A Comprehensive Empirical Study on Linear Subspace Methods for Facial Expression Analysis

Automatic facial expression analysis is a vital component of intelligent Human-Computer Interaction (HCI). In this paper, we present a extensive empirical study on linear subspace methods for facial expression analysis. Locality Preserving Projections (LPP) and Orthogonal Neighborhood Preserving Projections (ONPP) are first time applied to facial expression analysis. We systematically examine a number of linear subspace methods, and show that, in our comparative studies, the Supervised LPP (SLPP) is superior in supervised methods, while ONPP performs best in unsupervised learning.