Databases Reference
In-Depth Information
FIGURE 5.1: Pattern recognition processes using DHGN algorithm where a 7
x 5 bitmap of letter K is mapped as subpatterns over 7 hierarchically formed
GN sub-networks. (With kind permission from Springer Science+Business Me-
dia: AI 2008: Advances in Artificial Intelligence, “Single-Cycle Image Recog-
nition Using an Adaptive Granularity Associative Memory Network”, LNCS,
2008, 386-392, Muhamad Amin, A.H., and Khan, A.I., Fig.1, W. Wobcke and
M. Zhang (Eds.), http://dx.doi.org/10.1007/978-3-540-89378-3 39.)
subnet, we consider the number of neurons, n
gn
, required for a subpattern of
size s
sp
composed of v different element given by the following equation:
2
s
sp
+ 1
2
n
gn
= v
(5.1)
5.1.2.1
Network Generation
In order for the DHGN scheme to perform recognition on patterns, it must
first be generated. Network generation involves the construction of SI mod-
ule node and a collection of DHGN subnets. SI module node is a control
node, responsible for managing the inputs and outputs among the DHGN
subnets. The distribution of DHGN subnets within the network depends
on the pattern decomposition by the SI module. Given a pattern vector
P = {p
1
, p
2
, p
3
, . . . , p
m
} of size m, and subpattern length s
sp
. The number
of DHGN subnets n
sn
that needs to be generated is determined by Equation
5.2:
