Abstract: This paper proposes an image registration based on SIFT（Scale Invariant Feature Transform） and Krawtchouk moment invariants. Key points are extracted from images by applying SIFT. Then Krawtchouk moment invariants from the image region around key point are calculated, and these Krawtchouk moment invariants constitute feature vectors to describe the key point. Finally, key points are matched by calculating the Euclidean distance of feature vectors. The results of experiments show that the algorithm which has the same performance with the standard SIFT is more rapid than the standard SIFT.

Key words: image registration, Scale Invariant Feature Transform（SIFT）, feature description, Krawtchouk moment invariants

摘要： 提出了一种基于SIFT和Krawtchouk矩不变量的图像配准方法。通过SIFT关键点检测方法检测关键点；对每个关键点计算其邻域的Krawtchouk矩不变量，并将其构成描述关键点的特征向量；计算关键点特征向量之间的欧氏距离找出相匹配的关键点对。实验结果表明，该算法的配准性能与标准SIFT算法相当，而运算速度比标准SIFT算法有较大程度提高。

关键词: 图像配准, 尺度不变特征变换（SIFT）, 特征描述, Krawtchouk矩不变量