Age Estimation of Real-Time Faces Using Convolutional Neural Network

Abstract

Age classification of an individual from an unconstrained real-time face image is rapidly gaining more popularity and this is because of its many possible applications from security control, surveillance monitoring to forensic art. Several solutions have been proposed in the past few years in solving this problem. Many of the existing traditional methods addressed age classification from face images taken from a controlled environment, only a few studied an unconstrained imaging conditions problem from real-time faces. However, deep learning methods have proven to be effective in solving this problem especially with the availability of both a large amount of data for training and high-end machines. In view of this, we propose a deep learning solution to age estimation from real-life faces. A novel six-layer deep convolutional neural network (CNN) architecture, learns the facial representations needed to estimate ages of individuals from face images taken from uncontrolled ideal environments. In order to further enhance the performance and reduce overfitting problem, we pre-trained our model on a large IMDB-WIKI dataset to conform to face image contents and then tuned the network on the training portions of MORPH-II and OIU-Adience datasets to pick-up the peculiarities and the distribution of the dataset. Our experiments demonstrate the effectiveness of our method for age estimation in-the-wild when evaluated on OIU-Adience benchmark that is known to contain images of faces acquired in ideal and unconstrained conditions, where it achieves better performance than other CNN methods. The proposed age classification method achieves new state-of-the-art results with an improvement of $8.6\%$ (Exact) and $3.4\%$ (One-off) acccuracy over the best-reported result on OIU-Adience dataset.