Abstract: This paper presents a novel algorithm to design the descriptor of image feature points based on locality preserving projections (LPP). Firstly, an eigenmatrix was pre-produced by LPP, and a high-dimensional gradient vector was constructed by the gradient vectors of all neighborhood points around feature points. Then, the high-dimensional gradient vector was embedded into a lower dimensional manifold space with the eigenmatrix, and a low-dimensional descriptor of the feature points was generated. The proposed algorithm can preserve invariability on the geometric structure: the eigenvectors which are neighboring each other in the original space will maintain the same attribute in low-dimensional space; on the contrary, the unsimilar eigenvectors become apart farther each other. Therefore, the description generated by our algorithm can show the interrelationship between features and has strong robustness. Moreover, the comparative experiments illustrated that the proposed algorithm is more rapid and accurate than SIFT and PCA-SIFT.