基于卷积神经网络的人脸活体检测模型研究

Research on Face in Vivo Detection Model Based on Convolutional Neural Network

摘要: 针对以往身份验证系统容易被攻击活体欺诈的问题，提出了一种基于深度学习的活体检测算法。主要包括：1) 通过图像增强技术对现有少量真实和欺诈图像进行平移、旋转、翻转等几何变换自制数据集，用于模型训练、验证及测试。2) 针对现有算法对光照条件不足的照片识别准确率较低的问题，提出了一种基于照片的活体检测模型。通过检测照片中的人脸区域，并针对人脸区域中的像素、纹理以及人脸特征差异进行活体和非活体二分类预测。实验结果表明，本文提出的算法在光照条件不足的图像中准确率达到86%。同时与以往模型相比，在保证模型预测精度的基础上减少了参数的数量。

Abstract: Aiming at the problem that previous identity verification systems are easy to be attacked by vivi-section fraud, a vivisection detection algorithm based on deep learning was proposed. It mainly includes: 1) Self-made data sets for translation, rotation and reversal of a small number of existing real and fraudulent images by using image enhancement technology for model training, verification and testing. 2) Aiming at the problem of low recognition accuracy of existing algorithms for photos with insufficient illumination conditions, a photo-based in vivo detection model is proposed. By detecting the face region in the photo, we predict the difference of pixel, texture and face feature in the face region by living and nonliving dichotomies. Experimental results show that the accuracy of the proposed algorithm reaches 86% in images with insufficient illumination conditions. At the same time, compared with the previous models, the number of parameters is reduced on the basis of en-suring the prediction accuracy of the model.