Re: (image attached)how to segment the image like this?

Jerry Avins <[EMAIL PROTECTED]> wrote in message news:<[EMAIL PROTECTED]>...
> Robert W. Kuhn wrote:
> 
> > walala <[EMAIL PROTECTED]> wrote:
> > 
> > 
> >>That's a great idea! I will do that next time...
> > 
> > 
> > Please do it this time. Most of us have Newsservers that do not transfer
> > binaries.
> > 
> > 
> >>but could you please help me on this question as I am in need of it
> >>urgently? I just need to know based on what property/feature extraction I
> >>can segment the Lena image like shown in the image.
> > 
> > 
> > Than perhaps someone can answer this queston.
> > 
> > And please do not quote every time the whole posting.
> > 
> > Tschau - Robert
> 
> My news connection was broken for a time, so my posts didn't arrive. 
> Blocky Lena is at http://users.rcn.com/jyavins/walala-lena.jpg
> 
> Note to Walala:
> 
> Lena, who is not really a blockhead (she is still very attractive in
> advancing age), looks better interpolated. I divided each square into
> four smaller ones, and made a new image in which each new square is the
> weighted average of those around it. I used this set of weights:
> 
>       1 2 1
>       2 4 2
>       1 2 1
> 
> Divide the sum by 16, of course. Other weightings may be better: I 
> pulled these out of a hat.
> 
> Jerry

Are you trying to separate the girl from the background?  What's the
goal that you're trying to accomplish with this segmentation?  And why
did you scale up the image?

> Segmentation based on "non-homoguousnity"? Based on "grayness level" in the
> original lena image? Based on on "local smoothness"?

"Smoothness" of a texture is well correlated with its fractal
dimension.  The fractal dimension of a signal relates to the signal's
geometric complexity.  There are plenty of papers available on-line
that describe processes for estimating the fractal dimension of an
image.

Another thing you might want to look at is feature clustering methods,
such as described by the Fuzzy C-Means algorithm.  This algorithm does
a very good job defining clusters given several sets of feature
vectors.  Another approach is called K-Means clustering.  Again, there
are many papers on-line that you can find just by searching, but
here's a link to some Fuzzy-C code that you might find useful...

http://homepage.ntlworld.com/paul.mather/book.html

-- Brian Stone