Performance Evaluation of Machine Learning Based Face Recognition Techniques

Abstract

The robustness of machine-learning model-based face recognition techniques to image processing attacks using the quantization of extracted features is presented. Recently developed face recognition techniques based on machine learning models have been outperformed over traditional face recognition techniques. An efficient face recognition technology should be able to resist various image processing attacks. This paper presents the simulation results by evaluating ten variants of machine-learning-based face recognition techniques on ten well-known image processing attacks. The quality of face recognition techniques has been assessed on recognition accuracy. The performance has been evaluated on two well-known face databases viz. Bosphorus and University of Milano Bicocca (UMB) face database. The experimental results reveal that the Subspace discriminant ensemble-based face recognition model has consistently performed in most image processing attacks. All image processing attacks have been visually verified and presented.