The invention proposes a local linear migration and affine transformation based image feature matching method and system used for establishing correct matching by removing error matching in an initial matching point pair. The method comprises: establishing a model corresponding to geometric transformation of to-be-matched images for affine transformation of the to-be-matched images, establishing a model corresponding to a posterior probability of correct matching of the matching point pair, and solving model parameters based on nearest neighbor matching points, a least square method and an optimization method; and calculating the posterior probability of the correct matching of the initial matching point pair, and judging the correctness of the initial matching point pair according to a threshold. According to the method, the modeling is carried out for the affine transformation of the to-be-matched images, so that the error rate of matching is greatly reduced and good robustness is still kept even if large amounts of error matching exists in initial matching.