A Hybrid Approach to Face Recognition using Local and Global Features

In the domain of biometric feature analysis, facial recognition serves as a crucial technique applied across various applications, including telemonitoring, access control, and electronic system security. This study introduces a robust hybrid facial recognition method tailored for real-time applications, accommodating variations in position and luminosity. The proposed approach demonstrates a tolerance of ±1% to positional shifts under diverse lighting conditions. By integrating local features such as Local Binary Patterns (LBP), Histogram of Oriented Gradients (HOG), and Gabor wavelets with global features including Discrete Wavelet Transform (DWT) and Principal Component Analysis (PCA), the method achieves high precision in defining facial features, effectively mitigating illumination effects. A neural network classifier is employed to enhance accuracy, achieving nearly 99.40% accuracy with single-image training per class. The system has been rigorously tested on the FERET and Yale B datasets, yielding satisfactory results.