Thread Subject: back propagation neural network

hi
please help me, how i do Implementation image recognition using back propagation?
Thanks

On Jan 10, 7:40 am, "kourosh " <mah24...@gmail.com> wrote:
> hi
> please help me, how i do Implementation image recognition using back propagation?
> Thanks

It depends on what toolboxes you have.

Have you searched the archives with combinations of words like

image recognition neural

?

Hope this helps.

Greg

On 1/10/2012 6:40 AM, kourosh wrote:
> hi
> please help me, how i do Implementation image recognition using back propagation?
> Thanks

I do not exactly understand the question, but if you mean how to reconstruct
an image from the forward projections? then you can use the inverse radon transform
for that. I have a matlab demo for that. If you mean something else,
then may be you can be more clear.

--Nasser

On Jan 10, 10:34 pm, "Nasser M. Abbasi" <n...@12000.org> wrote:
> On 1/10/2012 6:40 AM, kourosh wrote:
>
> > hi
> > please help me, how i do Implementation image recognition using back propagation?
> > Thanks
>
> I do not exactly understand the question, but if you mean how to reconstruct
> an image from the forward projections? then you can use the inverse radon transform
> for that. I have a matlab demo for that. If you mean something else,
> then may be you can be more clear.
>
> --Nasser

Typically, image recognition refers to assigning an input image to one
of c predefined
classes of images. For example, assigning noisy images of typed single
upper and
lower case letters and digits to one of 62 (26+26+10) classes.

Hope this helps.

Greg
one of 36 classes.

On 1/10/2012 9:43 PM, Greg Heath wrote:
> On Jan 10, 10:34 pm, "Nasser M. Abbasi"<n...@12000.org> wrote:
>> On 1/10/2012 6:40 AM, kourosh wrote:
>>
>>> hi
>>> please help me, how i do Implementation image recognition using back propagation?
>>> Thanks
>>
>> I do not exactly understand the question, but if you mean how to reconstruct
>> an image from the forward projections? then you can use the inverse radon transform
>> for that. I have a matlab demo for that. If you mean something else,
>> then may be you can be more clear.
>>
>> --Nasser
>
> Typically, image recognition refers to assigning an input image to one
> of c predefined
> classes of images. For example, assigning noisy images of typed single
> upper and
> lower case letters and digits to one of 62 (26+26+10) classes.
>
> Hope this helps.
>
> Greg

thanks. I think I do understand that. But what does backprojection
have to do with image recognition then?

backprojection has to do with construction of an image from projections,
as in x-ray and such.

image recognition has to do with to which class a specific given image
belongs.

I do not see the connection between the two fields. I can see
one first reconstruction an image using backprojections, and then
doing image recognition on the result.

By using more angles in the backprojection, one obtains a better
image. But still this has nothing to do with image recognition?

any way, I am not an expert on this area. Just took one course
on it once.

--Nasser

--Nasser

On Jan 10, 11:16 pm, "Nasser M. Abbasi" <n...@12000.org> wrote:
> On 1/10/2012 9:43 PM, Greg Heath wrote:
>
>
>
>
>
> > On Jan 10, 10:34 pm, "Nasser M. Abbasi"<n...@12000.org>  wrote:
> >> On 1/10/2012 6:40 AM, kourosh wrote:
>
> >>> hi
> >>> please help me, how i do Implementation image recognition using back propagation?
> >>> Thanks
>
> >> I do not exactly understand the question, but if you mean how to reconstruct
> >> an image from the forward projections? then you can use the inverse radon transform
> >> for that. I have a matlab demo for that. If you mean something else,
> >> then may be you can be more clear.
>
> >> --Nasser
>
> > Typically, image recognition refers to assigning an input image to one
> > of c predefined
> > classes of images. For example, assigning noisy images of typed single
> > upper and
> > lower case letters and digits to one of 62 (26+26+10) classes.
>
> > Hope this helps.
>
> > Greg
>
> thanks. I think I do understand that. But what does backprojection
> have to do with image recognition then?
>
> backprojection has to do with construction of an image from projections,
> as in x-ray and such.
>
> image recognition has to do with to which class a specific given image
> belongs.
>
> I do not see the connection between the two fields. I can see
> one first reconstruction an image using backprojections, and then
> doing image recognition on the result.
>
> By using more angles in the backprojection, one obtains a better
> image. But still this has nothing to do with  image recognition?
>
> any way, I am not an expert on this area. Just took one course
> on it once.
>
> --Nasser
>
> --Nasser- Hide quoted text -
>
> - Show quoted text -

The term backpropagation refers to a method for minimizing the error
of a neural
network that has hidden layers of nodes that cannot be "seen" at the
input and output.

h = tanh(W1*x+b1);
y = W2*h+b2;

x input node vector
h hidden node vector
y output node vector

W1,W2 unknown variable weight matrices
b1,b2 unknown bias weight vectors

Let t be the target output when x is applied. To estimate small
changes in (W1,b1) that
will tend to decrease the squared error e^2 = (t-y)^2, the hidden node
error should be known.
However, it can't be known because the correct value of h is not
known. The impasse is
side stepped by assuming that if e is small enough, the error in h is,
approximately,
proportional to e.

This assumption is the cornerstone of the method "Error
backpropagation".


Hope this helps.

Greg

On 1/11/2012 3:42 AM, Greg Heath wrote:

> The term backpropagation refers to a method for minimizing the error
> of a neural
> network that has hidden layers of nodes that cannot be "seen" at the
> input and output.
>

ok, thanks. I was thinking of something completely different.

This explains everything now ;)

--Nasser