Age-Related Face Recognition Using Siamese Networks and Vision Transformers

Face recognition plays a crucial role in various applications, ranging from security to personal convenience. Recent advancements have emphasized the importance of recognizing individuals based on age-related facial features within this domain. This paper presents a comprehensive evaluation of two deep learning architectures for age-based face recognition: Siamese Convolutional Networks (SCNs) and Vision Transformers (ViTs). Convolutional Neural Networks (CNNs), which are critical in modern face recognition, serve as the backbone for Siamese Convolutional Networks (SCNs). SCNs are specifically designed to detect similarities between input pairs by emphasising local features crucial for age-related distinctions. In contrast, ViTs, initially developed for natural language processing, have demonstrated promising performance in image recognition, showcasing their aptitude for capturing global image context. This work investigates the performance of these distinct architectures in discerning age-related variations within facial data features. Performance comparisons were conducted on three established SCN models and two ViT architectures. The results revealed that the optimal SCNs primarily focused on the mouth, nose, and eye regions, indicating their reliance on local features for age estimation. Interestingly, the ViT models achieved superior performance despite lacking explicit feature localization. This suggests that a holistic understanding of the facial context may be more effective than focusing solely on isolated features for age-based recognition.