3D FACIAL LANDMARK-BASED DECEPTION DETECTION IN VIDEO USING GRU MODEL

Abstract

Deception detection is an interdisciplinary field that has researchers from psychology, criminology, and computer science. We propose the automated detection of deception based on facial micro expressions which occur spontaneously in response to the attempt to mask the inner emotion. It has received significant attention as an indicator of deceit, it reveals the genuine emotions that are concealed. In this paper, we first proposed a 3D 478 Mediapipe Face Mesh Model to extract facial landmarks that reflect facial micro expression, this is contrary to the traditional method, which relies on human judgment and the use of devices to detect facial micro expression. Second, a feature selection-based multivariate mutual information method was proposed to select facial landmarks that are most related to the deceptive cues and have critical influence on the classification task. Finally, a gated recurrent unit model was trained to predict deceptive behavior on a real-life trial dataset. The model successfully achieved 97% accuracy, outperforming other state-of-the-art methods.

Keywords

• facial micro expression
• facial action coding system
• mediapipe face mesh model
• feature selection-based multivariate mutual information method
• gated recurrent unit (gru) model