In this paper, a novel local invariant model based on scale invariant feature transform (SIFT) features is presented to accurately obtain and locate the local features of an image. After the local features of each image in the training set are extracted by the SIFT, we eliminate the unsteady factors in term of statistical results of all the SIFT features to establish the local invariant model. The experiments to evaluate the performance of the model are carried out, which prove that the method has the quality of high-repeatability and accuracy and achieves the power of accurately locating the similar objects in different scenes despite the rigid or non-rigid deformation on them. For further investigation, we combine the local invariant model with an active shape model for automatically initialization. Results show that the combined model achieves satisfactory performance.