Facial Expression Recognition Using Features of Density Distributions between Facial Images

Abstract

This paper presents a method of discriminating human facial expressions using 5 pattern classifiers, where features based on density distributions are extracted from target regions set in facial feature points. Observing human facial expressions in videos, we could find that wrinkles appear in regions of correlated mimic muscles. The proposed method extracts the degree of similarity based on Zero-Mean Normalized Cross-Correlation as features from target areas where wrinkles often appear. And, 5 pattern classifiers, namely, Nearest Neighbor, Random Forests, Logistic Regression, Support Vector Machine, and Neural Network, are applied to discrimination of 6 basic facial expressions. The efficiency of the proposed method has been confirmed using public facial expression databases.