A comprehensive comparison of features and embedding methods for face recognition

Abstract

Face recognition is an essential issue in modern-day applications since it can be used in many areas for several purposes. Many methods have been proposed for face recognition. It is a difficult task since variations in lighting, instantaneous mimic varieties, posing angles, and scaling differences can drastically change the appearance of the face. To suppress these complications, effective feature extraction and proper alignment of face images gain as much importance as the recognition method choice. In this paper, we provide an extensive comparison of the state-of-theart face recognition methods with the most well-known techniques used in feature representation. In order to test the performances of these various methods, we created a new face database, named the ESOGU face database, which includes frontal images of 100 people taken under different lighting and posing conditions. In addition to our new database, we also present experiments on the well-known AR face database to obtain more general and reliable results. Moreover, we investigate the automatic face detection and automatic normalization of the face images in the databases. Based on this, we discuss the use of such automatic methods for face recognition applications.