A Real-Time Gender and Age Prediction System Based on Facial Images Using Convolutional Neural Networks

Abstract

Real-time gender and age prediction based on facial images has become essential in applications such as surveillance, demographic analytics, and human-computer interaction. However, existing systems often struggle with challenges like inconsistent lighting, occlusions, pose variations, and dataset imbalances, which reduce accuracy and hinder deployment on resource-constrained devices. This research introduces a real-time gender and age prediction system that leverages Convolutional Neural Networks (CNNs), utilizing fine-tuned ResNet-34 for gender classification and a modified VGG-16 for age estimation. A robust preprocessing pipeline, integrating image normalization, histogram equalization, data augmentation, and synthetic data generation, enhances model generalization across diverse datasets. Efficient face detection techniques, combined with lightweight model architectures, enable the system to achieve inference speeds below 500 milliseconds per frame without GPU acceleration. Experimental evaluations across IMDB-WIKI, UTKFace, MORPH II, and Adience datasets revealed gender classification accuracies exceeding 94% and age prediction one-off accuracies above 90%. The proposed system demonstrates a scalable and efficient solution for real-time facial attribute recognition in dynamic and resource-limited environments.