University of Houston

Lying is one of the ubiquitous human behaviors. While it may be true that not everyone lies, there are still many that systematically exercise deception at every opportune moment. Throughout history, deception has been used for selfish well being and ill got gain. People may lie to protect themselves and others, or may do it for the sheer thrill to see if they can get away with the act of deception. Many verbal and nonverbal behavior analysis techniques as well as polygraph measurements techniques are available to detect lies. The formal techniques, however, require labor intensive efforts to analyze a small portion of the visual data and the later techniques offer uncomfortable and obtrusive measurements.

In this thesis, an alternate approach of detecting deception is proposed. In particular, emotional perspiration responses are extracted from the peri-nasal region of the face through image processing techniques and then analyzed using wavelet, statistical and machine learning approaches. In contrast to the traditional polygraph measurement, the proposed approach offers unobtrusive measurement via thermal imagery as well as automatic detection of the subjects' deceptiveness. We tested the method on thermal videos of 67 subjects who faced stressful interrogation for a mock crime. Total of 25 subjects are used to train the machine learning algorithms and remaining 43 subjects are used for testing.

The results show that the proposed approach has achieved a success rate of 78.5% in blind predictions. This research has opened a novel way of detecting lie in unstructured interrogation such as Behavior Analysis Interviews.