High-Stakes Deception Detection Based on Facial Expressions

摘要: During a forensic interview, high-stakes deception is very prevalent notwithstanding the heavy consequences that might result. This paper proposes an automated computer vision solution for detecting based on facial clues. Four deceptive cues (eye-blink, eyebrow motion, wrinkle occurrence and mouth motion) were identified integrated into single behavior pattern vector discerning honesty. A Random Forest classifier was trained using unconstrained video database applied to classify patterns either or truthful categories. The labeled we created open sources such as YouTube. interview videos used training testing selected basis of criminal situations, murder kidnapping, which later verified by trials. Despite many uncontrolled factors (illumination, head pose occlusion) in videos, have achieved accuracy 76.92% when discriminating liars from truth-tellers. compares well with 80.9% [1], best extant accurary obtained experienced interrogators.