Hybrid Facial Recognition System Using Histogram of Oriented Gradients and Deep Learning with Dimensionality Reduction

This paper presents a novel hybrid approach for facial recognition that integrates Histogram of Oriented Gradients (HOG) and FaceNet deep learning architecture with Principal Component Analysis (PCA) dimensionality reduction. Through comprehensive experimental validation on the Labeled Faces in the Wild (LFW)dataset with 40 subjects and 400 images per subject, the proposed system achieves 98.8% recognition accuracy while reducing computational complexity by11.4×compared to the combined feature approach without dimensionality reduction. The integration of reinforcement learning for hyperparameter optimization, homomorphic encryption for cloud security, and generative AI techniques demonstrates the system's robustness across varying lighting conditions, pose variations (0° to 90°), and occlusion scenarios. Performance metrics reveal training time reduction from 198.7 seconds to 43.8 seconds with memory optimization from 8.1GB to 1.9GB. This research demonstrates the effectiveness of multi-modal feature fusion and dimensionality reduction in real-time biometric systems deployed in resource-constrained environments.

Keywords: Facial Recognition, Histogram of Oriented Gradients, FaceNet, Principal Component Analysis, Deep Learning, Dimensionality Reduction, Cloud Security, Generative AI