Automatic Age Detection Through Facial Features Using Convolutional Neural Networks

A model capable of automatically detecting the age of a human being, based only on the image of his face, represents an added value for several sectors. Ageing is a natural and complex process in the human being development. This process is affected by intrinsic and extrinsic factors. The comprehension of this process is fundamental to detect age based on facial characteristics. Here, we propose the development of a facial image-based ageing detection model for users. The system comprises, in an initial phase, the preprocessing of the image datasets followed by the development of an age detection model through an extensive configuration of convolutional neural networks. Several models employing the Xception, VGG-16, and Inception-V4 networks were developed and assessed. The CNN model using the Xception architecture with the ADAM optimiser performed the best among all models. It was trained on multiple public accessible datasets using a sampling-based method to tackle data distribution imbalance. An accuracy rate of 88\% and MCC of 82\% were achieved across all age groups, encompassing children, young people, adults, and the elderly.