Media Releases

Computers spot false faces better than people

March 21, 2014

TORONTO, ON — A joint study by researchers at the Uni­ver­si­ty of Cal­i­for­nia San Diego and the Uni­ver­si­ty of Toron­to has found that a com­put­er sys­tem spots real or faked expres­sions of pain more accu­rate­ly than peo­ple can. The work, titled “Auto­mat­ic Decod­ing of Decep­tive Pain Expres­sions,” is pub­lished in the lat­est issue of Cur­rent Biol­o­gy.

“The com­put­er sys­tem man­aged to detect dis­tinc­tive dynam­ic fea­tures of facial expres­sions that peo­ple missed,” said Mar­i­an Bartlett, research pro­fes­sor at UC San Diego’s Insti­tute for Neur­al Com­pu­ta­tion and lead author of the study. “Human observers just aren’t very good at telling real from faked expres­sions of pain.”

Senior author Kang Lee, pro­fes­sor at the Dr. Eric Jack­man Insti­tute of Child Study at the Uni­ver­si­ty of Toron­to, said “humans can sim­u­late facial expres­sions and fake emo­tions well enough to deceive most observers. The computer’s pat­tern-recog­ni­tion abil­i­ties prove bet­ter at telling whether pain is real or faked.”

The research team found that humans could not dis­crim­i­nate real from faked expres­sions of pain bet­ter than ran­dom chance – and, even after train­ing, only improved accu­ra­cy to a mod­est 55 per­cent. The com­put­er sys­tem attains an 85 per­cent accu­ra­cy.

“In high­ly social species such as humans,” said 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.”

“By reveal­ing the dynam­ics of facial action through machine vision sys­tems,” said Bartlett, “our approach has the poten­tial to elu­ci­date ‘behav­ioral fin­ger­prints’ of the neur­al-con­trol sys­tems involved in emo­tion­al sig­nalling.

The sin­gle most pre­dic­tive fea­ture of fal­si­fied expres­sions, the study shows, is the mouth, and how and when it opens. Fak­ers’ mouths open with less vari­a­tion and too reg­u­lar­ly.

“Fur­ther inves­ti­ga­tions,” said the researchers, “will explore whether over-reg­u­lar­i­ty is a gen­er­al fea­ture of fake expres­sions.”

In addi­tion to detect­ing pain malin­ger­ing, the com­put­er-vision sys­tem might be used to detect oth­er real-world decep­tive actions in the realms of home­land secu­ri­ty, psy­chopathol­o­gy, job screen­ing, med­i­cine, and law, said Bartlett.

“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,” she said. “In addi­tion, our com­put­er-vision sys­tem 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.”

To read about the study in Cur­rent Biol­o­gy, vis­it:‑X.


For more infor­ma­tion, con­tact:

Kang Lee
Uni­ver­si­ty Dis­tin­guished Pro­fes­sor
Dr. Eric Jack­man Insti­tute of Child Study,
Uni­ver­si­ty of Toron­to
*** NOTE: Prof. Lee is out of the coun­try but is avail­able for inter­views via e‑mail***

Dominic Ali
Media Rela­tions Offi­cer
Uni­ver­si­ty of Toron­to
Tel: 416–978-6974