Comparison between Thermal and Visible Facial Features on a Verification Approach - Biomedical Engineering Systems and Technologies - page 274

Comparison between Thermal and Visible Facial
Features on a Verification Approach

Carlos M. Travieso, Marcos del Pozo-Baños, and Jesús B. Alonso

Signals and Communications Department (DSC),
Institute for Technological Development and Innovation in Communications (IDETIC),
University of Las Palmas de Gran Canaria (ULPGC),
Campus Universitario de Tafira, s/n, 35017, Las Palmas de Gran Canaria, Spain
{ctravieso,jalonso}@dsc.ulpgc.es

Abstract. A comprehensible performance analysis of a thermal and visible face
verification system based on the Scale-Invariant Feature Transform algorithm
(SIFT) with a vocabulary tree is presented in this work, providing a verification
scheme that scales efficiently to a large number of features. The image database
is formed from front-view thermal images, which contain facial temperature
distributions of different individuals in 2-dimensional format and the visible
image per subject, containing 1,476 thermal images and 1,476 visible images
equally split into two sets of modalities: face and head, respectively. The SIFT
features are not only invariant to image scale and rotation but also essential for
providing a robust matching across changes in illumination or addition of noise.
Descriptors extracted from local regions are hierarchically set in a vocabulary
tree using the k-means algorithm as clustering method. That provides a larger
and more discriminatory vocabulary, which leads to a performance improve-
ment. The verification quality is evaluated through a series of independent
experiments with various results, showing the power of the system, which satis-
factorily verifies the identity of the database subjects and overcoming limita-
tions such as dependency on illumination conditions and facial expressions. A
comparison between head and face verification is made for both ranges. This
approach has reached accuracy rates of 97.60% in thermal head images in
relation to 88.20% in thermal face verification. For visible range, 99.05% with
visible head images in relation to 97.65% in visible face verification. In this
proposal and after experiments, visible range gives better accuracy than thermal
range, and with independency of range, head images give the most discriminate
information.

Keywords: Thermal face verification, Visible face verification, Face detection,
Biometrics, SIFT parameters, Vocabulary tree, k-Means, Image processing,
Pattern recognition.

1 Introduction

Human recognition through distinctive facial features supported by an image database
is still an appropriate subject of study. We may not forget that this problem presents