Real-time Age-Group and Accurate Age Prediction with Bagging and Transfer Learning

Age prediction based on facial images has become a vital part in many scenarios since the human face contains most important bio-metric characteristics from race, gender to age. As a result, automatic age prediction has gained its potential importance. This paper addresses two approaches to this problem: accurate age and age-group prediction, both using convolutional neural networks (CNNs) and deep neural networks (DNNs) - two fundamental deep learning techniques, in combination with computer vision library, OpenCV to aid in classifying input images. This work is distinguished by two main key contributions. The first is feature extraction, where convolutional activation features could be obtained through using a novel hierarchical aggregation based neural networks to learn aging features from an abundant number of training data. The second contribution is the features classification part, in which we apply the Softmax Regression (SR) classifier to perform accurate age prediction whereas age-group prediction uses majority vote classifier. The results showed that the number and the quality of training images and subjects in the training dataset play a significant role in improving the accuracy of this work. The experimental results showed satisfactory and promising performance, which indicates another direction of further development and optimization of deep learning algorithms.