Exploring convolutional neural network for multi-spectral face recognition

Abstract

Multi-spectral face recognition has procured noteworthy consideration over a most recent times because of its potential capacity to acquire spatial and spectral information over the electromagnetic range, which cannot be obtained using traditional visible imaging techniques. With the advances in deep learning, Convolutional Neural Network (CNN) based approach has become an essential method in the field of face recognition. In this work, we present two face recognition techniques using face image at nine unique spectra ranging from Visible (VIS) to Near-Infra-Red (NIR) range of the electromagnetic spectrum. This paper is based on the application of using CNN as feature extractor along with Support Vector Machine (SVM) and k-Nearest Neighbor (k-NN) as a classifier on the images of nine different spectra ranging from 530 nm-1000 nm. The obtained performance evaluation results show highest Rank-1 recognition rate of 84.52 percent using CNN-KNN, demonstrating the significance of using CNN extracted features for improved accuracy.