Pauli Virtanen wrote:
Thu, 22 Jul 2010 00:47:20 -0400, Robin Kraft wrote: [clip]
Let's say the image looks like this: np.random.randint(0,2, 16).reshape(4,4)
array([[0, 0, 0, 1], [0, 0, 1, 1], [1, 1, 0, 0], [0, 0, 0, 0]])
I want to use a square, non-overlapping moving window for resampling, so that I get a count of all the 1's in each 2x2 window.
0, 0, 0, 1 0, 0, 1, 1 0 3 => 2 0 1, 1, 0, 0 0, 0, 0, 0
In another situation with similar data I'll need the average, or the maximum value, etc..
Non-overlapping windows can be done by reshaping:
x = np.array([[0, 0, 0, 1, 1, 1], [0, 0, 1, 1, 0, 0], [1, 1, 0, 0, 1, 1], [0, 0, 0, 0, 1, 1], [1, 0, 1, 0, 1, 1], [0, 0, 1, 0, 0, 0]])
y = x.reshape(3,2,3,2) y2 = y.sum(axis=3).sum(axis=1)
# -> array([[0, 3, 2], # [2, 0, 4], # [1, 2, 2]])
y2 = x.reshape(3,2,3,2).transpose(0,2,1,3).reshape(3,3,4).sum(axis=-1)
# -> array([[0, 3, 2], # [2, 0, 4], # [1, 2, 2]])
The above requires no copying of data, and should be relatively fast.
Actually, because of the use of reshape(3,3,4), your second example does make a copy. Warren
If you need overlapping windows, those can be emulated with strides:
http://mentat.za.net/numpy/scipy2009/stefanv_numpy_advanced.pdf http://conference.scipy.org/scipy2010/slides/tutorials /stefan_vd_walt_numpy_advanced.pdf