blocks|key|206052|text|如果您想重新采样，那么您应该查看Scipy的rebinning食谱。特别是，在末尾定义的congrid函数将支持反归或插值(相当于IDL中具有相同名称的函数)。如果你不想要插值，这应该是最快的选择。|type|unstyled|depth|inlineStyleRanges|offset|length|style|CODE|entityRanges|data|206053|也可以直接使用scipy.ndimage.map_coordinates，它将为任何类型的重采样(包括非结构化栅格)执行样条线插值。我发现对于大型数组(nx，ny+>+200)，map_coordinates速度很慢。|206054|对于结构化网格上的插值，我倾向于使用scipy.interpolate.RectBivariateSpline。您可以选择样条曲线的顺序(线性、二次、三次等)，甚至可以独立选择每个轴的顺序。举个例子：|206055|++++import+scipy.interpolate+as+interp
++++f+=+interp.RectBivariateSpline(x,+y,+im,+kx=1,+ky=1)
++++new_im+=+f(new_x,+new_y)|code-block|syntax|javascript|206056|在本例中，您正在进行双线性插值(kx+=+ky+=+1)。不支持“最近”类型的插值，因为它只支持矩形网格上的样条线插值。这也不是最快的方法。|206057|如果你想要双线性或双三次插值，通常做两个一维插值要快得多：|206058|++++f+=+interp.interp1d(y,+im,+kind='linear')
++++temp+=+f(new_y)
++++f+=+interp.interp1d(x,+temp.T,+kind='linear')
++++new_im+=+f(new_x).T|206059|您也可以使用kind='nearest'，但在这种情况下，请去掉横向数组。|206060|entityMap|0|LINK|mutability|MUTABLE|url|http://www.scipy.org/Cookbook/Rebinning|1|http://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.interpolation.map_coordinates.html|2|http://docs.scipy.org/doc/scipy/reference/generated/scipy.interpolate.RectBivariateSpline.html^0|18|7|M|9|0|0|7|T|7|T|1|0|I|11|I|11|2|0|0|F|D|0|0|0|6|E|0^^$0|@$1|2|3|4|5|6|7|18|8|@$9|19|A|1A|B|C]]|D|@$9|1B|A|1C|1|1D]]|E|$]]|$1|F|3|G|5|6|7|1E|8|@$9|1F|A|1G|B|C]]|D|@$9|1H|A|1I|1|1J]]|E|$]]|$1|H|3|I|5|6|7|1K|8|@$9|1L|A|1M|B|C]]|D|@$9|1N|A|1O|1|1P]]|E|$]]|$1|J|3|K|5|L|7|1Q|8|@]|D|@]|E|$M|N]]|$1|O|3|P|5|6|7|1R|8|@$9|1S|A|1T|B|C]]|D|@]|E|$]]|$1|Q|3|R|5|6|7|1U|8|@]|D|@]|E|$]]|$1|S|3|T|5|L|7|1V|8|@]|D|@]|E|$M|N]]|$1|U|3|V|5|6|7|1W|8|@$9|1X|A|1Y|B|C]]|D|@]|E|$]]|$1|W|3|-4|5|6|7|1Z|8|@]|D|@]|E|$]]]|X|$Y|$5|Z|10|11|E|$12|13]]|14|$5|Z|10|11|E|$12|15]]|16|$5|Z|10|11|E|$12|17]]]]

If you want to resample, then you should look at Scipy's cookbook for <a href="http://www.scipy.org/Cookbook/Rebinning">rebinning</a>. In particular, the <code>congrid</code> function defined at the end will support rebinning or interpolation (equivalent to the function in IDL with the same name). This should be the fastest option if you don't want interpolation.

You can also use directly <a href="http://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.interpolation.map_coordinates.html"><code>scipy.ndimage.map_coordinates</code></a>, which will do a spline interpolation for any kind of resampling (including unstructured grids). I find map_coordinates to be slow for large arrays (nx, ny > 200).

