Thermal Image Analysis for Polygraph Testing

Date: 
12-01-2002

Polygraph testing is a standard security procedure favored by the U.S. government.
The objective of polygraph testing is to ascertain if the subject under investigation
truthfully or deceitfully answers the questions at hand. Specially trained psychologists structure the questions to maximize elicitation. During the testing three physiological parameters are monitored
closely:
a) blood volume and pulse changes
b) respiratory changes
c) electrodermal activity.
They are all recorded using invasive
methods and produce scalar values over
time (signals). Then, a scoring system is
used to quantify the subject’s response
and classify it as deceitful or truthful. An
extensive account of state-of-the-art polygraph
methods and systems can be found
in [3].

The success rate for polygraph testing
varies, but on average it is in the neighborhood
of 90%. The U.S. government is interested
in increasing this rate through the
use of additional information channels
[4]. One very promising channel is the use
of infrared facial image analysis. There
are several advantages to this method.
1) It is noninvasive. This is very important
in the context of polygraph testing
where it is crucial for the subject to feel as
comfortable as possible.
2) After appropriate processing the
thermal imagery can yield physiological
information similar to one of the traditional
polygraph channels; i.e., blood flow
rate. The major difference is that this information is now two-dimensional (2-D)
and not one-dimensional (1-D) across the
time line. This is at least an order of magnitude
more information than the traditional
channel can generate.

In this article we will describe our experimental
design of an infrared facial image
analysis system, the image acquisition
setup, the method we follow to convert
raw thermal data to blood flow rate data,
the scheme we use to classify deceptive
from nondeceptive subjects, and an analysis
of our experimental results. We will
also discuss our plan for future research.

Read the full article.  Click here.

Authors: 
Ioannis Pavlidis, and James Levine