关键词: 配准, 印鉴边缘, 尺度不变特征变换（SIFT）, 变换模型, 真伪识别

Abstract: Accurate registration is the key premise of accurate verification. A SIFT（Scale Invariant Feature Transform） feature based registration algorithm is presented to prepare for the seal verification, especially for the verification of high quality counterfeit sample seals. The similarities and the spatial relationships between the matched SIFT features are combined for the seal image registration. SIFT features extracted from the edge images of the binary model seal and sample seal images are matched according to their similarities. False matches are eliminated according to their position relationship. The homography between the model seal and the sample seal is constructed by RANSAC. In experiments, 10 pairs of genuine seal imprints and 10 pairs of fake are tested by the presented registration method and other two classical methods. The average distance between non-overlapped edges is obtained to assess the accuracy of registration, and the maximum distance quantifies the biggest difference between two seal imprints. Experiment results show that the presented method can accomplish more accurate registration, and there is no limit to the seal shapes, stroke number and structures. The registration speed is doubled, compared to the registration method of using the SIFT in binary seal imprints.

Key words: registration, edge of seal imprint, Scale Invariant Feature Transform（SIFT）, homography, verification