Design of "Personalized" Classifier Using Soft Computing Techniques for "Personalized" Facial Expression Recognition

Abstract

We propose a design method of personalized classifier with soft computing techniques for automatic facial expression recognition. Motivated by the fact that even though human facial expressions of emotion are often ambiguous and inconsistent, humans are, in general, very good at classifying such complex images. In consideration of individual characteristics, we adopt a similar strategy of building a personalized classifier to enhance the recognition performance. For realization, we use a soft computing technique of neurofuzzy approach. Specifically, two core steps-"model building/modification" and "feature selection"-are applied to build a "personalized" classification structure. The proposed scheme of classifier construction achieves a higher classification rate, minimal network parameters, easy-to-extend structure, and faster computation time, among others. Four sets of facial expression data are chosen and image features are extracted from each of them to show effectiveness of the proposed method, which confirms considerable enhancement of the whole performance.