Abstract: Feature points matching of images is considered as a key process in digital protection of cultural heritage,how to decrease the outliers of the matching pairs is an important issue.The paper acquires initial matching pairs based on SIFT(Scale Invariant Feature Transform) descriptor and then triangulates the feature points,so each triangle is under non rigid transformation restricted by condition number.An energy function is defined to transform the feature matching problem into optimization problem which is solved by an iterative reweighting algorithm.It's shown by the results of experiment on terrcotta warrior data that the proposed method can decrease the ratio of outliers and increase matching accuracy.