This paper investigates the appearance manifold of facial expression: embedding image sequences of facial expression from the high dimensional appearance feature space to a low dimensional manifold. We explore Locality Preserving Projections (LPP) to learn expression manifolds from two kinds of feature space: raw image data and Local Binary Patterns (LBP). For manifolds of different subjects, we propose a novel alignment algorithm to define a global coordinate space, and align them on one generalized manifold. Extensive experiments on 96 subjects from the Cohn-Kanade database illustrate the effectiveness of the alignment algorithm. The proposed generalized appearance manifold provides a unified framework for automatic facial expression analysis. © Springer-Verlag Berlin Heidelberg 2005.
Shan, C., Gong, S., & McOwan, P. W. (2005). Computer Vision in Human-Computer Interaction. Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 3766, 221–230.