Face Liveness Detection Based On Multispectral Imaging

With the maturing of face recognition technology, its commercial products are more and more widely used in our life. However, in identity authentication and security protection system based on face recognition with with traditional RGB image, legitimate users’face can be easily simulated by photos, video, mask or three-dimensional model, so the method based on RGB image can not complete the task of face liveness detection. In this paper, we study a method based on multispectral imaging for face liveness detection. The main contents of research include:1. Aimed at human skin liveness detection, multispectral imaging system has been built for extractiong the special reflectivity information of human skin. Multispectral face database including photos, mask,3-D model are also established.2. On the basis of reflectivity information of facial skin in the multispectral image, combined with gradient information of contour and features of face, a method of multispectral face liveness detection based on gradient has been proposed. Among them, four gradient-based features are designed to conduct face liveness detection under various illumination conditions.3. Baesd on multispectral face database and feature extraction method above, train classifier with SVM(support vector machine), effectiveness of the above features has been verified, the true positive rate of method reached98.3%and false negative rate is1.3%. Also, further testing and analysis are performed with different bands and posture.Based on a cost-effective multispectral imaging system, this paper proposes a method for face liveness detection based on gradient and test the effectiveness of the method on multispectral database. Potential applications of multispectral imaging in the future is shown in this paper.