一种基于FAS-Transformer的人脸防伪方法  

A face anti-spoofing method based on FAS-Transformer

作　　者：魏鑫 马宏斌[1] 王英丽[1] WEI Xin;MA Hongbin;WANG Yingli(College of Electronic Engineering,Heilongjiang University,Harbin 150080,China)

机构地区：[1]黑龙江大学电子工程学院,哈尔滨150080

出　　处：《黑龙江大学自然科学学报》2023年第3期369-378,共10页Journal of Natural Science of Heilongjiang University

基　　金：黑龙江省自然科学基金(YQ2020F012);装备重大基础研究项目。

摘　　要：针对人脸防伪方法在应对不同表征攻击和未知表征攻击时,普遍存在有效性和泛化性差的问题,提出了一种人脸防伪方法。将在自然语言处理领域应用的注意力机制引入人脸防伪任务,获取特征之间的成对相似性关系,提升方法的有效性和泛化性。针对人脸防伪模型在训练过程中数据量不足的问题,引入迁移学习的思想,通过对FAS-Transformer预训练模型进行改进,使其快速地部署到二分类任务中。为验证所提出方法的有效性,分别设计了集内测试实验和集间测试实验,与主流方法进行了对比。实验结果表明,本方法获得了预期效果。Face anti-spoofing methods generally have poor effectiveness and generalization when dealing with different presentation attacks and unknown presentation attacks.To solve these problems,the face anti-spoofing method is proposed.The attention mechanism widely used in the field of natural language processing is introduced into the face anti-spoofing task to obtain the pairwise similarity relationship between features and improve the effectiveness and generalization of the method.Aiming at the problem of insufficient data in the training process of face anti-spoofing model,the idea of transfer learning is introduced.By improving the FAS-Transformer pre-trained model,it can be quickly deployed to the binary classification task.In order to verify the effectiveness and generalization of the proposed method,intra test experiments and inter test experiments are designed to compare with the state-of-the-art methods.The experimental results show that the proposed method achieves the expected results.

关 键 词：注意力机制 人脸防伪 迁移学习 FAS-Transformer 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]