In order to reduce the dimension of high-dimensional data, raised edge semi-supervised marginal discriminant embedding and local preserving algorithm for dimensionality reduction is proposed. By minimizing the distance between sample and the center of its category, the local topology of samples is maintained in the projection subspace. And by maximizing the distance between the edges of different categories, the inter scatter of classes is increased in the projection subspace. Experimental results show that the dimensionality reduction algorithm of semi supervised marginal discriminant embedding and local preserving can get a better projection subspace of the initial feature space.