基于图像色彩纹理的人脸活体检测算法研究

Research on Face Liveness Detection Method Based on Image Color LBP Features

对于人脸活体检测的研究主要集中在分析人脸图像的亮度,因此丢弃图像的色彩信息,然而色彩信息可用于区分活体人脸和伪人脸。提出在HSV和YCbCr色彩模式下,提取人脸图像LBP纹理信息的人脸活体检测方法。使用LBP特征提取器从图像亮度和色彩通道中提取纹理特征,并将各个通道的LBP特征融合,输入卷积神经网络,利用卷积神经网络进行分类。基于CASIA Face、Replay-Attack数据集的实验结果表明,该方法优于传统方法。

Research on face liveness detection is mainly focused on the analysis of the luminance of image that discards the color information which can help distinguish between real face and fake face. In order to solve this problem, proposes a method based on image color LBP features in this essay by putting the fusion features of LBP features extracted from channels of luminance and color separately when image is transformed to HSV or YCrCb color space into convolutional neural network instead of RGB color image. The evaluation of our method on two challenging bench mark face spoofing databases, namely the CASIA Face Anti-Spoofing Database, the Replay-Attack Database showed that the proposed method outperforms traditional methods.