Biomedical Engineering Reference
In-Depth Information
0.14
SD
0.12
SD
0.1
0.08
RCE
CE
CE
RCE
0.06
0.04
0.02
0
−
15
−
10
−
5
0
5
10
15
Z Translation (mm)
Figure 10.16: The cross-entropy (CE), reversed cross-entropy (RCE), and sym-
metric divergence (SD) profiles (with a likely priori) of an MR/Em image
pair in the vicinity of a manual registration when the z-translation changes
independently.
image registration, which quantify the difference between the true joint dis-
tribution and its prior estimation of the voxel value pair in two images at corre-
sponding points. The experimental results indicate that all three measures are
suitable as registration criteria.
The cross-entropy, reversed cross-entropy, and symmetric divergence can be
maximized and minimized, which is an unusual property for similarity measures.
Almost all other similarity measures can be either maximized or minimized, but
not both. If a likely priori pdf is given, we want our estimated pdf at registra-
tion to be as close to the priori as possible. Therefore, the similarity measures
are minimized. If an unlikely priori pdf is given, we want our estimated pdf at
registration to be different from the priori as much as possible. Thus, those simi-
larity measures are maximized. There are many unlikely priori pdf. This chapter
only considered the most popular one, i.e., the voxel values in two images at
