激光与光电子学进展, 2019, 56 (3): 031002, 网络出版: 2019-07-31   

基于多特征融合的3D打印面具攻击检测 下载: 1252次封面文章

3D Printing Mask Attacks Detection Based on Multi-Feature Fusion
作者单位
1 北京工业大学北京未来网络科技高精尖创新中心, 北京 100124
2 北京工业大学北京市数字化医疗3D打印工程技术研究中心, 北京 100124
摘要
针对人脸认证系统的欺骗攻击,传统欺骗攻击方式主要包括照片和视频攻击。随着三维(3D)打印技术的快速发展,使用3D面具进行欺骗攻击逐渐成为新威胁。在剪切波变换基础上,结合人脸3D几何特征和局部区域纹理变化,针对3D面具欺骗攻击提出一种利用多层自编码网络进行特征融合分类来识别攻击面具的方法。通过非下采样剪切波变换从目标人脸3D图像中提取低频子带和高频子带。在低频子带上利用尺度空间函数对特征点进行检测、定位及方向分配,生成特征描述子。将所生成的特征描述子与高频子带上提取的纹理特征输入栈式自编码器和softmax分类器进行瓶颈特征融合分类。在基于柔性TPU材质3D打印面具BFFD数据库上的实验结果表明,相比于以往单独使用纹理特征的方法,加入3D几何特征的多特征融合方法对反3D面具攻击的准确率有显著提升。
Abstract
Aim

ing at the spoofing attacks for the current face authentication systems, the traditional spoofing attacks include displaying printed photos and replaying recorded videos. With the rapid development of three-dimensional (3D) printing technology, the 3D mask spoofing attack is becoming a new threat. On the basis of the shearlet transform and combining with the 3D geometric attributes and the local regional texture changes, a method by utilizing the multilayer autoencoder network to conduct the feature fusion-based classification to identify the attack mask is proposed for the 3D mask spoofing attack. The low-frequency sub-band and several high-frequency sub-bands are extracted from the 3D image of the target face by the non sub-sampled shearlet transform method. The scale space function is used to detect, locate and distribute the feature points and then to generate feature operators in the low-frequency sub-band . Then, the generated feature operators and the texture features extracted from the high-frequency sub-band are combined in series and fed into the stacked autoencoder network and the softmax classifier to conduct the bottleneck feature fusion-based classification. The experimental results in the BFFD database based on the flexible TPU material 3D print mask shows that, the multi-feature fusion method added the 3D geometric feature has an obvious improvement for the accuracy of the anti-spoofing performance against 3D mask attacks to compare with the previous method of using the texture feature alone.

陆经纬, 陈鹤天, 马肖攀, 陈继民. 基于多特征融合的3D打印面具攻击检测[J]. 激光与光电子学进展, 2019, 56(3): 031002. Jingwei Lu, Hetian Chen, Xiaopan Ma, Jimin Chen. 3D Printing Mask Attacks Detection Based on Multi-Feature Fusion[J]. Laser & Optoelectronics Progress, 2019, 56(3): 031002.

本文已被 2 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!