Palm vein recognition is a new biometric identification technology. The horizontal rotation, translation, tilting and loss of local vein information of palm vein image greatly affect recognition rate. To solve the above problems, this paper respectively extract four kinds of local invariant feature, Scale Invariant Feature Transform(SIFT), Affine-SIFT(ASIFT), Harris-Laplace and Maximally Stable Extremal Region(MSER), on palm vein images which occurred affine transformation. Then obtain the matching feature points by comparing the Euclidean distance. Through the comparison of the number of feature points and the matching feature points, ASIFT and Harris-Laplace algorithm have good resistance to affine transformation.