Age Estimation from Face Image Leveraging Concatenated Features of Vision Transformer Along with Resnet-50

Age is a biometric trait used in various applications such as recommendation, access control, criminal detection, so on. Age estimation algorithms primarily rely on facial features, with deep learning often showing promising results in this domain. However, due to the wide range of faces and numerous other extrinsic and intrinsic features, biological variances, and uncertainty make it complicated to estimate age. Thus, a single deep learning-based approach may downplay the significance of some extensive face data and other facial age features. Deep residual networks and Vision Transformer models are commonly used for facial age estimation, but their combined effectiveness and contrast in estimating human chronological age have not been thoroughly examined. To address these challenges, this work proposes a novel hybrid model combining the strengths of two state-of-the-art architectures, ResNet50 and Vision Transformer(ViT), capture local and global contextual information, improving age estimation robustness which outperforms standalone ResNet50 and Vision Transformer models, achieving state-of-the art accuracy. The analysis evaluates the effectiveness of multi-feature fusion from face images, including HOG-ResNet50-ViT features, comparing different machine learning and deep learning algorithms. The proposed method utilizes the UKTFace dataset, involving 23708 images, and attains 4.88 MAE by addressing age estimation as a regression problem.