ased13
January 26th, 2008, 01:42 PM
Can somebody help me with the matlab code for the following network?
I also attach a jpg with the stucture of the gabor wavelet network.
Any help on this is more than welcome!
The mapping function of a GWN has the following form:
gwn(x,y)=Σ[wi*go(x,y)]+gwn
where (x,y) denotes the input vector of the network,
gwn(x,y) refers to the network output, wi denotes the
network weight connecting the ith hidden node and the
output node, gi(●,●) denotes the transfer function of the
ith hidden node, N is the number of the hidden nodes and
gwn is introduced to ensure that the average value of the
objective function is zero.
The GWN adopts the following form as
its transfer function (i.e., the imaginary part of the Gabor wavelet):
g(x,y)=(1/2pi*sx*sy)*exp(-0.5*[(x'/sx)^2+(y'/sy)^2])*sin(2pi*v*x')
where (sx,sy) refers to the radial frequency bandwidths of
the Gabor wavelet, v denotes the central frequency, and (x’,y’)
are the (x,y) coordinates rotated by θ and translated
by (tx,ty) in accordance with the equation:
x’=cosθ(x-tx)-sinθ(y-ty)
y’=sinθ(x-tx)+cosθ(y-ty)
Assuming that IM(x,y) represents the
image being studied which is the objective function of the
GWN, the input vector (x,y) of the network is the position
of an image pixel in the image IM(x,y), and the network
output is the corresponding gray level of the image pixel.
The objective function of the network training process (the
network optimization) is defined as
E=min||IM(x,y)-Σ[wi*go(x,y)]-gwn||^2
I also attach a jpg with the stucture of the gabor wavelet network.
Any help on this is more than welcome!
The mapping function of a GWN has the following form:
gwn(x,y)=Σ[wi*go(x,y)]+gwn
where (x,y) denotes the input vector of the network,
gwn(x,y) refers to the network output, wi denotes the
network weight connecting the ith hidden node and the
output node, gi(●,●) denotes the transfer function of the
ith hidden node, N is the number of the hidden nodes and
gwn is introduced to ensure that the average value of the
objective function is zero.
The GWN adopts the following form as
its transfer function (i.e., the imaginary part of the Gabor wavelet):
g(x,y)=(1/2pi*sx*sy)*exp(-0.5*[(x'/sx)^2+(y'/sy)^2])*sin(2pi*v*x')
where (sx,sy) refers to the radial frequency bandwidths of
the Gabor wavelet, v denotes the central frequency, and (x’,y’)
are the (x,y) coordinates rotated by θ and translated
by (tx,ty) in accordance with the equation:
x’=cosθ(x-tx)-sinθ(y-ty)
y’=sinθ(x-tx)+cosθ(y-ty)
Assuming that IM(x,y) represents the
image being studied which is the objective function of the
GWN, the input vector (x,y) of the network is the position
of an image pixel in the image IM(x,y), and the network
output is the corresponding gray level of the image pixel.
The objective function of the network training process (the
network optimization) is defined as
E=min||IM(x,y)-Σ[wi*go(x,y)]-gwn||^2