Scholarlink

<p>Face anti-spoofing (FAS) plays a vital role in securing the face recognition systems based on video replay. Recent works have revealed the usefulness of 3D convolutional neural networks (3D-CNNs) to classify videos between bonafide and attacks. These 3D-CNN are characterized to obtain a high accuracy at expense of a high computational cost. In this context, this work presents a new strategy for visualizing and classifying video data using 3D-CNN applied to FAS challenge. It describes an original visualization method, named feature map cube visualization, based on using histograms of the inner-layer network output (feature map). This visualization method has served as the motivation for developing a novel FAS classification method, called histogram skewness classification (HiSkew), which classifies FAS videos based on skewness histograms derived from spatiotemporal feature maps. The accuracy and computational cost of HiSkew are compared with some representative spatiotemporal convolutional models of the state of the art. This experimentation reveals that HiSkew is able to obtain an accuracy similar to the state of the art with a much lower computational cost.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

HiSkew: a novel histogram skewness classification applied to face anti-spoofing

  • 期刊论文

  • 发表日期：

    2025年4月30日

查看全文
  • Vitor Luiz da Silva,
  • Francesc Giné,
  • Magda Valls

摘要

Face anti-spoofing (FAS) plays a vital role in securing the face recognition systems based on video replay. Recent works have revealed the usefulness of 3D convolutional neural networks (3D-CNNs) to classify videos between bonafide and attacks. These 3D-CNN are characterized to obtain a high accuracy at expense of a high computational cost. In this context, this work presents a new strategy for visualizing and classifying video data using 3D-CNN applied to FAS challenge. It describes an original visualization method, named feature map cube visualization, based on using histograms of the inner-layer network output (feature map). This visualization method has served as the motivation for developing a novel FAS classification method, called histogram skewness classification (HiSkew), which classifies FAS videos based on skewness histograms derived from spatiotemporal feature maps. The accuracy and computational cost of HiSkew are compared with some representative spatiotemporal convolutional models of the state of the art. This experimentation reveals that HiSkew is able to obtain an accuracy similar to the state of the art with a much lower computational cost.