A Scalable Face Recognition Framework: Integrating HOG Features, PCA, and CNN for Offline and Web-Based Systems

Most works on web-based face recognition systems have often focused on integrating Python-based machine learning models, which might result in a complex and multilevel environment needed for non-technical users.In this paper, two face recognition systems are proposed.The first system is an online web-based face recognition system, consisting of a frontend interface, server integration, and a backend.The backend is employed based on MATLAB for preprocessing, feature extraction, and classification.Also, the online system includes a key feature that allows adding new faces to the system and classifying them without retraining again.This key feature ensures that the system is scalable and adapting to new faces, as this is a very important feature in real-time applications.The second system is offline, serving as the same backend with the same MATLAB functions but without the web interface.Both systems contain the same preprocessing, feature extraction, and classification.The preprocessing step includes face detection, resizing the detected faces, and data splitting (training & testing).HOG is used for feature extraction, while PCA reduces dimensionality for feature selection.Finally, the extracted features are fed to CNN for classification purposes.The offline system achieved a rate of 98.57% and 98.73% on the FEI and ORL databases, respectively.These results indicate that using HOG features combined with PCA and classifying through CNN yields excellent performance compared with those results obtained by other approaches.The web interface highlights the benefits of integrating MATLAB in the backend for the face recognition system.This integration reduces the need for high computing power, enables remote server access, and ensures scalability for broader use.