For interpolation on structured grids, I tend to use <a href="http://docs.scipy.org/doc/scipy/reference/generated/scipy.interpolate.RectBivariateSpline.html"><code>scipy.interpolate.RectBivariateSpline</code></a>. You can choose the order of the spline (linear, quadratic, cubic, etc) and even independently for each axis. An example:

<pre><code> import scipy.interpolate as interp
 f = interp.RectBivariateSpline(x, y, im, kx=1, ky=1)
 new_im = f(new_x, new_y)
</code></pre>

In this case you're doing a bi-linear interpolation <code>(kx = ky = 1)</code>. The 'nearest' kind of interpolation is not supported, as all this does is a spline interpolation over a rectangular mesh. It's also not the fastest method. 

If you're after bi-linear or bi-cubic interpolation, it is generally much faster to do two 1D interpolations:

<pre><code> f = interp.interp1d(y, im, kind='linear')
 temp = f(new_y)
 f = interp.interp1d(x, temp.T, kind='linear')
 new_im = f(new_x).T
</code></pre>

You can also use <code>kind='nearest'</code>, but in that case get rid of the transverse arrays.

blocks|key|3059525|text|你看过Scikit-image吗？它的transform.pyramid_*函数可能对您有用。|type|unstyled|depth|inlineStyleRanges|offset|length|style|CODE|entityRanges|data|3059526|entityMap|0|LINK|mutability|MUTABLE|url|http://scikit-image.org/docs/dev/api/api.html^0|J|J|3|C|0|0^^$0|@$1|2|3|4|5|6|7|N|8|@$9|O|A|P|B|C]]|D|@$9|Q|A|R|1|S]]|E|$]]|$1|F|3|-4|5|6|7|T|8|@]|D|@]|E|$]]]|G|$H|$5|I|J|K|E|$L|M]]]]

Have you looked at <a href="http://scikit-image.org/docs/dev/api/api.html" rel="noreferrer">Scikit-image</a>? Its <code>transform.pyramid_*</code> functions might be useful for you.

blocks|key|2846769|text|根据您的描述，您需要scipy.ndimage.zoom。|type|unstyled|depth|inlineStyleRanges|offset|length|style|CODE|entityRanges|data|2846770|双线性插值将为order=1，最近点为order=0，默认为立方(order=3)。|2846771|zoom专门用于要重新采样到新分辨率的规则网格数据。|2846772|举个简单的例子：|2846773|import+numpy+as+np
import+scipy.ndimage

x+=+np.arange(9).reshape(3,3)

print+'Original+array:'
print+x

print+'Resampled+by+a+factor+of+2+with+nearest+interpolation:'
print+scipy.ndimage.zoom(x,+2,+order=0)


print+'Resampled+by+a+factor+of+2+with+bilinear+interpolation:'
print+scipy.ndimage.zoom(x,+2,+order=1)


print+'Resampled+by+a+factor+of+2+with+cubic+interpolation:'
print+scipy.ndimage.zoom(x,+2,+order=3)|code-block|syntax|javascript|2846774|结果是：|2846775|Original+array:
[[0+1+2]
+[3+4+5]
+[6+7+8]]
Resampled+by+a+factor+of+2+with+nearest+interpolation:
[[0+0+1+1+2+2]
+[0+0+1+1+2+2]
+[3+3+4+4+5+5]
+[3+3+4+4+5+5]
+[6+6+7+7+8+8]
+[6+6+7+7+8+8]]
Resampled+by+a+factor+of+2+with+bilinear+interpolation:
[[0+0+1+1+2+2]
+[1+2+2+2+3+3]
+[2+3+3+4+4+4]
+[4+4+4+5+5+6]
+[5+5+6+6+6+7]
+[6+6+7+7+8+8]]
Resampled+by+a+factor+of+2+with+cubic+interpolation:
[[0+0+1+1+2+2]
+[1+1+1+2+2+3]
+[2+2+3+3+4+4]
+[4+4+5+5+6+6]
+[5+6+6+7+7+7]
+[6+6+7+7+8+8]]|2846776|2846777|编辑:正如Matt+S所指出的，缩放多波段图像有几个注意事项。我几乎是一字不差地从我的一个earlier+answers复制下面的部分|2846778|缩放也适用于3D+(和nD)阵列。但是，请注意，如果您缩放2倍，例如，您将沿所有轴缩放。|2846779|data+=+np.arange(27).reshape(3,3,3)
print+'Original:\n',+data
print+'Zoomed+by+2x+gives+an+array+of+shape:',+ndimage.zoom(data,+2).shape|2846780|这会产生以下结果：|2846781|Original:
[[[+0++1++2]
++[+3++4++5]
++[+6++7++8]]

