Feature Extraction through Cross-Phase Congruency for Facial Expression Analysis

Human face analysis has attracted a large number of researchers from various fields, such as computer vision, image processing, neurophysiology or psychology. One of the particular aspects of human face analysis is encompassed by facial expression recognition task. A novel method based on phase congruency for extracting the facial features used in the facial expression classification procedure is developed. Considering a set of image samples comprising humans expressing various expressions, this new approach computes the phase congruency map between the samples. The analysis is performed in the frequency space where the similarity (or dissimilarity) between sample phases is measured to form discriminant features. The experiments were run using samples from two facial expression databases. To assess the method’s performance, the technique is compared to the state-of-the art techniques utilized for classifying facial expressions, such as Principal Component Analysis (PCA), Independent Component Analysis (ICA), Linear Discriminant Analysis (LDA), and Gabor jets. The features extracted by the aforementioned techniques are further classified using two classifiers: a distance-based classifier and a Support Vector Machine based classifier. Experiments reveal superior facial expression recognition performance for the proposed approach with respect to other techniques.