Biomedical Engineering Reference
In-Depth Information
Chapter 4
Stereo and Temporal Retinal Image
Registration by Mutual Information
Maximization
Xiao-Hong Zhu,
1
Cheng-Chang Lu,
2
and Yang-Ming Zhu
3
4.1
Introduction
Image registration is the process of determining a one-to-one mapping be-
tween the coordinates in one space and those in another such that points in
the two spaces that correspond to the same point are mapped to each other.
The mappings, which are also called transformation, are two dimensional (2D)
for 2D spaces and three dimensional (3D) for 3D spaces. The simplest exam-
ples are the rigid-body transformations, which are transformations in which
the distances among all points are preserved. The rigid-body transformations
are typically used to compensate for different imaging orientations of rigid ob-
jects. Other possible transformations include affine where parallel lines remain
parallel, projective that is from 3D to 2D, and warping which is nonlinear in
general.
The most prominent application of image registration is multimodality med-
ical image fusion. Different image modalities can provide different informa-
tion about the imaged organs, and most of the time this information is comple-
mentary in nature. For example, computer tomography (CT) image reveals the
anatomical structure of the organ, magnetic resonance (MR) image is capable of
