A novel feature extraction method based on sub-pattern Multi-directional two-dimensional linear discriminateanalysis (Sp-MD2DLDA) for face recognition is presented in this paper. In the proposed method, firstly, we apply directional2DLDA (D2DLDA) to extract features in some initial directions, and then choose the effective directions from theinitial directions for feature fusion after an evaluation. Secondly, divide the original images into small regions and applyD2DLDA to a set of partitioned sub-patterns to obtain features in the selected effective directions which complement eachother. Finally, fuse these complementary features and use nearest neighbor classifier for classification. Since the proposedmethod not only can extract local features and reduce the impact of the variations in expression and illumination by dividingthe original images into smaller sub-images, but also extract features in many more directions, we expect that it canimprove the recognition performance. The experimental results on Yale and ORL databases show that the proposed Sp-MD2DLDA method has better classification performance than that of the other related methods.
展开▼
