AfriNet: A Facial Age Estimation Model for Age-Restricted Services
Keywords:
Facial Age Estimation, Deep Learning MobileNet AfriNet Demographic Diversity Classification and RegressionAbstract
This article introduces a lightweight, custom-designed convolutional neural network (CNN) called AfriNet for real-time facial age estimation through both regression and classification. The model integrates demographic diversity, especially African facial features. AfriNet was trained and tested with UTKFace, CASIA Africa Face, and a locally collected dataset from Baze University and the Federal College of Education, Zaria. The model makes a strong hit as it is capable of achieving up to 99% accuracy in the classification of age groups and a mean absolute error (MAE) as low as 0.65 years on CASIA Africa Face. Besides, AfriNet is comparably much quicker in its inference than MobileNet, VGG, and ResNet, thus, it can be easily deployed on resource-constrained devices. The paper talks about AfriNet’s potential use in age-restricted access control as a tool for the child online protection’s (COP) enhancement in Nigeria, while at the same time, it is addressing the ethical and fairness issues.
