A computer approach for face aging problems

Abstract

This paper first presents a novel age-estimation approach combining Active Appearance Models (AAMs) and Support Vector Regression (SVR) to yield the highest accuracy of age recognition rates of all comparable published results both in overall Mean Absolute Error (MAE) and Mean Absolute Error per decade of life (MAEd). The combination of AAMs and AVR is used again for a newly proposed face age-progression method. The familial information of siblings is also collected so that the system can predict the future faces of an individual based on parental and sibling facial traits. Especially, a new longitudinal familial face database is presented. Compared to other databases, this database is unique in that it contains family-based longitudinal images. It contains not only frontal faces but also the corresponding profiles. It has the largest number of pre-adult face images per subject on average. © 2010 Springer-Verlag Berlin Heidelberg.