Facial expression recognition with subpattern-based approaches

Abstract

In this paper Principal Component Analysis (PCA), Subpattern-based PCA (spPCA) and Linear Discriminants Analysis (LDA) methods are used as feature extractors with the combination of the preprocessing techniques of histogram equalization and mean-and-variance normalization in order to nullify the effect of illumination changes which are known to significantly degrade recognition performance. The recognition performance of the holistic PCA, subpattern-based PCA and approach is compared with the performance of subpattern-based PCA and LDA in order to demonstrate the performance differences and similarities between these two types of approaches. To be consistent with the research of others, our work has been tested on three facial expression databases namely JAFFE FGnet and Cohn Kanade. Person-dependent and person-independent experiments are performed on these databases separately to represent the recognition performances of the holistic and subpattern-based approaches and LDA.