Databases Reference
In-Depth Information
TABLE 4.3: Bias Array Entries for All Active GN in the HGN Composition
Illustrated in Figure 4.3 When the Pattern “uvwxy” Is Introduced
Layer
Active GN
Bias Array Entries
U1
1(#, V2)
1(U1, W3)
2(Z1, W3)
V2
Base
W3
1(V2, X4)
1(W3, Z5)
2(W3, Y5)
X4
Y5
1(X4, #)
1(#, 1, 1)
2(#, 2, 1)
1V1
Middle
1(1, 1, 1)
2(2, 1, 2)
3(1, 1, 2)
1W2
1(1, 1, #)
2(1, 2, #)
1X3
1
2
3
Top
TW
nodes in a physical network. As mentioned previously, the patterns used in
the HGN recognition scheme must be odd-size patterns. Therefore, the base
layer of the HGN network must also fulfill this requirement. To analyze the
number of neurons required for an HGN network to conduct recognition on
patterns of size S, we use and extend the methods described in [52].
In HGN pattern recognition, the number of neurons required to process one-
dimensional patterns of size S = x comprising v different pattern elements,
n (x), is obtained from the following equation:
n (x) = vx + v (x−2) + v (x−4) + . . . + v
(
x−
2
)
n(x) = v
(x−2i)
(4.1)
i=0
2
x + 1
2
n (x) = v
For two-dimensional patterns of size S = x × y, the number of neurons
required, n (x, y), is obtained as follows:
