DSPRelated.com
Forums

Re: image signatures

Started by Euphoria Stone February 4, 2006
Hi,

The best way to learn MATLB is to explore it and check
out its features.

Image signatures are basically a simple representation
of your image or object- rather than looking at a
object in 2 dimensions you can extract the important
structural information of the image and represent it
as a vector. This simplifies your problem. 

I've already sent a mail about image segmentation- you
can try that. Once you have segmented your object of
interest, you can apply a boundary extraction
algorithm to get the object boundaries. After getting
the boundary, just find those pixels that make up the
boundary. 

You can use the following code:

%------------
% Image segmentation code goes here...

% Code to get the perimeter pixels of the binarized 
% image bw
perim=bwperim(bw);

% Return in r and c the row and column coordinates of 
% the  boundary pixels
[r c]= find(bw==1);
%---------------

An alternative would be to thin the image- this will
give u the underlying skeleton of your object of
interest and is also a good approach to signatures.

Use the following code
%---------------
% Image segmentation code goes here...

% Code to get thinned version of binarized image bw

Bthin=bwmorph(bw,'thin',Inf);

% Return in r and c the row and column coordinates of 
% the  boundary pixels
[r c]= find(Bthin==1);
%----------------

The first approach will give you the object boundary
coordinates whereas the the 2nd approach will give you
the skeleton coordinates. The boundary using the first
approach is usually more representative of what your
object looks like, but skeletons are also very good at
describing the underlying structure.

E.S
	--- preeti aggarwal <pree_agg2002@pree...> wrote:

> Hello,
>     I am new to MATLAB. I want to generate the image
> signatures of an image. Signatures means the vector
> representation of image. for that these are the
> steps:
>   1) Image segmentation
>   2) Segment Isolation
>   3)Segment Comparison
>   4)Clustering
>   5) Vector Representation of those segments
>    
>   So if possible please tell me any of the above
> steps. Please help me.
>    
>   Regards,
	_____________