamese network, neural network

中图分类号: