Facial expression recognition by combination of classifiers
Résumé
In this paper, we present a classifier fusion solution for automatic facial expression recognition. We represent our data using a sorted Principal Component Analysis, followed by a Linear Discriminant Analysis: the selection of principal components first performs a dimensionally reduction by improving discriminant capacities and then, a Linear Discriminant Analysis provides a class representation subspace where new samples can be classified. Using a fuzzy integral method, the classification is operated by combining, the outputs of three classifiers (using Mahalanobis distance, Euclidean distance and a Bayes rule based criterion). This method gives, for a new sample, a probabilistic interpretation of the different classifier outputs to generate a fuzzy measure vector for each considered facial expression class. The sample is then classified into class with maximum fuzzy posterior probability.