Training Algorithm
The relation of an original pattern P and it sunpatterns p1, p2, ...p16
is defined by P=(p1,p2,..pi,..p16) and for each pi , pi=(pi1,pi2,..
pik,..pi16) , ?pik={0,1},
where pik is the graylevel of a pixel. The value 1 indicates
that the corresponding pixel is black, while the value 0 indicates that
the corresponding pixel is white. For each vector pi, there exists a weight
vector wi=(wi1,wi2,.. wik,..wi16)
and a threshold calculated
by the equations wik=2 pik - 1
. These weight vectors are produced for remembering the training patterns
in training phase. The threshold value
will inhibit the neuron‘s firing for all input vectors pjpi
.That is . However, it
is possible defining a reliable range in order to increase the tolerance
of noisy. By experience, we take .
The proposed neural networks are guaranteed to converge in a single
sweep. It provide a short time dealing with the training phase, the same
result was seen during the identification phase.
nextindexback