Human Facial Expression Recognition Using Region-based Motion Estimation

Abstract

Facial expressions are the indicators to a person’s emotional state and also the principle means of human non-verbal communication. If
computers would have the ability to perceive and respond to facial expressions, the human-computer interaction would become much more
spontaneous. Although several approaches have been proposed to recognize human facial expressions, none have gained universal
acceptability due to their time complexity and difficulty of implementation. This paper proposes a simple method based on region-based
motion estimation that can identify facial expressions with acceptable precision in real time. The proposed method identifies six regions of
the human face that incur significant visual change caused by the facial motion of expressions. The amount of motion in each region,
estimated by the amount of variation in intensity level, is observed and recorded. The results of the observation are used to describe a rulebased
classifier that identifies a given emotion as belonging to one of the six universal expressions. The performance of the proposed
method is tested on several participants, showing 76% accuracy in recognizing facial expressions.