I need to read a single-channel 32-bit integer TIFF image from disk using Python 2 to perform some image analysis. I tried image.imread from matplotlib but I was unable to get the code to work because data were read as 4-channel 8-bit integer image:
>>> import numpy as np
>>> import matplotlib.image as mpimg
>>> img = mpimg.imread('my_image.tif')
>>> img.shape
(52, 80, 4)
>>> img[0:2, 0:2]
array([[[255, 255, 255, 255],
[255, 255, 255, 255]],
[[255, 255, 255, 255],
[255, 255, 255, 255]]], dtype=uint8)
QUESTION: Is it possible to read a single-channel 32-bit integer image using matplotlib?
I'm aware there are alternative ways to read such an image in Python, for example using Image.open from PIL:
>>> from PIL import Image
>>> img = np.asarray(Image.open('my_image.tif'))
>>> img.dtype
dtype('int32')
>>> img.shape
(52, 80)
>>> img[0:2, 0:2]
array([[8745, 8917],
[8918, 9479]])
Another possibility would be using io.imread from scikit-learn:
>>> from skimage import io
>>> img = io.imread('my_image.tif')
Yet another approach is to utilize the imread function from OpenCV. But in this case data have to be casted to 32-bit integer:
>>> import cv2
>>> img = cv2.imread('my_image.tif', -1)
>>> img[0:2, 0:2]
array([[ 1.22543551e-41, 1.24953784e-41],
[ 1.24967797e-41, 1.32829081e-41]], dtype=float32)
>>> img.dtype = np.int32
>>> img[0:2, 0:2]
array([[8745, 8917],
[8918, 9479]])
According to this tutorial, there's no way to read a 32-bit integer image using matplotlib:
Matplotlib plotting can handle float32 and uint8, but image reading/writing for any format other than PNG is limited to uint8 data.
For reference, I found out one more workaround based on ndimage.imread from SciPy:
from scipy import ndimage
img = ndimage.imread('my_image.tif', mode='I')
The approach based on tifffile (suggested by @Warren Weckesser) also works fine:
from tifffile import TiffFile
with TiffFile('my_image.tif') as tif:
img = tif.asarray()