Age Invariant Face Recognition using Feature Level Fused Morphometry of Lip-Nose and Periocular Region Features

Aging variation poses an interesting challenge for the task of automatic face recognition. Most face recognition studies that have addressed this problem focused on age estimation or aging simulation. Designing an appropriate feature representation and an effective matching framework for age invariant face recognition remains an open problem. In this research study, a novel age-invariant face recognition framework that is built based on two face biometric traits is proposed. The first trait is a set of anthropometric measurements acquired from the lip-nose complex. Lip-nose complex measurements have been known in several physiological studies to be discriminative among different ethnicities and among different genders within the same ethnicity. The second trait is based on extracting features of the periocular region using two robust descriptors, namely local binary patterns rotation invariant descriptor (LBPV) and GIST descriptors. The periocular area is considered as the most discriminative face area and is known to preserve its stability with aging. The two biometric face traits were combined at the feature level after being normalized separately using the Z-score rule and projected into a principle component analysis (PCA) subspace. Eight algorithms were derived as part of the proposed framework for performing age invariant face verification and identification.  Furthermore, the proposed framework was used to produce demographic information, namely: age group, gender, and ethnicity from aging faces. Experimental results show that the proposed framework reported an EER of 6.51% over the MORPH album 2 database (compared to 16.49% reported by Mahalingam et al), which is the largest public face aging database and 7.22% over the FG-NET database (compared to 24.08% reported by Mahalingam et al). The proposed framework achieved an identification accuracy of more than 95% (compared to 66.40% reported by Park et al) over the MORPH album 2 database, which is the largest public face aging database and 93% over the FG-NET database (compared to 38.10% reported by Geng et al ).

The authors declare no conflict of interest.

Not applicable

The datasets used in this study are openly available at [repository link] and the source code is available on GitHub at [GitHub link].

This work did not receive any external funding.