An Innovative Approach to Cross-Age Face Recognition: Combining Deformable Convolutional Networks with VGG-16 Network

Facial features undergo changes over time, posing increasing challenges for face recognition tasks. Cross-age face recognition aims to improve the accuracy of these tasks and can be broadly categorized into two approaches: generative models and discriminative models. This paper focuses on the latter, specifically employing a discriminative model for cross-age face recognition. The VGG-16 Network has demonstrated excellent performance in image classification and recognition. This paper enhances the ability of VGG-16 networks to extract age-invariant features by utilizing the attention mechanism of Deformable Convolutional Networks (DCN). By integrating this model with cosine similarity, we develop a deep learning method for cross-age face recognition, referred to as the VGG16-DCN method. This approach utilizes Deformable Convolutional Networks to facilitate the model's ability to automatically discern temporal variations in facial features while extracting identity characteristics that remain stable across different ages, thereby enhancing the model's accuracy in cross-age face recognition. We evaluate this method on the VGGFace and FG-NET datasets. Given the small size of the FG-NET dataset, we apply data augmentation to expand the original dataset to seven times its initial size, thereby enhancing the model's generalization ability and robustness. Our approach achieves an accuracy of 93.45% on the VGGFace dataset and 80.5% on the FG-NET dataset. Additionally, we compared our model with several other models, consistently achieving higher accuracy than the others, which demonstrates the strong performance of our method in cross-age face recognition.