Zheng Yuanjie, Hunter Allan A, Wu Jue, Wang Hongzhi, Gao Jianbin, Maguire Maureen G, Gee James C
PICSL, Department of Radiology, University of Pennsylvania, Philadelphia, PA, USA.
Inf Process Med Imaging. 2011;22:674-85. doi: 10.1007/978-3-642-22092-0_55.
In this paper, we address the problem of landmark matching based retinal image registration. Two major contributions render our registration algorithm distinguished from many previous methods. One is a novel landmark-matching formulation which enables not only a joint estimation of the correspondences and transformation model but also the optimization with linear programming. The other contribution lies in the introduction of a reinforced self-similarities descriptor in characterizing the local appearance of landmarks. Theoretical analysis and a series of preliminary experimental results show both the effectiveness of our optimization scheme and the high differentiating ability of our features.
在本文中，我们解决了基于地标匹配的视网膜图像配准问题。我们的配准算法有两个主要贡献，使其区别于许多先前的方法。一是一种新颖的地标匹配公式，它不仅能够联合估计对应关系和变换模型，还能通过线性规划进行优化。另一个贡献在于引入了一种增强的自相似性描述符来表征地标的局部外观。理论分析和一系列初步实验结果表明了我们优化方案的有效性以及我们特征的高区分能力。
