Databases Reference
In-Depth Information
TABLE 5.1: Recalled Indices Retrieved from the DHGN Subnets after Each
Pattern Input
DHGN Subnets
k
1
k
2
k
3
k
4
k
5
k
6
k
7
P
A
1
1
1
1
1
1
1
P
E
2
2
2
2
2
2
2
Patterns
P
U
3
3
3
3
1
1
2
P
A
4
1
1
1
1
1
1
Each pattern is decomposed into subpatterns and is sent to the DHGN
subnets for the first level recognition process. In this example, each character
pattern is decomposed into seven subpatterns; each subpattern represents a
row of binary values, as shown below:
0
1
P
A
= (00100)
P
A
= (01010)
P
A
= (10001)
P
A
= (11111)
P
A
= (10001)
P
A
= (10001)
P
A
= (10001)
0 0 1 0 0
0 1 0 1 0
1 0 0 0 1
1 1 1 1 1
1 0 0 0 1
1 0 0 0 1
1 0 0 0 1
@
A
→
P
A
=
The results of the recognition process from the subpattern level are sent
back to the SI module node. These results, in the form of recalled/new indices
for each subnet, sn, are received by the SI module node and are represented
by a voting matrix, V, shown in Table 5.1.
The results of the recognition processes show that when the character pat-
tern A is introduced, all subnets respond with index 1. This shows that all
subnets agree that this is a newly stored pattern. Similarly, when pattern E
is being introduced, all subnets give feedback with an increase in the index
value, i.e., index 2. Consequently, pattern U obtains various results from the
DHGN subnets. Four out of seven subnets produce a new index, two subnets
recall the index of pattern A, and one gave the index of pattern E. In this case,
the maximum number of recalled/new indices is chosen as the recalled/new
pattern. Similarly, for the distorted pattern A, the index recalled most often
is index 1, which correlates with pattern A. Therefore, pattern A is recalled.
Consider that P is an array of stored patterns, P = {p
1
, p
2
, p
3
, . . . , p
m
},
where m represents the number of patterns being stored. For any pattern p
x
to
be recalled, the maximum vote V
p
max
, is obtained using the following equation:
V
p
max
= arg max (w
x
) ,
x ∈ m
(5.8)