Databases Reference
In-Depth Information
FIGURE 5.4: Samples of binary character images.
messages are received, the actual recognition process is carried out. There are
two stages involved at this level.
1. All of the indices received from the DHGN subnets for original patterns
are stored in a two-dimensional vector matrix, S = {s
11
, s
12
, . . . , s
mn
}.
The width of the matrix is equivalent to the size of the pattern, i.e., m;
the height corresponds to the number of stored patterns, n.
2. The frequency of the indices is calculated for each test pattern. All of the
indices for the test pattern are stored in a vector, R = {r
1
, r
2
, . . . , r
m
}.
The width of the matrix is equivalent to the size of the pattern. If an
entry in vector R gives the list of indices as {1, 2, 2, 2, 1}, this indicates
that three subnets have given a recall result of pattern 2; two subnets
have given a recall result of pattern 1. Therefore, based on the voting
approach, the pattern will be recalled as pattern 2.
To describe the voting mechanism, a simple pattern recognition problem fol-
lows. Figure 5.4 shows four binary character images: A, E, U, and a distorted
version of A.
The binary patterns for characters A, E, U, and the distorted image A, P
A
,
P
E
, P
U
, and P
A
are shown below:
0
@
1
A
P
E
=
0
@
1
A
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
1 1 1 1 1
1 0 0 0 0
1 0 0 0 0
1 1 1 1 0
1 0 0 0 0
1 0 0 0 0
1 1 1 1 1
P
A
=
0
1
0
1
1 0 0 0 1
1 0 0 0 1
1 0 0 0 1
1 0 0 0 1
1 0 0 0 1
1 0 0 0 1
1 1 1 1 1
0 1 1 1 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
=
@
A
P
U
=
