[SciPy-User] down-sampling an array by averaging - vectorized form?
Sat Feb 11 13:23:32 CST 2012
I'd like to thank everyone for their responses- they were really helpful in
thinking about the problem. All the solutions people posted were faster
than my original brutish algorithm, but all had subtle differences as well.
None of them are 'vectorized' persay but all are more clever or effeicent
ways of getting at the same problem. I thought I'd write a couple of quick
Convolution with a kernel is a good idea (one that I should have thought of
since I've been working with various types of filtering recently) and works
quite quickly. Depending on the type and size of the kernel it yields
slightly different local averages at each point which can be sub-sampled to
yield a smaller array. I didn't play with too many types of kernels but
I'd say it's important to consider the nature of the kernel and the
relation of the subsampling frequency to the size of the kernel in order to
get the best results.
The binning followed by scalar division is wicked fast and yields results
that are very close to my original algorithm. The reshaping seems very
clever and I am going to read it more carefully to learn some lessons
there, I think.
The ndimage.zoom approach is a very general approach (and roughly as quick
as the others). As far as I can tell, that function uses spline
interpolation for zoom factors > 1, and I'm unsure how it deals with zoom
factors < 1. It might do nearest neighbor or something like that, I wasn't
able to quickly determine from glancing at the source. If anyone knows, it
would be cool to hear.
I think I'll probably go with binning for now- We'll be dealing with
hundreds of wide-field microscopy images (of this rough size) in every
experiment and speed is a factor.
Thanks again everyone!
On Sat, Feb 11, 2012 at 10:40, Tony Yu <email@example.com> wrote:
> On Fri, Feb 10, 2012 at 10:11 PM, Andrew Giessel <
> firstname.lastname@example.org> wrote:
>> Hello all,
>> I'm looking to down-sample an image by averaging. I quickly hacked up
>> the following code, which does exactly what I want, but the double loop is
>> slow (the images I'm working with are ~2000x2000 pixels). Is there a nice
>> way to vectorize this? A quick profile showed that most of the time is
>> spend averaging- perhaps there is a way to utilize np.sum or np.cumsum,
>> divide the whole array, and then take every so many pixels?
>> This method of down-sampling (spatial averaging) makes sense for the type
>> of data I'm using and yields good results, but I'm also open to
>> alternatives. Thanks in advance!
>> import numpy as np
>> def downsample(array, reduction):
>> """example call for 2fold size reduction: newImage =
>> downsample(image, 2)"""
>> newArray =
>> np.empty(array.shape/reduction, array.shape/reduction)
>> for x in range(newArray.shape):
>> for y in range(newArray.shape):
>> newArray[x,y] =
>> np.mean(array[x*reduction:((x+1)*reduction)-1, y*reduction:((y+1)*reduction)-1])
>> return newArray
> I think `scipy.ndimage.zoom` does what you want. Or actually, it does the
> opposite: your 2fold size reduction example would be
> >>> from scipy import ndimage
> >>> small_image = ndimage.zoom(image, 0.5)
> SciPy-User mailing list
Andrew Giessel, PhD
Department of Neurobiology, Harvard Medical School
220 Longwood Ave Boston, MA 02115
ph: 617.432.7971 email: email@example.com
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the SciPy-User