Age-invariant Face Recognition Based on Sample Enhancement of Generative Adversarial Networks

Age-invariant face recognition is very challenging. To solve the problem that it is difficult to collect cross-age face data of the same person, this paper proposes to use generative adversarial networks to generate the same person's young and elderly age data for sample enhancement, and use these data to train age-invariant face recognition model. The proposed method is verified on public face-aging datasets: FGNET and CADA-VS. Rank-1 recognition rates reaches 85.04% on FGNET data sets and verification accuracies reaches 96.99% on CADA-VS data sets.

I. Introduction

Age-invariant face recognition is an important field, and it has widespread applications, for instance, identifying criminals, passport identification and finding missing children. Age-invariant face recognition is very challenging because the same person has large differences in face at different ages and it is difficult to collect age-invariant data from the same person.