融合DQ_CoALBP和LPQ算子的人脸活体检测

Face Anti-Spoofing Liveness Detection Combining DQ_CoALBP with LPQ Descriptors

面向图片与视频攻击下的人脸活体检测任务,提出了一种差分量化相邻局部二值模式(DQ_CoALBP)算子,综合不同方向上的图像局部中心点与周围点之间的差值,同时为了更加充分地描述人脸的彩色纹理信息,在颜色空间通道上将该算子与局部相位量化(LPQ)直方图特征相融合,并利用支持向量机(SVM)分类器实现人脸反欺诈判别。在公开CASIA-FASD与Replay-Attack数据集上的实验结果表明,DQ_CoALBP算子的表现均优于LBP、LPQ、CoALBP与DQ_LBP四种算子。采用YCbCr颜色空间在融合DQ_CoALBP与LPQ算子时,CASIA-FASD数据集上的等错误率(EER)和半错误率(HTER)分别降至2.5%和3.7%,Replay-Attack数据集上实现了无差错检测,优于一些深度卷积神经网络模型。

Aiming at the task of face anti-spoofing liveness detection under image and video attacks,a descriptor of differ-ent quantization co-occurrence of adjacent local binary pattern(DQ_CoALBP)is proposed,which comprehensively con-siders the difference between local center point from different directions and surrounding points of the image.In addition,in order to describe the color texture information of the face more fully,the descriptor is combined with local phase quanti-zation(LPQ)histogram features in the channel of color space.And support vector machine(SVM)classifier is used for face anti-spoofing discrimination.The experimental results on the public CASIA-FASD and Replay-Attack datasets show that the performance of DQ_CoALBP descriptor is better than LBP,LPQ,CoALBP and DQ_LBP.When DQ_CoALBP descriptor is combined with LPQ on YCbCr color space,the equal error rate(EER)and half total error rate(HTER)on the CASIA-FASD dataset are reduced to 2.5%and 3.7%respectively,and error free detection is realized on the Replay Attack dataset,which is better than some models based on deep convolution neural network.