In order to realize the fast and accurate matching of bullet marks

a new method based on scaleinvariant feature transform（ SIFT） algorithm and an empirical trace matching method are proposed. The algorithm of SIFT is analyzed in detail and verified step by step

including feature point extraction

feature vector generation and initial matching based on European distance. Finally

the method of random sample consensus（ RANSAC） is used to purify and eliminate mismatched point pairs. On the basis of this

an empirical trace matching method is summarized

namely the method of dense feature points. It can extract the dense area of feature points and calculate the percentage of the total area of the breech face

which is used as the basis for matching. The experiment based on the recommended samples

shows that η is set to 12 %

and the matched and known unmatched bullet marks can be effectively distinguished.