+[[+9+10+11]
++[12+13+14]
++[15+16+17]]

+[[18+19+20]
++[21+22+23]
++[24+25+26]]]
Zoomed+by+2x+gives+an+array+of+shape:+(6,+6,+6)|2846782|在多波段图像的情况下，您通常不希望沿"z“轴插值，从而创建新的波段。|2846783|如果您有想要缩放的3波段RGB图像，可以通过指定一个元组序列作为缩放因子来执行此操作：|2846784|print+'Zoomed+by+2x+along+the+last+two+axes:'
print+ndimage.zoom(data,+(1,+2,+2))|2846785|2846786|Zoomed+by+2x+along+the+last+two+axes:
[[[+0++0++1++1++2++2]
++[+1++1++1++2++2++3]
++[+2++2++3++3++4++4]
++[+4++4++5++5++6++6]
++[+5++6++6++7++7++7]
++[+6++6++7++7++8++8]]

+[[+9++9+10+10+11+11]
++[10+10+10+11+11+12]
++[11+11+12+12+13+13]
++[13+13+14+14+15+15]
++[14+15+15+16+16+16]
++[15+15+16+16+17+17]]

+[[18+18+19+19+20+20]
++[19+19+19+20+20+21]
++[20+20+21+21+22+22]
++[22+22+23+23+24+24]
++[23+24+24+25+25+25]
++[24+24+25+25+26+26]]]|2846787|entityMap|0|LINK|mutability|MUTABLE|url|https://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.zoom.html|1|https://stackoverflow.com/a/16984081/325565^0|A|I|A|I|0|0|7|7|J|7|X|7|0|0|4|0|0|0|0|0|0|19|F|1|0|0|0|0|0|0|0|0|0|0^^$0|@$1|2|3|4|5|6|7|1O|8|@$9|1P|A|1Q|B|C]]|D|@$9|1R|A|1S|1|1T]]|E|$]]|$1|F|3|G|5|6|7|1U|8|@$9|1V|A|1W|B|C]|$9|1X|A|1Y|B|C]|$9|1Z|A|20|B|C]]|D|@]|E|$]]|$1|H|3|I|5|6|7|21|8|@$9|22|A|23|B|C]]|D|@]|E|$]]|$1|J|3|K|5|6|7|24|8|@]|D|@]|E|$]]|$1|L|3|M|5|N|7|25|8|@]|D|@]|E|$O|P]]|$1|Q|3|R|5|6|7|26|8|@]|D|@]|E|$]]|$1|S|3|T|5|N|7|27|8|@]|D|@]|E|$O|P]]|$1|U|3|-4|5|6|7|28|8|@]|D|@]|E|$]]|$1|V|3|W|5|6|7|29|8|@]|D|@$9|2A|A|2B|1|2C]]|E|$]]|$1|X|3|Y|5|6|7|2D|8|@]|D|@]|E|$]]|$1|Z|3|10|5|N|7|2E|8|@]|D|@]|E|$O|P]]|$1|11|3|12|5|6|7|2F|8|@]|D|@]|E|$]]|$1|13|3|14|5|N|7|2G|8|@]|D|@]|E|$O|P]]|$1|15|3|16|5|6|7|2H|8|@]|D|@]|E|$]]|$1|17|3|18|5|6|7|2I|8|@]|D|@]|E|$]]|$1|19|3|1A|5|N|7|2J|8|@]|D|@]|E|$O|P]]|$1|1B|3|12|5|6|7|2K|8|@]|D|@]|E|$]]|$1|1C|3|1D|5|N|7|2L|8|@]|D|@]|E|$O|P]]|$1|1E|3|-4|5|6|7|2M|8|@]|D|@]|E|$]]]|1F|$1G|$5|1H|1I|1J|E|$1K|1L]]|1M|$5|1H|1I|1J|E|$1K|1N]]]]

