Comparative study between Deep Face, Autoencoder and Traditional Machine Learning Techniques aiming at Biometric Facial Recognition

Abstract

Biometric technology is increasingly present in our daily lives whether in mobile devices or commercial sectors because it is an approach where there is great difficulty in being circumvented, unlike traditional models of security and identification. Biometrics is the means by which these technologies can identify individuals and uses physical or behavioral characteristics of the human being and the physical characteristics can be: the iris, face, palm, fingerprint, among others. The behavioral ones can be: way of walking and typing dynamics. With the emergence of Deep Learning, a number of problems that were once solved with traditional machine learning models, have come to better results with this approach, but in the face recognition environment there is still no evidence that Deep Learning can achieve better results than traditional models, with different extractors of characteristics, when applied in different databases facial data. Therefore, the objective of this work is to perform a comparative study between traditional models of machine learning and Deep Learning focusing on Convolutional Neural Networks and Autoencoders for facial recognition.