This work focuses on the most common and cheapest face spoofing methods, i.e., photo attacks (including the printed photo on a paper or a photo demonstrated on an electronic screen). Many previous works [3-6] propose to classify genuine and fake samples based on frontal face images and achieve good performance on several face spoofing databases. However, in real applications, the imposter will try his best to fool the system and the texture difference between the genuine and fake samples is usually very small. In order to achieve robust face anti-spoofing performance, other cues like 3D face structure and motion pattern can be incorporated. In this work, we propose to detect spoofing photo attacks based on a sequence of rotated face images. Both the structure and texture information from the rotated face sequence are exploited. In practice, the users are only asked to take simple movement (i.e., rotate their faces). As pointed in [7], this head rotation requirement is much simpler than traditional challenge-response-based face anti-spoofing method, in which a combination of multiple movements is usually necessary. The proposed anti-spoofing method is applicable to face recognition applications such as face access control and remote authentication on mobile devices. The simple head rotation requirement is acceptable in these applications.

Chapter Contents:

• 14.1 Introduction
• 14.2 Related works
• 14.3 Overview of the proposed method
• 14.4 Sparse 3D facial structure recovery
• 14.4.1 Initial recovery from two images
• 14.4.1.1 Intrinsic parameter estimation
• 14.4.1.2 Facial structure reconstruction
• 14.4.2 Facial structure refinement
• 14.4.3 Key frame selection
• 14.5 Face anti-spoofing classification
• 14.5.1 Structure-based anti-spoofing classifier
• 14.5.2 Texture-based anti-spoofing classifier
• 14.6 Experiments
• 14.6.1 Database description
• 14.6.2 Evaluation protocols
• 14.6.3 Results of structure-based method
• 14.6.3.1 Intra-sensor evaluation
• 14.6.3.2 Cross-sensor evaluation
• 14.6.3.3 False classified sample analysis
• 14.6.4 Results of texture-based method
• 14.6.5 Combination of structure and texture clues
• 14.6.6 Computational cost analysis
• 14.7 Conclusion
• References

Inspec keywords: image texture; mobile computing; face recognition; visual databases; message authentication; image sequences

Other keywords: face photo spoofing attacks; rotated face sequences; structure information; face spoofing databases; face recognition; frontal face images; texture information

Subjects: Image recognition; Mobile, ubiquitous and pervasive computing; Computer vision and image processing techniques; Spatial and pictorial databases; Data security