Matplotlib: force aspect ratio in series of plots

Question

I have a series of images that represent physical regions of space, ie they are images of objects. They do not have a uniform amount of pixels, but in EVERY case, the resolution along the y-axis is the same for every image, and the length along the x-axis is the same (specifically: pixels are always 150 um across in the y dimension and images are always 11mm across in the x dimension).

I would like to generate a series of plots whose sizes accurately reflect the spatial sizes of the objects in relation to one another, so images that have a greater number of rows correspond to being larger in the y-dimension spatially, but every image is the same size in the x-dimension regardless of how many pixels there are in that dimension (since resolution in x is not constant).

To do this, my script calculates the x and y dimensions in spatial units (for example, 11mm x 4.2 mm or 11mm x 4.5 mm) and then forces the aspect ratio in plt.imshow using the aspect setting to the ratio of the y axis size to the x axis size.

This does not work for some reason. Also, setting the figure size doesn't work. Here's my code and images:

for f in mats: # Separate filename from extension name, ext = os.path.splitext(f)

# Get the age of the sample lens
age = int(name.split('_', 1)[0])

bshiftm = io.loadmat(f)['bshiftm']  # Un-interpolated image
bshiftm2 = io.loadmat(f)['bshiftm2']  # Interpolated image

# Load the data about the physical dimensions of images
xscan = int(io.loadmat(f)['xscan'])  # in mm
zscan = int(io.loadmat(f)['zscan'])  # in mm
zframes = int(io.loadmat(f)['zframes'])

# Determine resolution data so we can scale plot axes
xres = float(xscan)/bshiftm.shape[1]  # x dim. is the number of columns
zres = float(zscan)/zframes  # In mm/pixel

# All images have same x dimensions, but z dimensions are all different,
# meaning we have to figure them out based on the resolution of the image
# in the z direction and the number of rows in the raw image
xsize = xscan  # In mm
zsize = zres*bshiftm.shape[0]  # In mm

# Using the physical dimensions of the image, find the resolutions of the
# interpolated images so we can scale the plot axes
xresi = float(xsize)/bshiftm2.shape[1]
zresi = float(zsize)/bshiftm2.shape[0]

# Aspect ratio of image
ar = float(zsize)/xsize

# Start plotting
fig = plt.figure()

# Locations and labels for x tics
xlocs = np.arange(0, bshiftm.shape[1], bshiftm.shape[1]/5)
xlabs = map(str, np.round(xlocs*xres, 1))
plt.xticks(xlocs, xlabs)
plt.xlabel('x (mm)').set_weight('bold')

# Locations and labels for z tics
zlocs = np.arange(0, bshiftm.shape[0], bshiftm.shape[0]/2)
zlabs = map(str, np.round(zlocs*zres, 1))
plt.yticks(zlocs, zlabs)
plt.ylabel('z (mm)').set_weight('bold')

# Set the title
#plt.title(str(age) + ' year old lens, XZ').set_weight('bold')

# Plot the uninterpolated data
plt.imshow(bshiftm,
           cmap='spectral',
           aspect=ar,
           vmin=8,
           vmax=9.6,
           origin='upper',
           interpolation='none')

# Set up colorbar
cbar = plt.colorbar()
#cbar.set_label('Brillouin Shift (GHz)', rotation=270)

plt.savefig(os.path.join(raw, name))
plt.close()

It seems to me that this should work, and I have verified that this script correctly calculates the spatial dimensions of every image, along with the aspect ratio that each should have. I should point out that the third image is 110 pixels across in the x direction, while the other two are only 55.

Also, if anybody has advice on how to place ticks more robustly, that would be wonderful to hear.

Answer 1

At least one rather robust way is to do the image scaling by using the extent keyword.

ax.imshow(image, extent=[0, 10.4, 0, 4.2], aspect=1)

This should handle everything you need. (Of course, change 10.4 and 4.2 to whatever the physical dimensions are). It should be noted that in this case the aspect=1 does not necessarily make square pixels, instead it forces the units on both axis to be equal in scale - which seems to be what you want.

Just as a small example:

import matplotlib.pyplot as plt
import numpy as np

fig = plt.figure()
ax = fig.add_subplot(111)
ax.imshow(np.random.random((20, 20)), extent=(0, 10.5, 2, 4.7), aspect=1, interpolation='nearest')

This gives: