numpy-discussion

[Numpy-discussion] adaptive thresholding: get adacent cells for each pixel

[stephen e. <ste...@gm...>]

I'm just starting with numpy (via scipy) and I'm wanting to perform
adaptive thresholding
(http://www.cee.hw.ac.uk/hipr/html/adpthrsh.html) on an image.
Basically that means that I need to get a threshold for each pixel by
examining the pixels around it. In numpy this translates to finding
the adjacent cells for each cell (not including the value of the cell
we are examining) and getting the mean, or median of those cells.

I've written something that works, but is terribly slow. How would
someone with more experience get the adjacent cells for each cell
minus the cell being examined?

Thanks
Stephen Emslie

Re: [Numpy-discussion] adaptive thresholding: get adacent cells for each pixel

[Filip W. <fi...@ft...>]

Hi,

> I'm just starting with numpy (via scipy) and I'm wanting to perform
> adaptive thresholding
> (http://www.cee.hw.ac.uk/hipr/html/adpthrsh.html) on an image.
> Basically that means that I need to get a threshold for each pixel by
> examining the pixels around it. In numpy this translates to finding
> the adjacent cells for each cell (not including the value of the cell
> we are examining) and getting the mean, or median of those cells.

> I've written something that works, but is terribly slow. How would
> someone with more experience get the adjacent cells for each cell
> minus the cell being examined?

You can get the mean value of surrounding cells by filtering.

import numpy
from scipy import signal
im = numpy.ones((10,10), dtype='d') * range(10)
fi = numpy.ones((3,3), dtype='d') / 8
fi[1,1]=0
print fi
#[[ 0.125  0.125  0.125]
# [ 0.125  0.     0.125]
# [ 0.125  0.125  0.125]]

signal.convolve2d(im, fi, mode='same', boundary='symm') # or correlate2d in this case


Also check help(signal.convolve2d) for information on various
parameters this function takes.


cheers,
fw

Re: [Numpy-discussion] adaptive thresholding: get adacent cells for each pixel

[Johannes L. <a.u...@gm...>]

Hi,

> I'm just starting with numpy (via scipy) and I'm wanting to perform
> adaptive thresholding
> (http://www.cee.hw.ac.uk/hipr/html/adpthrsh.html) on an image.
> Basically that means that I need to get a threshold for each pixel by
> examining the pixels around it. In numpy this translates to finding
> the adjacent cells for each cell (not including the value of the cell
> we are examining) and getting the mean, or median of those cells.
> 
> I've written something that works, but is terribly slow. How would
> someone with more experience get the adjacent cells for each cell
> minus the cell being examined?

regarding the mean value, you can take a look at
scipy.signal.convolve2d. If you convolve with an array like this:

[[0.125 0.125 0.125]
 [0.125 0.0   0.125]
 [0.125 0.125 0.125]]

you get the 3x3 mean value (btw why leave out the center pixel?).

For the median, I can not think of any good method right now.

Also another method springs to my mind (just substract the top row and
add a new bottom row to the averaging window), but I have no idea how to
do this in an efficient way.

Generally, always try to find a way to process the whole array as one.
If you perform anything on an array elementwise, it will be dead slow.


Best regards,
Johannes

Re: [Numpy-discussion] =?utf-8?q?adaptive_thresholding=3A_get_adacent?= =?utf-8?q?_cells_for=09each_pixel?=

[Alan G I. <ai...@am...>]

On Sat, 10 Jun 2006, stephen emslie apparently wrote: 
> I'm just starting with numpy (via scipy) and I'm wanting to perform 
> adaptive thresholding 
> (http://www.cee.hw.ac.uk/hipr/html/adpthrsh.html) on an image. 

The ability to define a function on a neighborhood,
where the neighborhood is defined by relative coordinates,
is useful other places too.  (E.g., agent based modeling.
Here the output should be a new array of the same dimension 
with each element replaced by the value of the function on
the neighborhood.) I am also interested in learning how 
people handle this.

Cheers,
Alan Isaac