Neural networks for facial age estimation: a survey on recent advances

Abstract

Soft biometrics has emerged out to be a new area of interest for the researchers due to its growing real-world applications. It includes the estimation of demographic traits like age, gender, scars, ethnicity. Moreover, researchers are trying to develop models which can accurately estimate the age or the age group of a person using different biometric traits. Presently, neural networks proves out to give the best classification results for age estimation using human faces. Hence, in this paper, we have surveyed and compared all the neural network models developed and implemented for facial age estimation from 2010 to 2019. We have precisely compared all twenty-three different research works done so far to estimate age from human faces using neural networks. Most of the works are based on convolutional neural networks and a few are based on feed forward back propagation and autoencoders. Important details, issues and results of each work are thoroughly discussed for better knowledge of interested researchers. This paper also includes details on other classification techniques for facial age estimation to give an overall idea of all additional techniques adopted by the scientists till date. Details like neural network model names, datasets used, main contributions, evaluation metrics and results are adopted for a tabular and easy to understand comparison study. Finally, the paper concludes by mentioning the other relevant future research tasks that can be done in this challenging area of research.