Facial age estimation using pre-trained CNN and transfer learning

Abstract

This paper tackled the problem of human facial age estimation using transfer learning of some pre-trained CNNs, namely VGG, Res-Net, Google-Net, and Alex-Net. Those networks have been fine-tuned with transfer learning and undergone many experiments to get the optimum number of outputs and the optimum age gap. Based on those experiments, a novel hierarchical network that generates high age estimation accuracy was developed. This new network consists of a set of pre-trained 2-classes CNNs (Google-Net) with an optimum age gap which can better organize the face images in the age group they belong to. To show its effectiveness, it was compared with other states of the art techniques on the FGNET and the MORPH databases.