Multi-scale Feature Fusion and Multi-task Learning for Cross-age Face Recognition

Cross-age face recognition has long been a critical challenge in the field of biometric identification, as facial appearance can change significantly with age. Addressing the issue of how to mitigate the influence of age-related variations during model training and enabling the model to produce age-invariant identity features is crucial. In this paper, we propose Multi-scale & Multi-task Cross-age Face Recognition (MMCFR) that incorporates several key techniques, including multi-scale feature fusion and Triplet loss, to improve the model's ability to extract fundamental facial features. These techniques help ensure the robustness of the features by capturing both local and global facial information at different scales. Moreover, during the subsequent multi-task learning process, we introduce an attention mechanism to guide the model in focusing selectively on age-independent features. This allows the model to better differentiate identity-related attributes from age-related changes, thus refining the feature extraction process. In addition, adversarial learning is employed to further enhance the model's ability to extract discriminative features, ensuring that the generated feature representations are more distinct and less susceptible to variations due to aging. Our proposed models demonstrate superior performance compared to conventional approaches, as evidenced by improved results on widely-used benchmark datasets such as CACD-VS, FG-NET, and LFW. Furthermore, our models show potential for broader applications, including the recognition of individuals in old photographs and the identification of missing persons over time. Experimental results on the AgeDB-30 datasets and CALFW datasets further validate the effectiveness of our approach, confirming its robustness and versatility in handling cross-age face recognition tasks.