计算机工程

• 人工智能及识别技术 • 上一篇    下一篇

基于加速鲁棒特征的眼底图像配准改进方法

魏丽芳,林甲祥   

  1. (福建农林大学计算机与信息学院,福州 350002)
  • 收稿日期:2013-08-19 出版日期:2013-12-15 发布日期:2013-12-13
  • 作者简介:魏丽芳(1981-),女,讲师、博士,主研方向:数字图像处理与分析,模式识别;林甲祥,讲师、博士

Improved Fundus Image Registration Method Based on Speeded-up Robust Feature

WEI Li-fang, LIN Jia-xiang   

  1. (College of Computer and Information, Fujian Agriculture and Forestry University, Fuzhou 350002, China)
  • Received:2013-08-19 Online:2013-12-15 Published:2013-12-13

摘要: 为保证眼底图像配准精度,同时降低时间损耗,提出一种改进的基于加速鲁棒特征的眼底图像配准方法。该方法在提取眼底图像加速鲁棒特征的基础上,利用BBF算法和特征的方向特性和空间一致性检测得到初始匹配特征序列,并给出层次估计与模型选择技术相结合的方法,以求解图像之间的变换参数。通过进一步配准修正获得更好的变换参数。实验结果表明,该方法获得的配准精度均方根误差值均小于1,并能够在满足精度要求的同时提高效率。

关键词: 图像配准, 加速鲁棒特征, 特征提取, 特征匹配, 变换模型, 层次估计, 模型选择

Abstract: The improved method based on Speeded-up Robust Feature(SURF) is proposed to ensure the accuracy and reduce time loss of the fundus image registration. The method is based on SURF feature extraction. The BBF algorithm is used to match feature point, and the keypoints’ orientations and the geometrical size of matches are used to exclude the incorrect matches. The hierachical estimator and model selection are combined to calculate the transformation parameter. And the registration correction is done to yield better transformation parameter. Experimental results show that the mean square error value of registration accuracy is less than 1, and can meet the accuracy requirements and improve the processing speed.

Key words: image registration, Speeded-up Robust Feature(SURF), feature extraction, feature match, transformation model, hierachical estimation, model selection

中图分类号: