Abstract: A new mosaic method is proposed using Scale Invariant Feature Transform （SIFT） and Otsu matching for single-mode fundus images. SIFT features are extracted for fundus images to overcome the influences of uneven illumination. Otsu algorithm is utilized to remove false matching points for improving the accuracy of feature matching. Then, the matrix of affine transformation is calculated among the matching points to complete image registration by space transformation. Finally, the registrated images are fused. The experiment results show that the proposed method is robust and can achieve high precision automatic single-mode fundus images mosaic.

Key words: color fundus image, image mosaic, Scale Invariant Feature Transform（SIFT）, Otsu matching, image fusion

摘要： 针对多幅单模彩色眼底图像的拼接问题，提出一种基于尺度不变特征变换（Scale Invariant Feature Transform，SIFT）与最大类间方差（Otsu）匹配的拼接方法。为克服光照不均对特征提取造成的影响，采用SIFT变换提取眼底图像特征点；利用Otsu剔除误匹配点，提高特征点的匹配精度；在此基础上，计算匹配点对之间的仿射变换矩阵，进行图像空间变换实现图像配准，并对配准图像进行融合。结果表明，提出的方法可实现对多幅单模彩色眼底图像的高精度自动拼接，具有很强的鲁棒性。

关键词: 彩色眼底图像, 图像拼接, SIFT特征, Otsu匹配, 图像融合