Title: Face recognition performance using linear discriminant analysis and deep neural networks

Authors: Xhevahir Bajrami; Blendi Gashi; Ilir Murturi

Addresses: Department of Mechatronics, University of Prishtina 'Hasan Prishtina', Bregu i Diellit, pn. 10000, Prishtina, Kosovo ' Department of Software Design, University of Prizren 'Ukshin Hoti', Rruga e Shkronjave, nr. 1, 20000, Prizren, Kosovo ' Department of Computer Engineering, University of Prishtina 'Hasan Prishtina', Bregu i Diellit, pn. 10000, Prishtina, Kosovo

Abstract: The face recognition applications deal with large amounts of images and remain difficult to accomplish due to when displayed with images taken in unlimited conditions. Linear discriminant analysis (LDA) is a supervised method that uses training samples to obtain the projection matrix for feature extraction, while deep neural networks are trainable for supervised and unsupervised tasks. In this paper, we present our results of experiments done with linear discriminant analysis (LDA) and deep neural networks (DNN) for face recognition, while their efficiency and performance are tested on labelled faces in the wild (LFW) dataset. We used two methods of DNN, k-nearest neighbours algorithm (k-NN) and support vector machine (SVM). Experimental results show that the DNN method achieves better recognition accuracy and recognition time is much faster than the LDA method in large-scale datasets. Deep learning methods have shown high accuracy even for images coming out of the dataset.

Keywords: face recognition; linear discriminant analysis; LDA; deep neural network; DNN.

DOI: 10.1504/IJAPR.2018.094818

International Journal of Applied Pattern Recognition, 2018 Vol.5 No.3, pp.240 - 250

Received: 21 May 2018
Accepted: 12 Jul 2018
Published online: 23 Sep 2018 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article