[Numpy-discussion] [SciPy-User] Simple pattern recognition

[David Warde-Farley]

I think Zachary is right, ndimage does what you want:

In [48]: image = array(
[[0,0,0,1,1,0,0],
[0,0,0,1,1,1,0],
[0,0,0,1,0,0,0],
[0,0,0,0,0,0,0],
[0,1,0,0,0,0,0],
[0,1,1,0,0,0,0],
[0,0,0,0,1,1,0],
[0,0,0,0,1,1,1]])

In [57]: import scipy.ndimage as ndimage

In [58]: labels, num_found = ndimage.label(image)

In [59]: object_slices = ndimage.find_objects(labels)

In [60]: image[object_slices[0]]
Out[60]:
array([[1, 1, 0],
        [1, 1, 1],
        [1, 0, 0]])

In [61]: image[object_slices[1]]
Out[61]:
array([[1, 0],
        [1, 1]])

In [62]: image[object_slices[2]]
Out[62]:
array([[1, 1, 0],
        [1, 1, 1]])

David

On 21-Sep-09, at 2:04 PM, Gökhan Sever wrote:

> ndimage.label works differently than what I have done here.
>
> Later using find_objects you can get slices for row or column basis.  
> Not
> possible to construct a dynamical structure to find objects that are  
> in the
> in both axis.
>
> Could you look at the stackoverflow article once again and comment  
> back?
>
> Thanks.
>
> On Mon, Sep 21, 2009 at 12:57 PM, Zachary Pincus <zachary.pincus at yale.edu 
> >wrote:
>
>> I believe that pretty generic connected-component finding is already
>> available with scipy.ndimage.label, as David suggested at the
>> beginning of the thread...
>>
>> This function takes a binary array (e.g. zeros where the background
>> is, non-zero where foreground is) and outputs an array where each
>> connected component of non-background pixels has a unique non-zero
>> "label" value.
>>
>> ndimage.find_objects will then give slices (e.g. bounding boxes) for
>> each labeled object (or a subset of them as specified). There are  
>> also
>> a ton of statistics you can calculate based on the labeled objects --
>> look at the entire ndimage.measurements namespace.
>>
>> Zach
>>
>> On Sep 21, 2009, at 1:45 PM, Gökhan Sever wrote:
>>
>>> I asked this question at
>> http://stackoverflow.com/questions/1449139/simple-object-recognition
>>> and get lots of nice feedback, and finally I have managed to
>>> implement what I wanted.
>>>
>>> What I was looking for is named "connected component labelling or
>>> analysis" for my "connected component extraction"
>>>
>>> I have put the code (lab2.py) and the image (particles.png) under:
>>> http://code.google.com/p/ccnworks/source/browse/#svn/trunk/AtSc450/
>>> labs
>>>
>>> What do you think of improving that code and adding into scipy's
>>> ndimage library (like connected_components())  ?
>>>
>>> Comments and suggestions are welcome :)
>>>
>>>
>>> On Wed, Sep 16, 2009 at 7:22 PM, Gökhan Sever
>>> <gokhansever at gmail.com> wrote:
>>> Hello all,
>>>
>>> I want to be able to count predefined simple rectangle shapes on an
>>> image as shown like in this one:
>> http://img7.imageshack.us/img7/2327/particles.png
>>>
>>> Which is in my case to count all the blue pixels (they are ice-snow
>>> flake shadows in reality) in one of the column.
>>>
>>> What is the way to automate this task, which library or technique
>>> should I study to tackle it.
>>>
>>> Thanks.
>>>
>>> --
>>> Gökhan
>>>
>>>
>>>
>>> --
>>> Gökhan
>>> _______________________________________________
>>> SciPy-User mailing list
>>> SciPy-User at scipy.org
>>> http://mail.scipy.org/mailman/listinfo/scipy-user
>>
>> _______________________________________________
>> SciPy-User mailing list
>> SciPy-User at scipy.org
>> http://mail.scipy.org/mailman/listinfo/scipy-user
>>
>
>
>
> -- 
> Gökhan
> _______________________________________________
> SciPy-User mailing list
> SciPy-User at scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-user