A facial expression recognition method based on cubic spline interpolation and HOG features

摘要

Facial expression recognition plays important roles in many applications. This paper presents a new facial expression recognition method. The method uses the cubic spline interpolating coefficients of landmark points together with HOG (Histogram of Oriented Gradients) of selected areas as representing features, and uses support vector machine (SVM) to build classification models for facial expression recognition. The adoption of spline interpolating coefficients for geometrical representations reduces the dimensionality of feature vectors significantly while keeping the accuracy. Experiments also show that the integration of these two features at decision-level sometimes performs better than that of feature-level fusion with respect to facial expression recognition.