Face Recognition from Thermal Images Using Wavelet and PCA-based Feature Selector and Backpropagation Neural Networks
Lina, Arlends Chris 

Procedings IASTED International Conference on Computational Bioscience, pp.291-295
ISBN : 978-0-88986-901-1

Abstract:

This paper represents the development of a 3D face recognition system from thermal images using Haar-wavelet, PCA-based feature selector, and Backpropagation Neural Networks. The Haar-wavelet transformation is used to analyze thermal images, and the Euclidian Distance Maximizing Method (EDMM) combined with the Principal Component Analysis (PCA) is utilized to choose feature vectors and reduce its dimensions. The face recognition system is then applied in several different scenarios concerning with visible images and thermal images with frontal faces and faces with various viewpoints. Experimental results show that the developed system gives high recognition accuracies with 100% for frontal faces from visible images, while 93% for visible images with various viewpoints. The system also gives high recognition accuracies for thermal images with 96% for frontal faces and 86% for thermal images with various face viewpoints.