GPSO versus neural network in facial emotion detection

Recently, we have proposed the Guided Particle Swarm Optimization (GPSO) algorithm as a novel approach in facial emotion recognition. GPSO was a modification to the Particle Swarm Optimization (PSO) algorithm, which is widely recognized as an efficient optimization algorithm with applicability in many areas. While the results we obtained from the real-time system that we developed based on the said algorithm were very good, the question that still remained was, how does this method compare with the more conventional classification approaches, such as neural network? With the aim of answering this question, we have now re-implemented our emotion recognition system using the Back Propagation Neural Network (BPNN). The BPNN used has 3 layers, consisting of the input layer of 20 neurons representing the x and y coordinates of same 10 Facial Points (FPs) used in our previous experiments; the output layer has 7 neurons representing the six basic emotions plus Neutral and a hidden layer of 20 neurons. The same data (video clips) of 20 subjects used in previous experiments were used, randomly partitioning the data in the ratio of 60-40 to train and test the network respectively. The results show that while the BPNN has its own merits in terms of speed of detection, the GPSO method performed better in accuracy of detection for all but one of the six basic emotions.