Based on your description, you want <a href="https://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.zoom.html" rel="nofollow noreferrer"><code>scipy.ndimage.zoom</code></a>.
Bilinear interpolation would be <code>order=1</code>, nearest is <code>order=0</code>, and cubic is the default (<code>order=3</code>).
<code>zoom</code> is specifically for regularly-gridded data that you want to resample to a new resolution.
As a quick example:
<pre><code>import numpy as np
import scipy.ndimage

x = np.arange(9).reshape(3,3)

print 'Original array:'
print x

print 'Resampled by a factor of 2 with nearest interpolation:'
print scipy.ndimage.zoom(x, 2, order=0)


print 'Resampled by a factor of 2 with bilinear interpolation:'
print scipy.ndimage.zoom(x, 2, order=1)


print 'Resampled by a factor of 2 with cubic interpolation:'
print scipy.ndimage.zoom(x, 2, order=3)
</code></pre>
And the result:
<pre><code>Original array:
[[0 1 2]
 [3 4 5]
 [6 7 8]]
Resampled by a factor of 2 with nearest interpolation:
[[0 0 1 1 2 2]
 [0 0 1 1 2 2]
 [3 3 4 4 5 5]
 [3 3 4 4 5 5]
 [6 6 7 7 8 8]
 [6 6 7 7 8 8]]
Resampled by a factor of 2 with bilinear interpolation:
[[0 0 1 1 2 2]
 [1 2 2 2 3 3]
 [2 3 3 4 4 4]
 [4 4 4 5 5 6]
 [5 5 6 6 6 7]
 [6 6 7 7 8 8]]
Resampled by a factor of 2 with cubic interpolation:
[[0 0 1 1 2 2]
 [1 1 1 2 2 3]
 [2 2 3 3 4 4]
 [4 4 5 5 6 6]
 [5 6 6 7 7 7]
 [6 6 7 7 8 8]]
</code></pre>
<hr />
Edit: As Matt S. pointed out, there are a couple of caveats for zooming multi-band images. I'm copying the portion below almost verbatim from one of my <a href="https://stackoverflow.com/a/16984081/325565">earlier answers</a>:
Zooming also works for 3D (and nD) arrays. However, be aware that if you zoom by 2x, for example, you'll zoom along all axes.
<pre><code>data = np.arange(27).reshape(3,3,3)
print 'Original:\n', data
print 'Zoomed by 2x gives an array of shape:', ndimage.zoom(data, 2).shape
</code></pre>
This yields:
<pre><code>Original:
[[[ 0 1 2]
 [ 3 4 5]
 [ 6 7 8]]

 [[ 9 10 11]
 [12 13 14]
 [15 16 17]]

 [[18 19 20]
 [21 22 23]
 [24 25 26]]]
Zoomed by 2x gives an array of shape: (6, 6, 6)
</code></pre>
In the case of multi-band images, you usually don't want to interpolate along the &quot;z&quot; axis, creating new bands.
If you have something like a 3-band, RGB image that you'd like to zoom, you can do this by specifying a sequence of tuples as the zoom factor:
<pre><code>print 'Zoomed by 2x along the last two axes:'
print ndimage.zoom(data, (1, 2, 2))
</code></pre>
This yields:
<pre><code>Zoomed by 2x along the last two axes:
[[[ 0 0 1 1 2 2]
 [ 1 1 1 2 2 3]
 [ 2 2 3 3 4 4]
 [ 4 4 5 5 6 6]
 [ 5 6 6 7 7 7]
 [ 6 6 7 7 8 8]]

