Abstract: To improve the accuracy of the traditional feature-based methods of fundus image registration，a new method is proposed.First，the feature matches of the pictures are extracted by SIFT（Scale Invariance Feature Transform）.Then，the quadratic transformation model which is fit for the fundus images is used to register the pictures.The parameters of the transformation are obtained by M-estimate.Experimental results show that the method is effective to both normal and non-normal fundus image registration.

Key words: Scale Invariance Feature Transform（SIFT）, quadratic transformation model, RANdom SAmple Consensus（RANSAC）, M-estimate

摘要： 针对传统的基于特征的眼底图像配准方法配准精度不高的问题，提出了一种新的眼底图像配准方法。通过具有仿射不变性的尺度不变特征变换（Scale Invariance Feature Transform，SIFT）方法提取待配准图像的特征点匹配对。采用适合眼底图像特点的曲面变换模型，实现图像的配准，变换模型参数通过M估计获得。实验结果表明，该算法提高了配准精度，对正常眼和非正常眼的眼底图像配准都是有效的。

关键词: 尺度不变特征变换（SIFT）, 曲面变换模型, 随机抽样一致性（RANSAC）, M估计