Biomedical Engineering Reference
In-Depth Information
Figure 6.6
(a) Image data for the index
z
=
3. (b) The 3D median filtered data at
z
=
3.
Detailed results are shown in Figures 6.5 to 6.7. A 3D median filter of size
5
5 voxels was used. Figure 6.6 shows a sequence of five slices taken at in-
creasing depth, indexed by increasing
z
values. In Figure 6.6(a) the image in the
2D plane at
z
×
5
×
3 is shown for the sake of comparison. Figure 6.6(b) contains the
result of applying the 3D median filter, where only the
z
=
3 plane is shown for
clarity. Figure 6.7 shows enlarged views around a circular neural structure in Fig-
ure 6.6. Note that the high-frequency noise has been filtered out. This is a useful
operation, because the subsequent stages of contour extraction rely on image gra-
dient information. Since the gradient involves differentiation, which is well known
to amplify noise, it is important to first apply a smoothing filter.
=
6.5.2 Contour Extraction
The next processing step is to extract contours corresponding to neural structures,
such as dendrites and axons. Though, there are many possible solutions to this
Figure 6.7
(a) Image data for the index
z
3, shown enlarged. (b) The 3D median filtered data
=
at
z
3, shown enlarged.
=
Search WWH ::
Custom Search