A face recognition algorithm based on contextual constraints generalized two-dimensional FLD

In this paper, an improved subspace learning method using contextual constraints based linear discriminant analysis (CCLDA) is proposed for face recognition. CCLDA first transforms an image matrix to a vector which causes huge dimensionality and computational complexity, and it may lead to small sample size problem. While, our new method combines the contextual constraints in images with G-2DFLD method, therefore, the size of the feature matrix is much smaller. Then it can reduce the computational complexity and avoid the singular within-class scatter matrix problem. Moreover, it fully makes use of the correlation of image pixels structure, which will provide useful information for classification. Experimental results obtained on ORL and XM2VTS databases show the effectiveness of the method.