56
2: Raghuveer M. Rao, Manoj K. Arora
image. The widely used nearest neighbor resampling technique simply chooses
the gray value of that pixel of the original image whose center is closest to the
new pixel center. Other graphical interpolation techniques such as bilinear or
cubic interpolation may also be used (Richards and Jia 1999).
The above approach may also be used to register images acquired from
two different sensors or from two different sources or at two different times.
The process of image registration has widespread application in image fusion,
change detection and GIS modeling, and is described next.
2.4
Image Registration
The registration of two images is the process by which they are aligned so
that overlapping pixels correspond to the same entity that has been imaged.
Either feature based or intensity based image registration may be performed.
The feature based registration technique is based on manual selection of GCP
as described in the previous section. The feature-based technique is, however, laborious, time intensive and a complex task. Some automatic algorithms
have been developed to automate the selection of GCP to improve the efficiency. However, the extraction of GCP may still suffer from the fact that
sometimes too few a points will be selected, and further the extracted points
may be inaccurate and unevenly distributed over the image. Hence, intensity
based registration techniques may be more appropriate than the feature based
techniques.
We provide a basic overview of the principles of intensity based image registration in this chapter. An advanced intensity based technique using mutual
information as the similarity measure has been dealt in Chaps. 3 and 7.
The problem of intensity based registration of two images containing translation errors can be posed for two scalar-valued images It (m, n) and h(m, n) as
finding integers k and I such that II (m-k, n-l) is as close as possible to I 2 (m, n).
If the images are of the same size and the relative camera movement between
the two image-capture positions is mostly translational, then registration can
be attempted by shifting until a close match is obtained.
If the mean squared error is used as a criterion of fit, it is then equivalent to maximizing the cross-correlation between the two images, that is, k
and I are chosen to maximize the correlation between the images It and h
Finding the cross-correlation over all possible offsets is computationallyexpensive. Therefore, knowing approximately the range of offsets within which
best registration is achieved helps to keep the computational load small. In
many instances, a control point is chosen in the scene and a window of pixels
around the control point is extracted from the image to be registered and is
then correlated with the other image in the vicinity of the estimated location of
the control point in it. Instead of computing the correlation, one can also compute the sum of the least absolute differences between corresponding pixels of
the two images over the window. The window position for which the summed
difference is the smallest corresponds to the best translation.
2: Raghuveer M. Rao, Manoj K. Arora
image. The widely used nearest neighbor resampling technique simply chooses
the gray value of that pixel of the original image whose center is closest to the
new pixel center. Other graphical interpolation techniques such as bilinear or
cubic interpolation may also be used (Richards and Jia 1999).
The above approach may also be used to register images acquired from
two different sensors or from two different sources or at two different times.
The process of image registration has widespread application in image fusion,
change detection and GIS modeling, and is described next.
2.4
Image Registration
The registration of two images is the process by which they are aligned so
that overlapping pixels correspond to the same entity that has been imaged.
Either feature based or intensity based image registration may be performed.
The feature based registration technique is based on manual selection of GCP
as described in the previous section. The feature-based technique is, however, laborious, time intensive and a complex task. Some automatic algorithms
have been developed to automate the selection of GCP to improve the efficiency. However, the extraction of GCP may still suffer from the fact that
sometimes too few a points will be selected, and further the extracted points
may be inaccurate and unevenly distributed over the image. Hence, intensity
based registration techniques may be more appropriate than the feature based
techniques.
We provide a basic overview of the principles of intensity based image registration in this chapter. An advanced intensity based technique using mutual
information as the similarity measure has been dealt in Chaps. 3 and 7.
The problem of intensity based registration of two images containing translation errors can be posed for two scalar-valued images It (m, n) and h(m, n) as
finding integers k and I such that II (m-k, n-l) is as close as possible to I 2 (m, n).
If the images are of the same size and the relative camera movement between
the two image-capture positions is mostly translational, then registration can
be attempted by shifting until a close match is obtained.
If the mean squared error is used as a criterion of fit, it is then equivalent to maximizing the cross-correlation between the two images, that is, k
and I are chosen to maximize the correlation between the images It and h
Finding the cross-correlation over all possible offsets is computationallyexpensive. Therefore, knowing approximately the range of offsets within which
best registration is achieved helps to keep the computational load small. In
many instances, a control point is chosen in the scene and a window of pixels
around the control point is extracted from the image to be registered and is
then correlated with the other image in the vicinity of the estimated location of
the control point in it. Instead of computing the correlation, one can also compute the sum of the least absolute differences between corresponding pixels of
the two images over the window. The window position for which the summed
difference is the smallest corresponds to the best translation.