Research On Presentation Attack Detection Of Facial Multi-modal Biometrics

Among all biological characteristics of the human,the iris is extremely stable and distinguishable,so the iris recognition system is gradually popularized in our life.However,many iris recognition systems still cannot guarantee full reliability when resisting various types of spoofing attacks,thus hindering the deployment of iris recognition in advanced security scenarios.Therefore,iris presentation attack detection(Presentation Attack Detection,PAD)has become one of the most concerned problems in biometrics.To this end,an iris PAD method based on enhanced gray-scale image space is proposed.The proposed method mainly enhances the difference in texture blur between genuine and fake irises through space mapping.The experimental results show that the analysis and transformation of the image space can significantly solve the problem that the genuine and various fake irises are difficult to distinguish in the original gray space,and the detection network can accurately discriminate the remaining unknown types of fake iris images to achieve the state-of-the-art performance in iris PAD.Furthermore,the proposed method improves the generalization of the iris PAD method.In recent years,the attack pattern of 3D masks has become a new challenge and has attracted more researchers' interest.At the same time,the existing PAD algorithm only focuses on spoofing attacks on global faces,while ignoring the fine iris texture that is difficult to counterfeit.Because the realistic degree between the 3D masks and real faces is very high,it is clear that the existing PAD algorithms based on planar face attacks such as printed face and replay-attack are not enough to break the spoofing attack of the realistic 3D mask,then it is necessary to accurately detect the iris information on the face to break the 3D mask spoofing attack.Therefore,a new PAD database of 3D masks that can be detected fine iris textures of faces is established.This database not only uses realistic 3D masks as the main attack pattern,but also devotes to precisely detect the fine iris textures of real and fake faces,which is the first PAD database that fuses multiple biometric modals.However,3D masks are expensive and the number of purchases is very limited.Therefore,additional collection variables are needed to simulate rich realworld scenarios.The proposed database has 15 subjects and 3 different types of attacks.In addition,it includes 7 cameras from stationary and mobile devices and 6 lighting settings covering typical lighting conditions,so the total number of images in database is(10-15)subjects × 4 attributes(1 real + 3 fake)× 7 collection devices × 6 lighting types.Finally,utilizing two types of benchmark experiments to measure the image quality of the database.