A diagram of the proposed age estimation system. Source: Agbo-Ajala and ViririResearchers from the University of Kwazulu-Natal in South Africa are using machine learning to estimate a person’s age.

The system estimates a person’s age by analyzing facial images captured in moments when the subject is in a real-life setting, whereas similar age-estimating models are typically applied to images taken in controlled settings, such as in a photo studio or lab.

The researchers employed a deep learning method based on a deep convolutional neural network (CNN) multilayer architecture. The CNN model was first taught what facial images were the most critical for estimating a subject’s age, and learned to focus specifically on those. To do this, the CNN-based model was pre-trained on a dataset called IMDB-WIKI, which consists of more than 500,000 face images from both Wikipedia and IMDB. Each of the images used to train the system were labeled with the subject’s age.

Additionally, the team further fine-tuned the model by training it on two more databases — MORPH-II and OUI-Adience — to help the model detect differences and peculiarities in subjects' faces. The images held in the OUI-Adience database were taken of subjects in real-life settings.

Once trained, the team analyzed their model on images of subjects captured in uncontrolled settings and discovered that the model outperformed other CNN-based models for estimating a subject’s age.

"Our experiments demonstrate the effectiveness of our method for age estimation in the wild when evaluated on the OUI-Adience benchmark, which is known to contain images of faces acquired in ideal and unconstrained conditions," the researchers wrote. "The proposed age classification method achieves new state-of-the-art results, with an improvement in accuracy of 8.6 percent (exact) and 3.4 percent (one-off) over the best-reported result on the OUI-Adience dataset."

The team, which recently presented their findings at the International Conference on Computational Collective Intelligence (ICCCI) 2019 held in Hendaye, France, intends to further improve the age-estimating model’s performance by training it on even more datasets.

The results appear in the article Age Estimation of Real-Time Faces Using Convolutional Neural Network.