Age Detection Using Deep Learning Techniques: A Comparative Study of CNN, VGG19, and ResNet.

Age Detection Using Deep Learning Techniques: A Comparative Study of CNN, VGG19, and ResNet Md. Mehedi Hassan, Md. Mahfuzur Rahman Department of Computer Science and Engineering, Daffodil International University, Dhaka, Bangladesh Corresponding Author: mehedi15-113[email protected] Abstract This research focuses on developing an automated system to estimate human age using facial images based on deep learning algorithms. The study compares the performance of three major architectures: Convolutional Neural Network (CNN), VGG19, and ResNet. A dataset consisting of 13,300 facial images was collected from open-source platforms such as Kaggle and Google, divided into nine distinct age categories. After extensive preprocessing and training using optimizers Adam and RMSProp, the VGG19 model achieved the highest accuracy of 88.24%, followed by CNN (88.14%) and ResNet (82.63%). These results demonstrate that deep convolutional architectures are effective in facial-based age estimation. This study contributes a comparative evaluation framework and highlights the importance of data diversity and model selection in deep learning-based age prediction. Keywords Deep Learning, CNN, VGG19, ResNet, Age Detection, Facial Recognition 1. Introduction The ability to estimate human age from facial features has become a significant area of research within computer vision and artificial intelligence. With the rapid advancement of machine learning, age detection systems can now assist in numerous applications including biometric verification, security systems, and social media moderation. Despite extensive studies, challenges remain due to variations in facial expressions, lighting, and other environmental factors. This research explores multiple deep learning architectures to address these challenges and improve the accuracy of facialbased age estimation. 2. Related Work Several previous studies have explored age estimation using facial features. Dehshibi and Bastanfard (2010) proposed a method achieving 86.64% accuracy on a limited dataset. Alonso et al. (2016) utilized the FG-NET and Adience datasets, attaining comparable results. Kwon and Lobo (1993) introduced early age classification methods using wrinkle analysis, while Horng et al. (2001) categorized 230 images into four age groups with 81.58% accuracy. Recent studies applying convolutional architectures have improved performance, but limitations remain due to dataset constraints and model generalization. Our study expands upon this by integrating larger datasets and testing multiple deep learning architectures. 3. Methodology This study used a dataset of 13,300 facial images from open-source repositories such as Kaggle and Google. Images were divided into nine age groups ranging from 0 to 100 years. The dataset was split into 60% for training, 10% for testing, and 30% for validation. Data preprocessing included resizing, augmentation, and normalization. Three deep learning models were implemented—CNN, VGG19, and ResNet50—each trained using TensorFlow with optimizers Adam and RMSProp on Google Colab GPU. The performance was evaluated using metrics including precision, recall, F1-score, and overall accuracy. 4. Results and Discussion Experimental analysis showed that the VGG19 model achieved the best accuracy among the tested models, reaching 88.24% with RMSProp optimizer. The CNN model performed closely, achieving 88.14% accuracy with RMSProp and 85.47% with Adam. ResNet yielded 82.63% accuracy, indicating comparatively lower generalization. The findings suggest that deeper architectures with transfer learning capabilities like VGG19 outperform shallower models for complex facial datasets. However, model performance also depended on preprocessing quality and the optimizer used during training. 5. Conclusion and Future Work This research demonstrated that deep learning-based architectures can achieve high accuracy in age estimation from facial images. Among the compared models, VGG19 proved most effective, highlighting the strength of deep convolutional feature extraction. Future research will focus on increasing dataset diversity, integrating facial expression recognition, and optimizing computational efficiency. In addition, developing a real-time age estimation application could further enhance practical usability in biometric and social applications. References [1] M.M. Dehshibi and A. Bastanfard, 'A new algorithm for age recognition from facial images,' Signal Processing, 2010. [2] J.B. Alonso et al., 'Automatic age detection based on facial images,' IEEE CCIS, 2016. [3] Y.H. Kwon and N.V. Lobo, 'Locating facial features for age classification,' SPIE, 1993. [4] W.B. Horng, C.P. Lee, and C.W. Chen, 'Classification of age groups based on facial features,' Tamkang Journal of Science and Engineering, 2001. [5] A.L. Ingole and K.J. Karande, 'Automatic age estimation from face images using facial features,' IEEE GCWCN, 2018.