Automatic Facial Age Estimation

Abstract

In recent times, human facial age estimation attracted a lot of attention in the fields of computer vision and pattern recognition because of its significant applications in many domains such as biometrics, law enforcement, health care, security control and surveillance. Numerous approaches were offered and studied by researchers concerning this problem. Nevertheless, age estimation systems that can be globally used for every situation have not been reached yet. In our thesis, we presented an in-depth analysis of facial age estimation. In the proposed solution, three feature extraction algorithms are evaluated, including a Handcraft technique (LBP), a learned hand-craft technique (CBFD), and a deep learning technique (DCTNet). A classification phase using MLP and SVM classifiers is implemented in order to test the extracted features. Through multiple experiments, our proposed DCTNet achieved 3.84 MAE (mean absolute error) for exact age estimation using the MORPH-II facial image database. In this regard, the proposed method provided a very respectable performance compared to existing state-of-the-art methods.