Comparative Analysis for The Performance of Convolutional Neural Network Architectures in Facial Recognition Insights From LFW and ORL Databases

Abstract

Facial recognition technology based on artificial intelligence, machine learning techniques and deep learning techniques is frequently used in security systems through facial biometric authentication, law enforcement, arresting suspects and criminals using surveillance and face detection, attendance monitoring systems on real-time, and other various use cases. Deep learning (DL) techniques have lately made important contributions to face recognition technology, particularly the convolutional neural network (CNN); which is most widely used for image processing due to its high accuracy for modeling complex phenomena. There are diverse CNNs that are characterized by their architecture. In this article, four architectures are performance analyzed: LeNet, AlexNet, ResNet, and DenseNet. To determine which architecture is the best, two performance measures are used for comparison; accuracy and time. These architectures were trained and tested on the LFW and ORL databases. The best performing architecture was DenseNet, with an accuracy of 100% On ORL face database and it took the least time to implement. This can have good potential to be developed Systems and applications that use facial recognition in their work.