Abstract: For the traditional linear discriminant analysis method,there are usually three questions:1) In order to ensure that the within-class scatter matrix is nonsingular,the principal component analysis must firstly is performed,which limits the effect of multidimensional space．2) If the number of training samples per person is single,the within-class scatter matrix is generally singular,and the method does not work．3)Without considering the partial correlation between pixels．To address these problems,this paper proposed a different linear discriminant analysis based on Laplacian orientations．The usage of the Laplacian orientations results in a more robust dissimilarity measures between images．The introduction of the difference scatter matrix avoids the singularity of the within-class scatter matrix．Experiments show that the proposed method has better robustness for facial expressions,illumination changes and different occlusions,and achieves a higher recognition rate．Especially for illumination changes,the effect is better.

Key words: Laplacian orientations,Dimensionality reduction,Linear discriminant analysis,Robust dissimilarity measures