[Numpy-discussion] Simple pattern recognition

[Gökhan Sever]

On Thu, Sep 17, 2009 at 10:14 AM, Gökhan Sever <gokhansever at gmail.com>wrote:

>
> I use PIL to read my png file (after cropped the initial image to the
> column of my interest) Like:
>
> from PIL import Image
> myim = Image('seccol.png)
> imdata = np.array(myim.getdata())
>
> From this on, I am not sure what to provide to the structure parameter
>
> label(input, structure=None, output=None)
>
> I am correct on PIL usage or just a simple binary ready work, too?
>
>
Confirming myself :) This is the right approach to take. However this is
only the beginning.


>
>
>> Once you've run label() you can use scipy.ndimage.find_objects() to
>> get slices into the entire column that contain the contiguous region.
>>
>> If you want to try and count individual rectangles that may overlap,
>> there are likely dynamic programming algorithms that can find the
>> biggest rectangles. The truth is you can probably even do something
>> pretty naive and then compile it with Cython and it'll go blazing fast.
>>
>
I used find_object() correctly as well. Again, this query turns to not so
simple object counting :) Because I should be able identify complex objects
in that labeled array (ie. whether a single shadow or a part of compound
object; look up, down, right, left neighbourhoods of each object to see it
is by itself or not). And after such identification, I should get the
maximum height and widths of each object. This is somewhat easier but the
identification seems like a bit challenging for me. (many loops and some
intelligent logic needs to applied). I am aware this is a very good
algorithm design exercise, however I would be happy to hear if any similar
work has been done before, since this is just the tip of my lab report.

Thanks again.

-- 
Gökhan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/numpy-discussion/attachments/20090917/9dcd40f5/attachment.html>