Prev: How to use my own function within a guide file?
Next: function to add an array to another array
From: Faris on 26 Apr 2010 18:36
I'm working on Face Recognition Project and I have many images with 1728x2592 size..

Now I wanna compress all these images to 30x32 using one loop in MATLAB to allow me to use them as inputs to Neural Network..

So, may anyone help me please ..??

Because I'm beginner in MATLAB..
From: ImageAnalyst on 26 Apr 2010 18:39
Faris:
How about using a for loop (over all images) and the function
imresize() to do the resizing?
From: Walter Roberson on 27 Apr 2010 14:16
Faris wrote:
> I'm working on Face Recognition Project and I have many images with
> 1728x2592 size..
>
> Now I wanna compress all these images to 30x32 using one loop in MATLAB
> to allow me to use them as inputs to Neural Network..

Are the entire large images of a single face? Or is there a face somewhere in
each image and you need to chop off the non-face data first?

Is it mandatory that what you feed in to the Neural Network be the compressed
face image? I don't know much about NNs, but when I see such a large input and
such a small bit of information to be put into the NN, what I think of is that
what you should probably be doing is creating feature vectors and training on
the feature vectors rather than on the images themselves.

Feature vectors would be measurements of some kind. They might include, for
example, wavelet coefficients, or they might include distances from (say) the
tip of the nose to each of several other key features (tip of the chin, middle
of each eye, whatever.) They might include measurements of (say) the
parameters of the superellipse that you used to fit the shape of the face.
http://en.wikipedia.org/wiki/Superellipse
From: Faris on 27 Apr 2010 16:07
ImageAnalyst <imageanalyst(a)mailinator.com> wrote in message <81c0351f-5de9-40da-ba29-df4ee2025f9f(a)o1g2000vbe.googlegroups.com>...
> Faris:
> How about using a for loop (over all images) and the function
> imresize() to do the resizing?

ImageAnalyst:

I've tried that , but it didn't work..
because these images in gray-scale and aren't processed yet..
From: Faris on 27 Apr 2010 16:17
Walter Roberson <roberson(a)hushmail.com> wrote in message <hr79m6$55c$1(a)canopus.cc.umanitoba.ca>...
> Faris wrote:
> > I'm working on Face Recognition Project and I have many images with
> > 1728x2592 size..
> >
> > Now I wanna compress all these images to 30x32 using one loop in MATLAB
> > to allow me to use them as inputs to Neural Network..
>
> Are the entire large images of a single face? Or is there a face somewhere in
> each image and you need to chop off the non-face data first?
>
> Is it mandatory that what you feed in to the Neural Network be the compressed
> face image? I don't know much about NNs, but when I see such a large input and
> such a small bit of information to be put into the NN, what I think of is that
> what you should probably be doing is creating feature vectors and training on
> the feature vectors rather than on the images themselves.
>
> Feature vectors would be measurements of some kind. They might include, for
> example, wavelet coefficients, or they might include distances from (say) the
> tip of the nose to each of several other key features (tip of the chin, middle
> of each eye, whatever.) They might include measurements of (say) the
> parameters of the superellipse that you used to fit the shape of the face.
> http://en.wikipedia.org/wiki/Superellipse


Walter:
I really appreciate your help..

these images don't contain the single face, they have to be chopped off..
because ANN needs a compressed image to reduce the number of inputs..
and about the features, they will be extracted later, because what I need now the compressing..
I found many things but I'm confused now..

Best Regards,,
 |  Next  |  Last
Pages: 1 2
Prev: How to use my own function within a guide file?
Next: function to add an array to another array