Fusion of Visual and Thermal Face Recognition Techniques: A ...

Appearance-based approaches are commonly used for IR face recognition systems
[36]. In contrast to visual face recognition algorithms that mostly rely on the eye location,
thermal IR face recognition techniques present difficulties in locating the eyes. Initial
research approaches to thermal face recognition extracts and matches thermal contours
for identification. Such techniques include elemental shape matching and the Eigenface
method. Elemental shape matching techniques use the elemental shape of thermal face
images. Several different closed thermal contours can be observed in each face.
Variations in defining the thermal slices from one image to another has the effect of
shrinking or enlarging the resulting shapes while keeping the centroid location and other
features of the shapes constant. Perimeter, area, x and y coordinates of the centroid,
minimum and maximum chord length through the centroid and between perimeter points,
and standard deviation of that length are being considered. Automated face recognition
using elemental shapes in real time has reported 96% accuracy for cooperative access
control applications. A non-cooperative, non-real-time, faces-in-the-crowd version of
thermal face recognition achieved 98% accuracy with no false positives for more than
100 people represented in a database of 500 images [37]. Yoshitomi et al [38] also
proposed thermal face identification using Neural Networks. They classified based on
combining various information such as gray level histograms, mosaic images, and shape
factors.
1.4 Fusion Approaches to Face Recognition
The best classifier in the world, human beings, also do not rely on a single modality
to recognize objects if they are producing different information. Information fusion
11