Tensor Rank One Discriminant Locally Linear Embedding for facial expression classification
Shuai Liu, Qiuqi Ruan · 2010
In this paper we propose the Tensor Rank one Discriminant Locally Linear Embedding algorithm (TR1DLLE), which accept tensors as input for classification. TR1DLLE integrates the tensor rank one Analysis (TRIA) and a recently proposed graph embedding algorithm Discriminant Locally Linear Embedding (DLLE). The merits of TR1DLLE include: (1) representing data in their native structure without losing spatial locality information; (2) avoiding the curse of dimensionality and small sample size problems; (4) inheriting the excellent characters of DLLE about intraclass manifold preservation and interclass discrimination; (5) having better learning capacity especially when the size of the training sample is small; (6) converge well. In the experiments, we apply TR1DLLE to the facial expressions classification and compared it with the former related algorithms.