Symmetric and transitive registration of image sequences.
Symmetric and transitive registration of image sequences.
复制标题
DOI：
10.1155/2008/686875
复制
发表时间：
2008
影响因子：
7.6
通讯作者：
Tagare H
中科院分区：
文献类型：
--
作者：
Skrinjar O;Bistoquet A;Tagare H
针对图像配准算法不具备对称性和传递性的问题，提出了一种构造对称传递性图像序列配准算法的方法。该方法既适用于刚体配准，也适用于非刚体配准，适用于线性或周期图像序列。对称性和传递性属性完全满足(直到机器精度)，也就是说，只要计算的变换是可逆的，它们总是保持不变，无论图像类型、质量和配准算法如何。这两个特性在运动跟踪应用中尤其重要，因为如果配准算法不是对称和可传递的，则可能获得物理上不正确的变形。使用两种不同的非刚性图像配准算法在两个心脏磁共振图像序列上测试了该方法。证明了当计算的变换是可逆的时，算法的对称性和传递性修改的传递性和对称性误差可以任意小，而未修改的算法的相应误差在像素大小的量级上。此外，对称和传递修正算法对两个图像序列的配准精度均高于未修正算法。
This paper presents a method for constructing symmetric and transitive algorithms for registration of image sequences from image registration algorithms that do not have these two properties. The method is applicable to both rigid and nonrigid registration and it can be used with linear or periodic image sequences. The symmetry and transitivity properties are satisfied exactly (up to the machine precision), that is, they always hold regardless of the image type, quality, and the registration algorithm as long as the computed transformations are invertable. These two properties are especially important in motion tracking applications since physically incorrect deformations might be obtained if the registration algorithm is not symmetric and transitive. The method was tested on two sequences of cardiac magnetic resonance images using two different nonrigid image registration algorithms. It was demonstrated that the transitivity and symmetry errors of the symmetric and transitive modification of the algorithms could be made arbitrary small when the computed transformations are invertable, whereas the corresponding errors for the nonmodified algorithms were on the order of the pixel size. Furthermore, the symmetric and transitive modification of the algorithms had higher registration accuracy than the nonmodified algorithms for both image sequences.