Abstract

We study the problem of cross-population age estimation.Human aging is determined by the genes and influencedby many factors. Different populations, e.g., males and fe-males, Caucasian and Asian, may age differently. Previousresearch has discovered the aging difference among differ-ent populations, and reported large errors in age estima-tion when crossing gender and/or ethnicity. In this paperwe propose novel methods for cross-population age esti-mation with a good performance. The proposed methodsare based on projecting the different aging patterns into acommon space where the aging patterns can be correlatedeven though they come from different populations. The proFjections are also discriminative between age classes due to mthe integration of the classical discriminant analysis techanique. Further, we study the amount of data needed in the ttarget population to learn a cross-population age estima-tor. Finally, we study the feasibility of multi-source cross-population age estimation. Experiments are conducted on qa large database of more than 21,000 face images selected wfrom the MORPH. Our studies are valuable to significantly ereduce the burden of training data collection for age estimantion on a new population, utilizing existing aging patterns seven from different populations. e t s