Abstract: PURPOSE: The eyes, as the windows to the soul, can reflect many internal mental activities. Can the eyes be the windows to the lies? Research has found that the pupil can be used as a cue for deception detection. In addition to the pupil, there is another feature of the eyes that can leak out inner mental information - the Eye Aspect Ratio (EAR, which is equal to 0 when the eyes are closed)
METHODS: This paper presents a development of non-intrusive video-based method that uses computer-vision to measure eyes features for identifying visible signs of deception. By using video footage of players lying and telling the truth in the game show of Golden ball, the computer vision software OpenFace was used to analyze the features of the players' eyes in both cases. The pupil size and Eye Aspect Ratio were calculated. These obtained features are statistically analyzed and fed into the machine learning software WEKA for distinguishing lies from truth.
RESULTS: The findings showed that the eyes’ pupil size was not different between the lying and truth-telling conditions, and the accuracy of machine learning classification using pupil size as a feature was below 60%; The Eye Aspect Ratio was different between the lying and truth-telling conditions (the Eye Aspect Ratio was greater during lying than it is while telling the truth), and the performances machine learning classification using Eye Aspect Ratio were all above 75%.
CONCLUSIONS: Eyes can be the windows to the lies.

Key words: Eye Aspect Ratio, deception detection, computer vision, machine learning