DocFace: Matching ID Document Photos to Selfies*

摘要

Numerous activities in our daily life, including purchases, travels and access to services, require us to verify who we are by showing ID documents containing face images, such as passports and driver licenses. An automatic system for matching ID document photos to live face images in real time with high accuracy would speed up the verification process and reduce the burden on human operators. In this paper, we propose a new method, DocFace, for ID document photo matching using the transfer learning technique. We propose to use a pair of sibling networks to learn domain specific parameters from heterogeneous face pairs. Cross validation testing on an ID-Selfie dataset shows that while the best CNN-based general face matcher only achieves a TAR=61.14% at FAR=0.1% on the problem, the DocFace improves the TAR to 92.77%. Experimental results also indicate that given sufficiently large training data, a viable system for automatic ID document photo matching can be developed and deployed.