SIFT is a commonly used matching algorithm, which is commonly used in image copy paste tampering detection. Based on the SIFT algorithm, the matching point pairs are obtained by the generalized 2NN test in the copy paste tampering detection method, there is an error match to produce the wrong match points. Therefore, this paper proposes a method to purify the matching points obtained from the generalized 2NN test by using the structural similarity between the matching points. The error matching points are eliminated and the matching accuracy is improved. Because the error matching point pairs are eliminated, the matching points are reduced. The time of clustering and geometrical assessment of subsequent operation is reduced, thereby the efficient execution of the algorithm is improved. The efficiency of the implementation of the entire algorithm is improved. Finally, the experiment shows that the improved algorithm has improved the performance