Face Recognition using Convolutional Neural Network

Abstract

Biometric systems have emerged as a highly dynamic and critical domain within information security, garnering substantial attention. These systems leverage inherent human attributes, which are uniquely distinctive across individuals, such as the iris, voice patterns, facial characteristics, palm prints, retinal scans, gait, and fingerprints. Among these, the recognition of human faces represents a particularly significant biometric technique, possessing extensive applicability across numerous real-world scenarios. However, the complexity of face recognition is considerably amplified by variations in pose, illumination conditions, and the effects of aging. Consequently, these challenges have spurred an escalating demand for, and intensive research into, robust face recognition systems.

This study first undertakes a comprehensive investigation into the state-of-the-art in face feature extraction and classification methodologies. Subsequently, we propose a novel face recognition system predicated on deep learning principles. We carefully constructed a dataset and trained a GoogleNet-based model, employing advanced deep learning techniques. Our experimental results demonstrate a high degree of efficiency, achieving a 91.66% accuracy rate in facial recognition. The expansive potential of deep learning applications is evident, and by harnessing vast quantities of information through significant computational capacity, the precision of such systems can be further augmented. We argue that our contribution offers strong insights into the deep learning paradigm, encompassing the entire process from dataset construction to the careful preparation and deployment of models. This holistic approach aims to substantially reduce human effort.