Enhancing Face Recognition Performance: A Comprehensive Evaluation of Deep Learning Models and a Novel Ensemble Approach with Hyperparameter Tuning

Abstract

In response to growing security concerns and the increasing demand for face recognition (FR) technology in various sectors, this research explores the application of deep learning techniques, specifically pre-trained Convolutional Neural Network (CNN) models, in the field of FR. The study leverages five pre-trained CNN models, including DenseNet201, ResNet152V2, MobileNetV2, SeResNeXt, and Xception, for feature extraction, followed by SoftMax classification. A novel weighted average ensemble model, optimized through a grid search technique, is introduced to enhance feature extraction and classification performance. Robust data pre-processing, encompassing resizing, data augmentation, splitting, and normalization, is emphasized to ensure reliable FR systems. The research systematically investigates hyperparameters across deep learning models, fine-tuning network depth, learning rate, activation functions, and optimization methods. Comprehensive evaluations are conducted on diverse datasets, including ORL, GTAV, GTF, FEI, LFW, F_LFW, and YTF, to assess the effectiveness of the proposed models. Key contributions of this work include the utilization of pre-trained CNN models for feature extraction, extensive evaluation across multiple datasets, the introduction of a weighted average ensemble model, emphasis on robust data pre-processing, systematic hyperparameter tuning, and the use of comprehensive evaluation metrics. The results showcase the superior performance of the proposed method, consistently outperforming all other models across key metrics, including Recall, Precision, F1 Score, Matthews Correlation Coefficient (MCC), and Accuracy. Receiver Operator Characteristic (ROC) curves further highlight the models' classification abilities. Notably, the proposed method achieves an exceptional accuracy of 99.48% on the LFW dataset, surpassing state-of-the-art benchmarks. In conclusion, this research presents a significant advancement in FR technology, offering a reliable and accurate solution supported by empirical evidence. The proposed method demonstrates the potential of pre-trained CNN models, ensemble learning, robust data pre-processing, and hyperparameter tuning in enhancing the accuracy and reliability of FR systems, with implications for various real-world applications.