GPSO versus GA in facial emotion detection

Abstract

We recently proposed a modification to the widely known Particle Swarm Optimization (PSO) algorithm so that it can be applied as a method for facial emotion recognition. We named our proposed modification to PSO as the Guided Particle Swarm Optimization (GPSO) algorithm. GPSO was used to implement a real-time facial emotion recognition software which was tested with 20 subjects of different ethnic backgrounds. The result was found to be good both in terms of recognition success rate (85% on the average) and recognition speed (31.58 frames per second). As a follow-up to this, we wanted to investigate how our novel (GPSO) approach compare with existing popular classification methods, such as Neural Network and Genetic Algorithm (GA). In this paper we report the results of our attempt to answer this question with respect to GA. We defined suitable GA objective functions and other GA elements and operators such as genes, chromosomes, crossover and mutation in terms of the emotion recognition problem and then used these to reimplement our emotion recognition software. The resulting software was tested using the video recordings of the same 20 subjects that were used to test the GPSO-based system. Our results show that while the recognition success rate using GA is still reasonable, the recognition speed is very slow, suggesting that the GA method may not be suitable for real-time emotion recognition applications.