A Review of Performance Evaluation on 2D Face Databases
Gabriel Castaneda, Taghi M. Khoshgoftaar · 2017
Face recognition methods are evaluated against face image databases. Recent face image databases provide an evaluation protocol for an impartial comparison and assessment of where a facial recognition algorithm stands compared to other methods. Unfortunately, many authors test their facial recognition methods using either restricted face databases, random subsets from public databases, or do not follow the established testing protocol for the specific face database. As a consequence, complete and accurate comparisons between different facial recognition methods is not possible. The established protocol for a face database allows for comparisons across research groups. In this paper, a review of the currently available 2D face databases with their evaluation protocols is described. We aim to improve the current understanding of testing protocols with facial recognition in an effort to help researchers replicate and compare results accurately. Also, common performance measures in facial recognition and recommendations are discussed for face databases that do not have an evaluation protocol.