Biomedical Engineering Reference
In-Depth Information
5
×
Figure 9.
Eight possible orientations of a stick with a length of five (
5 masks).
that is chosen locally as a function of the magnitude gradient of the intensity
function in the image for estimating the image structure. The function
c
(
x, y, t
)=
f
(
|∇
I
(
x, y, t
)
|
)
(13)
is chosen to satisfy
f
(
z
)
→
0 when
z
→∞
, so that diffusion processing is stopped
near the edges. Therefore, the speckle noise within the homogeneous region can
be blurred and the region boundary can be preserved.
After anisotropic diffusion, the speckle noise has been reduced effectively.
The next step is to enhance the tumor boundary, and we employed edge detection
with the stick operation to do so. The edge detection problem can be modeled as
a line process, because boundaries between tissue layers will appear as all sorts
of lines in an ultrasound image (see Czerwinski et al. [39-41]). The stick is a set
of short line segments of variable orientation that can be used to approximate the
boundaries and enhance edge information. For a square area with size
N
×
N
in
an image, there are 2
N
−
2 short lines with a length of
N
pixels passing through
the center. The sums of the pixel values along the lines are calculated for each line
with different orientation. Then the largest of these sums is selected, and the value
at the center of the square is assigned as this maximum number. After each pixel in
an image is replaced by the maximum sum of the lines passing through the pixel,
the contrast at the edges is enhanced and speckle is also reduced in the resulting
image. Taking the length of five as an example, eight possible line segments are
illustrated in Figure 9. Continuing from Figure 8, the result with the stick function
is depicted in Figure 10.
