BUFFALO, N.Y. — A joint study by researchers at the
University of California, San Diego, the University at Buffalo, and
the University of Toronto has found that a computer–vision
system can distinguish between real or faked expressions of pain
more accurately than can humans.
This ability has obvious uses for uncovering pain malingering
— fabricating or exaggerating the symptoms of pain for a
variety of motives — but the system also could be used to
detect deceptive actions in the realms of security,
psychopathology, job screening, medicine and law.
The study, “Automatic Decoding of Deceptive Pain
Expressions,” is published in the latest issue of Current
The authors are Marian Bartlett, PhD, research professor,
Institute for Neural Computation, University of California, San
Diego; Gwen C. Littlewort, PhD, co-director of the
institute’s Machine Perception Laboratory; Mark G. Frank,
PhD, professor of communication, University at Buffalo, and Kang
Lee, PhD, Dr. Erick Jackman Institute of Child Study, University of
The study employed two experiments with a total of 205 human
observers who were asked to assess the veracity of expressions of
pain in video clips of individuals, some of whom were being
subjected to the cold presser test in which a hand is immersed in
ice water to measure pain tolerance, and of others who were faking
their painful expressions.
“Human subjects could not discriminate real from faked
expressions of pain more frequently than would be expected by
chance,” Frank says. “Even after training, they were
accurate only 55 percent of the time. The computer system, however,
was accurate 85 percent of the time.”
Bartlett noted that the computer system “managed to detect
distinctive, dynamic features of facial expressions that people
missed. Human observers just aren’t very good at telling real
from faked expressions of pain.”
The researchers employed the computer expression recognition
toolbox (CERT), an end-to-end system for fully automated
facial-expression recognition that operates in real time. It was
developed by Bartlett, Littlewort, Frank and others to assess the
accuracy of machine versus human vision.
They found that machine vision was able to automatically
distinguish deceptive facial signals from genuine facial signals by
extracting information from spatiotemporal facial-expression
signals that humans either cannot or do not extract.
“In highly social species such as humans,” says Lee,
“faces have evolved to convey rich information, including
expressions of emotion and pain. And, because of the way our brains
are built, people can simulate emotions they’re not actually
experiencing so successfully that they fool other people. The
computer is much better at spotting the subtle differences between
involuntary and voluntary facial movements.”
Frank adds, “Our findings demonstrate that automated
systems like CERT may analyze the dynamics of facial behavior at
temporal resolutions previously not feasible using manual coding
Bartlet says this approach illuminates basic questions
pertaining to many social situations in which the behavioral
fingerprint of neural control systems may be relevant.
“As with causes of pain, these scenarios also generate
strong emotions, along with attempts to minimize, mask and fake
such emotions, which may involve ‘dual control’ of the
face,” Bartlett says.
“Dual control of the face means that the signal for our
spontaneous felt emotion expressions originate in different areas
in the brain than our deliberately posed emotion
expressions,” Frank explains, “and they proceed through
different motor systems that account for subtle appearance, and in
the case of this study, dynamic movement factors.”
The computer-vision system, Bartlett says, “can be applied
to detect states in which the human face may provide important
clues as to health, physiology, emotion or thought, such as
drivers’ expressions of sleepiness, students’
expressions of attention and comprehension of lectures, or
responses to treatment of affective disorders.”
The single most predictive feature of falsified expressions, the
study showed, is how and when the mouth opens and closes.
Fakers’ mouths open with less variation and too regularly.
The researchers say further investigations will explore whether
such over-regularity is a general feature of fake expressions.