Biomedical Engineering Reference
In-Depth Information
Fig. 9.3 Structure of the
Permutative Coding Neural
Classifier
Feature
Extractor
Encoder
Classifier
Fig. 9.4 Initial images
Fig. 9.5 Specific points
selected by the feature
extractor
where
br
ij
is the brightness of the pixel (
i
,
j
). If (
B
), then the pixel (
i
,
j
)
corresponds to the selected specific point on the image, where
B
is the threshold for
specific point selection.
Each feature is extracted from the rectangle of
h
l
w
size (Fig.
9.5
). The
p
positive
and the
n
negative points determine one feature and are randomly distributed in
the rectangle. Each point
P
rs
has the threshold
T
rs
that is randomly selected from
the range:
△
>
T
min
T
rs
T
max
(9.2)
where
s
stands for the feature number, and
r
stands for the point number.
The positive point is “active” if, in the initial image, it has brightness
b
rs
T
rs
:
(9.3)
