Title: On the Performance of Neural Networks for Face Recognition: Linear or Nonlinear Classifiers?
Authors: Lima, Rafael O.; Barreto, Guilherme A.
Abstract: The main goal of this study is to empirically evaluate linear and nonlinear neural network based classifiers to be embedded in mobile devices. For this purpose, this paper reports a comprehensive performance comparison study involving 10 neural network models for human face recognition. All the classifiers are evaluated on two benchmarking face databases. For the linear classifiers we evaluate eight variants of the LMS and perceptron learning rules, while for the nonlinear ones we evaluate two state-of-the-art classifiers. In addition, we also evaluate empirically the robustness of all classifiers to the presence of gaussian and impulsive noise in test images. The results of the experiments indicate that the linear classifiers perform as good as nonlinear ones, with the advantage of demanding much lower computational resources.
Keywords: Face recognition; neural networks; linear classifiers; nonlinear classifiers; embedded applications
Paper as PDF: st_34.4.pdf
BibTex file: st_34.4.bib