Virtual sample generation using concurrent-self-organizing maps and its application for facial expression recognition
Victor-Emil Neagoe, Adrian-Dumitru Ciotec · 2010
This paper is dedicated to the improvement of a pattern classifier generalization performances. One proposes the increasing of the training set size, by means of sample generation using a set of concurrent self-organizing maps (VSG-CSOM). We have evaluated the above proposed model for facial expression recognition. One uses Japanese female facial expression (JAFFE) database corresponding to seven emotion classes: happiness, sadness, surprise, anger, disgust, fear and neutral face. We have considered the following classifiers: nearest neighbour (NN), multilayer perceptron (MLP), and radial basis function (RBF) neural classifier. One obtains an obvious improvement in generalization performances for all the considered statistical/neural classifiers. For example, the recognition score evaluated on the test set as a consequence of virtual sample generation increases for the MLP from 67.14 % to 92.86 % and for the RBF from 87.14 % to 94.29 %.