Automatic deception detection system based on hybrid feature extraction techniques

Abstract

Human face is considered as a rich source of non-verbal features. These features have proven their efficiency, so they are used by the deception detection system (DDS) to distinguish liar from innocent subjects. The suggested DDS utilized three kinds of features, these are facial expressions, head movements and eye gaze. Facial expressions are simply encoded and represented in the form of action units (AUs) based on facial action coding system (FACS). Head movements are represented based on both transitions and rotation. For eye gaze features, the eye gaze directional angle in both x-axis and y-axis are extracted. The collected database used to prove validity and robustness of the suggested system contains videos for 102 subjects from both genders with age range 18-55 years. The detection accuracy of the suggested DDS based on applying the logistic regression classifier is equal to 88.0631%. The proposed system has proven its robustness and the achievement of the highest detection accuracy when compared with previously designed systems.