Biomedical Engineering Reference
In-Depth Information
(a)
(b)
Figure 18.
Illustration of the result of dual active contour model (a) initialization of the two
contours (shown in white); (b) result after convergence. Reprinted with permission from
[47]. Copyright
c
1997, IEEE.
lies in the background region, it should experience a contracting force. In this
particular situation, it is assumed that the object is entirely contained within the
background. Figure 18 illustrates some results using the dual active contour model,
where two contours attempt to integrate the information from a contour expanding
within a feature to a contour contracting from outside the feature [47]. Though
conceptually this is something the a-priori information is designed to accomplish,
this method in principle does not use any explicit form of prior statistical image
information to drive the snake. Object feature information can be used analogous
to how shape information has been incorporated within the snake framework. The
idea is to define the statistical distribution of the feature space from prior-known
segmented object-background data and define a regional force field using this
information.
The statistical snake proposed by Ivins and Porill [31] addressed this feature
by incorporating an energy term that generates a bidirectional pressure force de-
pending on a-priori information of the image data. A regional feature defines the
modified external energy over the area as follows:
E
region
=
−
G
(
f
(
x, y
))
dxdy,
(21)
R
where
G
is the function that measures the nature of the image data. For a unidi-
rectional snake with a balloon or like forces the value of
G
(
f
(
x, y
)) is set to +1
or -1 depending on whether the snake is expanding or contracting. In a statistical
snake, this measure will change the direction depending on the nature of
f
(
x, y
)
