Classification of facial emotions using guided particle swarm optimization I

Abstract

This paper presents a novel approach to facial emotion detection using a modified Particle Swarm Optimization algorithm, which we called Guided Particle Swarm Optimization (GPSO). The approach involves tracking the movements of 10 Action Units (AUs) placed at appropriate points on the face of a subject and captured in video clips. Two dimensional rectangular domains are defined around each of the AUs and Particles are then defined to have a component in each domain, effectively creating a 10- dimensional search space within which particles fly in search of a solution. Since there are more than one possible target emotions at any point in time, multiple swarms are used, with each swarm having a specific emotion as its target. At each frame in the video clip, the solution of the swarm that is nearest to its target is accepted as the solution. Our results so far show the approach to have a promising success rate.