A Hierarchical Estimation System for Human Age using Artificial Neural Networks

Abstract

Image analysis for human face recorded a wide range of attention by different authors over the last decades. Despite the significant accuracy yielded by Deep Learning Networks (DLN), Artificial Neural Networks (ANN) still has own contribution. This is due to the obstacle detected in DLN classification, where the almost all control on the type of extracted features is by the adopted DLN. On the other side, previous studies on age estimation showed that each period of human life has its own age-progression signs (features), which differ from others in different age periods. As a result controlling the type of extracted features, using DLN, is a challenging task, which can be controlled using the ANN networks. This paper proposes a pyramidal classification system of two stages, within the first on it uses texture features to classify the human age into four age periods child (0- 8) years, teenager (9- 19) years, adult (20- 49) years, and old (50 and above) years. The number and length these periods are determined due to the changes witnessed the period. In the second stage, the system estimates the age within each period by training and testing the features of the specific period. Due to the wide range of periods, the yielded results recorded high accuracy comparing with other works in age estimation and even with the second stage of this work. Determining the exact ages (±1 year) contains more false positive and false negative cases than the first stage, which means less accuracy. As benchmarking with state of art in the field of age estimation, this paper yielded superior results. The experimental results showed considerable performance yielding (97.75%) for the best accuracy of estimation with (96%) as average of yielded accuracy.