Detection of facial microexpressions

Abstract

Facial microexpressions are instantaneous features signaling various details regarding the emotional and mental state of human beings. A key property of such features is that their interpretation as signals is the same or closely similar for all people. Currently, their detection requires a human expert. The automation of this task would allow a more widespread use. In this paper, we propose a hybrid solution, which is based on a framework of landmark points identified by a machine learningbased method. Upon this, we designed an expert system which utilizes image processing and signal processing algorithms such as homomorphic filtering, RANSAC parabola fitting, Hessian based shape analysis and change detection in order to identify microexpression features such as gaze detection and eyebrow raising. We evaluate these algorithms in real videos and pictures, and examine their applicability in practical scenarios. Our longterm goal is to detect complex facial expressions and emotions with the help of the detected microexpressions.