Face recognition based on sub-pattern two-directional 2DPCA

To reduce the impacts on face recognition rate coursed by variations of pose, expression and illumination, a new face recognition approach based on sub-pattern two-directional 2DPCA is adopted. By dividing the original images into blocks, the approach can efficiently extract the local discriminant features of these images. At the same time, the two-directional 2DPCA ((2D)²PCA) method is used to extract the features on these blocks straightly, which avoids the conversion between matrix and vector. Therefore, the method can evaluate the eigenvectors of covariance matrix accurately and reduce the dimension of the discriminant features efficiently. Experimental results show that the sub-pattern two-directional 2DPCA has good performance when pose, lighting condition and face expression change.