ENSEMBLE METHOD USING FACENET AND CNN FOR FACE RECOGNITION

Abstract

Face recognition has become a vital application in the domains of computer vision and biometric identification, playing a crucial role in security, authentication, and surveillance systems. This research focuses on the development and evaluation of a face recognition system using deep learning, specifically comparing the performance of a traditional Convolutional Neural Network (CNN), the FaceNet architecture, and a hybrid ensemble model that integrates both. The objective was to assess each model’s effectiveness, accuracy, and generalization capability in recognizing human.. All models were trained using a facial image dataset and evaluated across ten epochs using standard performance metrics such as accuracy and loss. The CNN model achieved a peak accuracy of 94.69% but exhibited inconsistent learning and signs of overfitting. FaceNet performed with greater stability, reaching a peak accuracy of 97.43% and maintaining low loss values throughout training. The ensemble model, which combines outputs from CNN and FaceNet, surpassed both, achieving the highest peak accuracy of 98.20% and the lowest final loss. In practical evaluation, the ensemble successfully identified a known individual and accurately inferred demographic features such as gender and age range. The results demonstrate that combining models into an ensemble yields even greater performance, making it the most suitable approach for real-world face recognition applications.