© 2018 Mohamed Sayed and Faris Baker. This open access article is distributed under a Creative Commons Attribution (CC-
BY) 3.0 license.
Journal of Computer Science
Original Research Paper
Thermal Face Authentication with Convolutional Neural
Network
Mohamed Sayed and Faris Baker
Faculty of Computer Studies, Arab Open University, Kuwait
Article history
Received: 25-10-2018
Revised: 10-11-2018
Accepted: 08-12-2018
Corresponding Author:
Mohamed Sayed
Faculty of Computer Studies,
Arab Open University, Kuwait
Email: msayed@aou.edu.kw
Abstract: Matching thermal face images as a method of biometric
authentication has gained increasing interest because of its advantage of
tracking a target object at night and in total darkness. Therefore, for
security purposes, it has become highly favourable and has extensive
applications, for instance, in video surveillance at night. The aim of this
study is to present a simple and efficient deep learning model, which
accurately predicts person identification. A pre-trained Convolutional
Neural Network (CNN) is employed to extract the features of the multiple
convolution layers of the low resolutions’ thermal infrared images. To run
the program and evaluate the performance, we use a sample of 1500 resized
thermal images, each with resolution 181×161 pixels. The sample
comprises of images that were captured within different time-lapse and
with diverse emotions, poses and lighting conditions. The proposed
approach is effective compared to the state-of-the-art thermal face
recognition algorithms and achieves impressive accuracy of 99.6% with
less processing and training times.
Keywords: Deep Learning, Convolutional Neural Networks, Image
Processing, Face Recognition, Thermal Images
Introduction
Recognitions with a thermal camera is a challenge
that have been recently improved by adopting deep
learning methods using Convolutional Neural Networks
(CNNs). Thermal camera forms a picture by capturing
various heat levels emitted from objects. One of the
applications of visualizing and identifying objects is face
recognition. There are many useful applications for face
recognition; an important one is an application for security
at night where an intruder needs to be identified in the
absence of light. Furthermore, it can be used as an
additional tool to normal cameras for identification.
Ensemble method of using both thermal and normal
cameras adds more certainty weight to the proof of
identity. In this study, faces are identified in different
circumstances such as full light, partial light and dim light
as well as in different emotional status, all based on the
heat emitted from them rather than the light intensities
reflected from them; the technology also allows us to
measure the accuracy of the recognition using CNNs.
Image processing (Nixon and Aguado, 2002) is a
technique that is used to change over a picture into an
enhanced computerized shape. It plays out a few
operations together with a specific end goal to get an
improved picture or to concentrate some valuable data
from it. It is also a sort of flag regulation which
information is a picture similar to a video edge or photo
and might yield a picture or attributes related with that
picture. Typically, image-processing framework
incorporates pictures as two-dimensional signs while
effectively applying strategies of set flag handling to
them. It is among quickly developing advances today
together with its applications in different parts of a
business. Advanced processing methods help in
controlling the computerized pictures by utilizing
personal computers. The three general stages that a wide
range of information needs to experience while utilizing
computerized system are pre-handling, improvement and
show and data extraction.
CNNs are a special kind of multi-layer neural
networks trained with a back-propagation algorithm that
extracts important features, supplemented with filters
that prevent overfitting and eliminate noise modelling.
While a fully connected layer has a high cost of
parameters and high risks of overfitting, the
convolutional layer is more suitable for 2D spatial
information because it captures the spatial characteristics