Hierarchical Aggregation based Deep Aging Feature for Age Prediction

Paper in proceeding, 2015

We propose a new, hierarchical, aggregation-based deep neural network to learn aging features from facial images. Our deep-aging feature vector is designed to capture both local and global aging cues from facial images. A Convolutional Neural Network (CNN) is employed to extract region-specific features at the lowest level of our hierarchy. These features are then hierarchically aggregated to consecutive higher levels and the resultant aging feature vector, of dimensionality 110, achieves both good discriminative ability and efficiency. Experimental results of age prediction on the MORPH-II databases show that our method outperforms state-of-the-art aging features by a clear margin. Experimental trails of our method across race and gender provide further confidence in its performance and robustness.