Anti-Photo Spoof Methods In Face Recognition

Biometric techniques, which rely on the inherited biometric traits taken from the user himself for authentication, have gained wide range of applications recently. Unfortunately, once such biometric data is stolen or duplicated, the advantages of biometrics become disadvantages immediately. Spoofing with photograph or video is one of the most common manner to circumvent a face recognition system.In our work, we present a novel method for liveness detection against photo spoofing in face recog-nition, which is a real-time and non-intrusive method to address this based on individual images from a generic webcamera. The major research and innovation of this work can be summarized from the aspects below:1,The task is formulated as a binary classification problem, in which, however, the distribution of positive and negative are largely overlapping in the input space, and a suitable representation space is hence of importance.2,We investigate the different nature of imaging variability from a live human or a photograph based on the analysis of Lambertian model, which leads to a new strategy to exploit the information contained in the given image. We show that some current illumination-invariant face recognition algo-rithm can be modified to collect the needed latent samples. We propose two strategies to extract the essential information about different surface properties of a live human face or a photograph, in terms of latent samples.3,Based on these, we develop two new extensions to the sparse logistic regression model which allow quick and accurate spoof detection.4,We constructed a publicly available photograph imposter database using a generic cheap webcamera bought from an electronic market, which is accessible freely from http://parnec.nuaa.edu.cn/xtan/data/NuaaImposterdb.html.Experiments on a large photo imposter database show that the proposed method gives preferable detection performance compared to others. Although there are lots of related work in the field of texture analysis, their goal is different from ours.We believe that our work is the first one trying to use the learning technique to distinguish whether a given static image is from a live human or not. We are currently investigating the possibility to integrate various texture descriptors to further improve the performance.