Orthogonal Discriminant Neighborhood Preserving Embedding for facial expression recognition
Shuai Liu, Qiuqi Ruan, Rongrong Ni · 2010
In this paper, a new manifold learning algorithm called Orthogonal Discriminant Neighborhood Preserving Embedding (ODNPE) is proposed for facial expression recognition. The ODNPE pursues orthogonal projections vectors to preserve the local manifold within same classes and keep the separability between different classes. The obtained orthogonal projections vectors can keep the metric structure of the manifold embedded in high dimensional space such that the intrinsic dimensions of the manifold can be well learned. Furthermore, we design a novel penalty graph to describe the separability between pair-wise different classes. The proposed algorithm is compared with some other algorithms on two facial expression databases, and the experimental results show its effectivity.