How can I render a Matplotlib Axes object to an image (as a Numpy array)?

Question

Is there a way to render the contents of a particular Axes object to an image, as a Numpy array? I know you can do this with the entire figure, but I want to get the image of a particular Axes.

The Axes I'm trying to render contains an image (drawn with imshow), with some lines plotted on top. Ideally, the rendered ndarray would contain just these elements, and no tick marks, borders, etc.

Update:

I forgot to mention that I'm looking for a solution that doesn't require saving an image file.

I've written the following example that almost achieves this, except it doesn't preserve image resolution. Any hints as to what I may be doing wrong would be greatly appreciated:


import matplotlib.pylab as plt
import numpy as np

def main():
  """Test for extracting pixel data from an Axes.

  This creates an image I, imshow()'s it to one Axes, then copies the pixel
  data out of that Axes to a numpy array I_copy, and imshow()'s the I_copy to
  another Axes.

  Problem: The two Axes *look* identical, but I does not equal I_copy.
  """

  fig, axes_pair = plt.subplots(1, 2)

  reds, greens = np.meshgrid(np.arange(0, 255), np.arange(0, 122))
  blues = np.zeros_like(reds)
  image = np.concatenate([x[..., np.newaxis] for x in (reds, greens, blues)],
                         axis=2)
  image = np.asarray(image, dtype=np.uint8)

  axes_pair[0].imshow(image)
  fig.canvas.draw()

  trans = axes_pair[0].figure.dpi_scale_trans.inverted()
  bbox = axes_pair[0].bbox.transformed(trans)
  bbox_contents = fig.canvas.copy_from_bbox(axes_pair[0].bbox)
  left, bottom, right, top = bbox_contents.get_extents()

  image_copy = np.fromstring(bbox_contents.to_string(),
                             dtype=np.uint8,
                             sep="")

  image_copy = image_copy.reshape([top - bottom, right - left, 4])
  image_copy = image_copy[..., :3]  # chop off alpha channel

  axes_pair[1].imshow(image_copy)

  print("Are the images perfectly equal? {}".format(np.all(image == image_copy)))

  plt.show()


if __name__ == '__main__':
  main()

Do you want the resulting image to be as large (in pixels) as the original one, or do you not care about the resolution? — ImportanceOfBeingErnest
– ImportanceOfBeingErnest, Commented Mar 29, 2017 at 15:00
Yes, the rendered image should be the same resolution as what I see on the screen when plotting. — SuperElectric
– SuperElectric, Commented Mar 29, 2017 at 15:25

ImportanceOfBeingErnest · Accepted Answer · 2017-03-30 06:49:08Z

4

One idea can be to intermediately turn the axes off, find out the bounding box of the axes in inches and then save the figure using the bbox_inches argument to plt.savefig().

If a numpy array is wanted, one can then read in the saved image again using plt.imread.

In this solution the returned array has dimensions exactly as the axes has pixels when plotted on the screen.

import matplotlib.pyplot as plt
import numpy as np; np.random.seed(0)

im = np.random.rand(16,16)

x = np.arange(9)
y = np.random.randint(1,14, size=(9,))
y2 = np.random.randint(1,7, size=(9,))

fig, (ax1, ax2) = plt.subplots(1,2)

ax1.imshow(im[:16,:9], cmap="viridis")
ax2.imshow(im[7:16,7:], cmap="YlGnBu")
ax1.plot(x, y, color="C3")
ax1.scatter(x, y, color="w")
ax2.plot(x, y2, color="C1")
ax2.scatter(x, y2, color="k")
ax1.set_xlabel("YlFasOcto")

def save_ax(ax, filename, **kwargs):
    ax.axis("off")
    ax.figure.canvas.draw()
    trans = ax.figure.dpi_scale_trans.inverted() 
    bbox = ax.bbox.transformed(trans)
    plt.savefig(filename, dpi="figure", bbox_inches=bbox,  **kwargs)
    ax.axis("on")
    im = plt.imread(filename)
    return im

arr = save_ax(ax1, __file__+".png")
print(arr.shape)
plt.show()

In order to prevent saving a file to disk, one could use a Stream to save the data.

import io

def save_ax_nosave(ax, **kwargs):
    ax.axis("off")
    ax.figure.canvas.draw()
    trans = ax.figure.dpi_scale_trans.inverted() 
    bbox = ax.bbox.transformed(trans)
    buff = io.BytesIO()
    plt.savefig(buff, format="png", dpi=ax.figure.dpi, bbox_inches=bbox,  **kwargs)
    ax.axis("on")
    buff.seek(0)
    im = plt.imread(buff )
    return im

edited Mar 30, 2017 at 6:49

answered Mar 29, 2017 at 16:27

ImportanceOfBeingErnest

342k61 gold badges737 silver badges771 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

SuperElectric Over a year ago

Unfortunately, I haven't been able to run your code (our Matplotlib versions may differ). I've updated my question with some example code of my own, which borrows your idea of using dpi_scale_trans.inverted(), but doesn't need to save out to a file (a necessary property I forgot to mention). It mostly does what I want, except for preserving the resolution of the originally imshow'ed image. Which is a problem.

ImportanceOfBeingErnest Over a year ago

"I haven't been able to run your code" is not a sufficient problem description. Initially I did try a similar solution as the one from the question, using to_string() method, but the image was shifted compared to the original. I therefore came up with the solution in this answer. I updated the answer to include the case where no output file is desired.

Collectives™ on Stack Overflow

How can I render a Matplotlib Axes object to an image (as a Numpy array)?

Update:

1 Answer 1

2 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

Update:

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related