Face recognition based on Two-Dimensional Discriminant Locality Preserving Projection

Abstract

Locality Preserving Projection is a method which can extract the feature and reduce dimensionality effectively, which has been widely used in face recognition. However, it is also an unsupervised method, and it is an image vector-based method, needing to covert the face image into a vector. This conversion not only breaks the local structural information, but also brings lots of problems, such as the dimension of these converted vectors is too high and encounters the small sample size problem. And it is also an unsupervised method and has no directly relation to classification. In order to improve the performance of LPP, we present a method named Two-Dimensional Discriminant Locality Preserving Projection for extracting the feature and reduce dimensionality and apply it in face recognition. Experimental results on ORL and Yale databases suggest that the proposed 2DDLPP provides a better way to solve these problems and achieves lower error rates.