Using unsupervised deep learning for human age estimation problem

Abstract

Automatic age estimation from facial images is a challenging problem upcoming in recent years. In this paper, we propose to use unsupervised deep learning features to improve accuracy of existing age estimation algorithms. The active appearance model and the bio-inspired features with a cascade of support vector machine classifiers were chosen to be the basic approaches. Experiments on the FGNET age database demonstrated that adding unsupervised deep learning features improved accuracy for some basic models. For example, adding the features to the active appearance model gave the 5% gain.