Digital Signal Processing Reference
In-Depth Information
Fig. 2.
Basic PCNN neuron
In these equations,
S
ij
is the input stimulus, it is the normalized gray level of image
pixel in (
i, j
) position in image processing.
F
ij
[n]
is the feedback input of the neuron in
(
i, j
),
L
ij
[n]
is the linking item,
U
ij
[n]
is the internal activity of neuron,
T
ij
[n]
is the dy-
namic threshold.
Y
ij
[n]
is the pulse output of neuron with the binary value 0 or 1.
M
and
W
are the constant synaptic inter-connecting weight matrices for the feeding and
the linking input, which dependant on the distance between neurons. Generally,
M
and
W
refer to the Gaussian weight functions with the distance, normally
W=M
.
β
is the
linking coefficients; indicate the linking strength of PCNN.
α
F
,
α
L ,
α
T
are the attenua-
tion time constants of
F
ij
[n]
,
L
ij
[n], T
ij
[n]
, respectively.
V
F
,
V
L
,
V
T
denote the inherent
voltage potential of
F
ij
[n]
,
L
ij
[n], T
ij
[n]
, respectively.
For the feeding channel,
α
F
determines the decay rate of feeding channel. Larger
α
F
causes faster decay.
V
F
can enlarge or reduce the influence from surrounding neurons.
Matrix
W
refers to the mode of inter-connection among neurons in the feeding recep-
tive field. Generally, the size of
W
denotes the size of the feeding receptive field. The
value of matrix element
w
ijkl
determines the synaptic weight strength. In most cases,
this channel is simplified via
α
F
=0 and
V
F
=0.Different from the feeding channel, the
link channel usually keep itself as it is. The link channel also has three parameters (
α
L
,
V
L
, and
M
) that have the same function to the parameters (
α
F
,
V
F
, and
W
), respectively.
Usually, the inter-connection employs the Gaussian weight functions with the distance.
The linking coefficient
β
is an important parameter, because it can vary the weighting
of the linking channel in the internal activity. Hence, its value is usually depended on
different demands. For example, if much influence from the linking channel is ex-
pected,
β
should be given larger value. All neurons often have the same value of
β
.
3
Traffic Signs Feature Extraction
When the equations are running iteratively, the PCNN will output a sequence of binary
images. The output of PCNN varies in a period which is related with the properties of
the input image and the parameters of PCNN. The aim of our research is finding a
suitable parameter set and iteration times to produce an ideal segment binary image.
A 2-D image (
m×n
) can be thought as a PCNN neuromime with
m×n
neurons, and
the gray level of pixels is
,
the input of neuron as
S
ij
. One pixel's pulsating output
can activate other corresponding pixels having the approximate gray level in the
