Appearance Manifold of Facial Expression

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.

• Type: Conference
• Year: 2005
• Where: ICCV
• Authors: Caifeng Shan, Shaogang Gong, Peter W. McOwan
• Keywords: Appearance Manifold, Computer Vision, Facial Expression, ICCV 2005, Manifold