Reconhecimento Facial usando Wavelets de Gabor via Álgebra Multilinear

Abstract

In this work, we propose a facial recognition system that exploits the multilinear nature of images. The feature extraction occurs in two stages. First, Gabor wavelets are used for feature selection. Then, the higher order singular value decomposition (HOSVD) is applied to separate the multimodal factors of the images. Two approaches based on Euclidean and Mahalanobis distances are proposed, respectively, for the classification of the set of feature vectors. The proposed facial recognition approach exhibits higher average success rates than competing methods. By analyzing the results for various training settings, the proposed approach demonstrates stability when there is deficiency in the amount of training images.