 [[ 9 9 10 10 11 11]
 [10 10 10 11 11 12]
 [11 11 12 12 13 13]
 [13 13 14 14 15 15]
 [14 15 15 16 16 16]
 [15 15 16 16 17 17]]

 [[18 18 19 19 20 20]
 [19 19 19 20 20 21]
 [20 20 21 21 22 22]
 [22 22 23 23 24 24]
 [23 24 24 25 25 25]
 [24 24 25 25 26 26]]]
</code></pre>

blocks|key|2846910|text|我最近刚刚发现了scipy.ndimage.interpolation.zoom的一个问题，我已经以错误报告的形式提交了这个问题：https://github.com/scipy/scipy/issues/3203|type|unstyled|depth|inlineStyleRanges|entityRanges|offset|length|data|2846911|作为另一种选择(至少对我来说)，我发现scikit+image的skimage.transform.resize可以正常工作：http://scikit-image.org/docs/dev/api/skimage.transform.html#skimage.transform.resize|2846912|然而，它的工作方式与scipy的interpolation.zoom不同-不是指定一个乘法器，而是指定您想要的输出形状。这适用于2D和3D图像。|2846913|对于2D图像，可以像使用interpolation.zoom一样使用transform.rescale并指定倍增或比例。|2846914|entityMap|0|LINK|mutability|MUTABLE|url|https://github.com/scipy/scipy/issues/3203|1|http://scikit-image.org/docs/dev/api/skimage.transform.html#skimage.transform.resize^0|1T|16|0|0|1R|2C|1|0|0|0^^$0|@$1|2|3|4|5|6|7|T|8|@]|9|@$A|U|B|V|1|W]]|C|$]]|$1|D|3|E|5|6|7|X|8|@]|9|@$A|Y|B|Z|1|10]]|C|$]]|$1|F|3|G|5|6|7|11|8|@]|9|@]|C|$]]|$1|H|3|I|5|6|7|12|8|@]|9|@]|C|$]]|$1|J|3|-4|5|6|7|13|8|@]|9|@]|C|$]]]|K|$L|$5|M|N|O|C|$P|Q]]|R|$5|M|N|O|C|$P|S]]]]

I've recently just found an issue with scipy.ndimage.interpolation.zoom, which I've submitted as a bug report: <a href="https://github.com/scipy/scipy/issues/3203" rel="noreferrer">https://github.com/scipy/scipy/issues/3203</a>

As an alternative (or at least for me), I've found that scikit-image's skimage.transform.resize works correctly: <a href="http://scikit-image.org/docs/dev/api/skimage.transform.html#skimage.transform.resize" rel="noreferrer">http://scikit-image.org/docs/dev/api/skimage.transform.html#skimage.transform.resize</a>

However it works differently to scipy's interpolation.zoom - rather than specifying a mutliplier, you specify the the output shape that you want. This works for 2D and 3D images.

For just 2D images, you can use transform.rescale and specify a multiplier or scale as you would with interpolation.zoom.

blocks|key|3059711|text|此解决方案在不影响RGB通道的情况下缩放馈送图像的X和Y：|type|unstyled|depth|inlineStyleRanges|entityRanges|data|3059712|import+numpy+as+np
import+scipy.ndimage

matplotlib.pyplot.imshow(scipy.ndimage.zoom(image_np_array,+zoom+=+(7,7,1),+order+=+1))|code-block|syntax|javascript|3059713|希望这是有用的。|3059714|entityMap^0|0|0|0^^$0|@$1|2|3|4|5|6|7|K|8|@]|9|@]|A|$]]|$1|B|3|C|5|D|7|L|8|@]|9|@]|A|$E|F]]|$1|G|3|H|5|6|7|M|8|@]|9|@]|A|$]]|$1|I|3|-4|5|6|7|N|8|@]|9|@]|A|$]]]|J|$]]

This solution scales X and Y of the fed image without affecting RGB channels:

<pre class="lang-py prettyprint-override"><code>import numpy as np
import scipy.ndimage

matplotlib.pyplot.imshow(scipy.ndimage.zoom(image_np_array, zoom = (7,7,1), order = 1))
</code></pre>

