Painful loss in the human vs computer-vision battle

Share this on social media:

A computer-vision system has been found to be 30 per cent better than humans at identifying if an expression of pain is real or faked. This could be used to uncover pain fabrication within security, law, and medicine.

Human observers were asked to deduce the veracity of expressions of pain of individuals who were either undergoing cold presser tests -- which involves the hand of the individual being placed in ice water for a period of time -- or were told to fake painful expressions. In total 205 human observers were asked to decide whether the pain was real. The cold presser test is a common method to measure pain tolerance of a subject.

The research was published in the most recent edition of Current Biology. The authors were: Dr Marian Bartlett, research professor, Institute for Neural Computation, University of California, San Diego; Dr Gwen Littlewort, co-director of the institute’s Machine Perception Laboratory; Dr Mark Frank, professor of communication, University at Buffalo, and Dr Kang Lee, Dr Erick Jackman Institute of Child Study, University of Toronto.

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.

The research found that, even after training, the observers were only 55 per cent accurate at deciding whether the facial expression was a real or fake and before training were less successful than if the decisions had been left to chance. The vision system on the other hand made the correct decision 85 per cent of the time.

Bartlett commented: ‘[The 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.’ He continued to say that the 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.’

Recent News

02 May 2019

UK start-up, Photonic Vision, has developed what it considers a disruptive approach to time-of-flight sensing for lidar, able to reach high sensitivity without avalanche multiplication

02 May 2019

The University of Durham's Lucy Downes presented a terahertz imaging method 100-times faster than any previous terahertz imager at the UK KTN's Intelligent Imaging meeting in London on 29 April

09 April 2019

UK firm Winnow has launched an AI-enabled product incorporating vision that aims to help commercial kitchens reduce food waste

09 April 2019

Contract R&D organisation, Southwest Research Institute, has developed a vision solution that improves robot handling of shiny metallic objects, which it is showing at Automate in Chicago