Data Augmentation Enhanced Age-Invariant Face Recognition System

Though research on face recognition has advanced significantly, it is still unclear if it is possible to identify the same person's face when naturally ageing reliably. Intra-subject differences brought on by facial ageing impair the precision of face recognition algorithms. Convolutional neural networks (CNNs) are the most recent method that researchers have developed to increase the accuracy of Age-Invariant Face Recognition (AIFR) systems. This research examines the challenge of using CNNs to learn from an available standard training dataset for face ageing research. These all contributed to using data augmentation to improve the accuracy of the AIFR system. A heterogeneous dataset was also developed by adding the curated face images of African subjects to the Face and Gesture Recognition Network Ageing Dataset (FG-NET) dataset after obtaining all legal requirements and following all ethical guidelines for including the African subjects. The experimental setup used the heterogeneous dataset to train three different CNN architectures. The resultant effect was the creation of nine AIFR models. The experimental set-ups for ResNet-18, Inception-v3 and Inception-ResNet-v2 on datasets 1, 2 and 3 were used to create nine models. In conclusion, ResNet-18 on dataset 2 proved to be the best-performing AIFR model.