F I F T H P R O O F S “FAIR” 2009/10/8 page 1
i
i
i
i
i
i
i
i
Chapter 1
Introduction
1.1 Image Registration
Image registration is one of the challenging problems in image processing. Given
two images taken, for example, at different times, from different devices or perspec-
tives, the goal is to determine a reasonable transformation, such that a transformed
version of the first image is similar to the second one. A simplified registration ex-
ample is illustrated in Figure 1.1. Given are two three-dimensional (3D) magnetic
resonance scans of a human knee taken at two different times. The objective is to
compensate image differences introduced by the different poses of the knee. Fig-
ure 1.1 also displays two corresponding slices of the image volumes: the template
overlaid with a regular grid, and the difference before and after registration.
There is a large number of application areas which demand registration, in-
cluding art, astronomy, astrophysics, biology, chemistry, criminology, genetics, phys-
ics, and basically any area involving imaging techniques. More specific examples
include remote sensing (generating a global picture from different partial views), se-
curity (comparing current images with a data base), robotics (tracking of objects),
and, in particular, medicine, where computational anatomy, computer-aided diagno-
sis, fusion of different modalities, intervention and treatment planning, monitoring
of diseases, motion correction, radiation therapy, or treatment verification demand
registration. Since imaging techniques, such as computer tomography (CT), mag-
netic resonance imaging (MRI), positron emission tomography (PET), single-photon
emission computer tomography (SPECT), or ultrasound (US), have undergone re-
markable, fascinating, and ongoing improvements, in the last decade, a tremendous
increase in the utilization of the various modalities in medicine is taking place, and
more is to be expected for the future.
Unfortunately, no unified treatment or general theory for image registration
has been yet established. It appears that each application area has developed its
own approaches and implementations. Depending on the particular application,
the focus includes computing time (real-time applications in industrial inspection or
tracking), image features (remote sensing), memory (high-resolution 3D images) and
1
Copyright © by SIAM.
Unauthorized reproduction of this article is prohibited.