Hope this is useful.

blocks|key|208881|text|您可以使用interpolate.interp2d。|type|unstyled|depth|inlineStyleRanges|offset|length|style|CODE|entityRanges|data|208882|例如，考虑由numpy数组arr表示的图像，您可以将其调整为任意高度和宽度，如下所示：|208883|W,+H+=+arr.shape[:2]
new_W,+new_H+=+(600,300)
xrange+=+lambda+x:+np.linspace(0,+1,+x)

f+=+interp2d(xrange(W),+xrange(H),+arr,+kind="linear")
new_arr+=+f(xrange(new_W),+xrange(new_H))|code-block|syntax|javascript|208884|当然，如果您的图像有多个通道，则必须对每个通道执行插值。|208885|entityMap|0|LINK|mutability|MUTABLE|url|https://docs.scipy.org/doc/scipy/reference/generated/scipy.interpolate.interp2d.html#scipy.interpolate.interp2d^0|5|K|5|K|0|0|D|3|0|0|0^^$0|@$1|2|3|4|5|6|7|W|8|@$9|X|A|Y|B|C]]|D|@$9|Z|A|10|1|11]]|E|$]]|$1|F|3|G|5|6|7|12|8|@$9|13|A|14|B|C]]|D|@]|E|$]]|$1|H|3|I|5|J|7|15|8|@]|D|@]|E|$K|L]]|$1|M|3|N|5|6|7|16|8|@]|D|@]|E|$]]|$1|O|3|-4|5|6|7|17|8|@]|D|@]|E|$]]]|P|$Q|$5|R|S|T|E|$U|V]]]]

You can use <a href="https://docs.scipy.org/doc/scipy/reference/generated/scipy.interpolate.interp2d.html#scipy.interpolate.interp2d" rel="nofollow noreferrer"><code>interpolate.interp2d</code></a>.

For example, considering an image represented by a numpy array <code>arr</code>, you can resize it to an arbitrary height and width as follows:

<pre><code>W, H = arr.shape[:2]
new_W, new_H = (600,300)
xrange = lambda x: np.linspace(0, 1, x)

f = interp2d(xrange(W), xrange(H), arr, kind="linear")
new_arr = f(xrange(new_W), xrange(new_H))
</code></pre>

Of course, if your image has multiple channels, you have to perform the interpolation for each one.

I am looking for how to resample a numpy array representing image data at a new size, preferably having a choice of the interpolation method (nearest, bilinear, etc.). I know there is

<pre><code>scipy.misc.imresize
</code></pre>

which does exactly this by wrapping PIL's resize function. The only problem is that since it uses PIL, the numpy array has to conform to image formats, giving me a maximum of 4 "color" channels.

I want to be able to resize arbitrary images, with any number of "color" channels. I was wondering if there is a simple way to do this in scipy/numpy, or if I need to roll my own.

I have two ideas for how to concoct one myself:

<ul>
<li>a function that runs <code>scipy.misc.imresize</code> on every channel separately</li>
<li>create my own using <code>scipy.ndimage.interpolation.affine_transform</code></li>
</ul>

The first one would probably be slow for large data, and the second one does not seem to offer any other interpolation method except splines.

Resampling a numpy array representing an image

大数据

我正在寻找如何对表示新大小的图像数据的numpy数组进行重采样，最好有插值方法的选择(最接近的，双线性等)。我知道有scipy.misc.imresize这正是通过包装PIL的resize函数来实现的。唯一的问题是，由于它使用了PIL，numpy数组必须符合图像格式，这给了我最多4个“颜色”通道。我希望能够调整任意大小...

问对表示图像的numpy数组进行重采样
EN

社区

活动

资源

关于

腾讯云开发者

热门产品

热门推荐

更多推荐

问对表示图像的numpy数组进行重采样EN

社区

活动

资源

关于

腾讯云开发者

热门产品

热门推荐

更多推荐

问对表示图像的numpy数组进行重采样
EN