Open AccessDomain Adaptation for Robust Face Recognition Using Transfer Kernel Learning
Author(s) -
João Renato Manesco,
Aparecido Nilceu Marana
Publication year - 2020
Language(s) - English
Resource type - Conference proceedings
DOI - 10.5753/wvc.2020.13482
Subject(s) - biometrics , facial recognition system , computer science , domain adaptation , artificial intelligence , adaptation (eye) , three dimensional face recognition , kernel (algebra) , face (sociological concept) , transfer of learning , identification (biology) , face recognition grand challenge , computer vision , domain (mathematical analysis) , pattern recognition (psychology) , machine learning , face detection , mathematics , psychology , mathematical analysis , social science , botany , combinatorics , neuroscience , sociology , classifier (uml) , biology
In the last decades, for reasons of safety or convenience, biometric characteristics are increasingly being used to identify individuals who wish to have access to systems or places, and facial features are one of the most used characteristics for this purpose. For biometric identification to be effective, the recognition accuracy rates must be high. However, these rates can be very low depending on the difference (displacement) between the domain of the images stored in the database of the biometric system (source images) and the images used at the moment of identification (target images). In this work, we evaluated the performance of a domain adaptation method called Transfer Kernel Learning (TKL) in the face recognition problem. Results obtained in our experiments on two face datasets, ARFace and FRGC, corroborates that TKL is suitable for domain adaptation and that it is capable of improving significantly the accuracy rates of face recognition, even when considering facial images with occlusions, variations in illumination and complex backgrounds.