I have an array with 600×600×40 dimension that each band(from 40 band) represent a 600×600 image I want to save it to a multiple band .tif image. I have tried this functions from scikit-image and openCV but they can not save more than 3 band(as RGB).
import cv2
cv2.imwrite('image.tif',600by600_just3band_array)
tifffile (https://pypi.org/project/tifffile/) supports multi-channel .tiff's and has an API similar to the one of scikit-image or OpenCV:
In [1]: import numpy as np
In [2]: import tifffile
In [3]: # Channel dimension should come first
In [4]: x = np.random.randint(0, 255, 4*100*100).reshape((4, 100, 100))
In [5]: tifffile.imsave('test.tiff', x)
In [6]: y = tifffile.imread('test.tiff')
In [7]: np.all(np.equal(x, y))
Out[7]: True
You could save multiple images, each representing a single band (greyscale), or even multiple bands (colour) in a single TIFF file with PIL/Pillow like this:
from PIL import Image
# Synthesize 8 dummy images, all greyscale, all same size but with varying brightness
size=(480,640)
b1 = Image.new('L', size, color=10)
b2 = Image.new('L', size, color=20)
b3 = Image.new('L', size, color=30)
b4 = Image.new('L', size, color=40)
b5 = Image.new('L', size, color=50)
b6 = Image.new('L', size, color=60)
b7 = Image.new('L', size, color=70)
b8 = Image.new('L', size, color=80)
# Save all 8 to single TIFF file
b1.save('multi.tif', save_all=True, append_images=[b2,b3,b4,b5,b6,b7,b8])
If you now examine that file with ImageMagick at the command line, you can see all 8 bands are present:
magick identify multi.tif
multi.tif[0] TIFF 480x640 480x640+0+0 8-bit Grayscale Gray 2.34473MiB 0.000u 0:00.000
multi.tif[1] TIFF 480x640 480x640+0+0 8-bit Grayscale Gray 2.34473MiB 0.000u 0:00.000
multi.tif[2] TIFF 480x640 480x640+0+0 8-bit Grayscale Gray 2.34473MiB 0.000u 0:00.000
multi.tif[3] TIFF 480x640 480x640+0+0 8-bit Grayscale Gray 2.34473MiB 0.000u 0:00.000
multi.tif[4] TIFF 480x640 480x640+0+0 8-bit Grayscale Gray 2.34473MiB 0.000u 0:00.000
multi.tif[5] TIFF 480x640 480x640+0+0 8-bit Grayscale Gray 2.34473MiB 0.000u 0:00.000
multi.tif[6] TIFF 480x640 480x640+0+0 8-bit Grayscale Gray 2.34473MiB 0.000u 0:00.000
multi.tif[7] TIFF 480x640 480x640+0+0 8-bit Grayscale Gray 2.34473MiB 0.000u 0:00.000
In case you are using OpenCV or Numpy arrays for your processing, you can make an OpenCV or Numpy array into a PIL/Pillow image with:
PILimage = Image.fromarray(numpyImage)
and, going the other way, from a PIL/Pillow image to Numpy array:
NumpyImage = np.array(PILimage)
If you then want to read them back, you can do this:
# Open the multi image
im = Image.open('multi.tif')
# Iterate through frames
for frame in ImageSequence.Iterator(im):
frame.show()
If you want to move to a specific band, you can seek like this:
im = Image.open('multi.tif')
im.seek(3)
im.show()
You can also extract band3 from the TIF and save as a PNG with ImageMagick at the command line with:
magick multi.tif[3] band3.png
Or make a band 1, 2, 7 RGB composite with:
magick multi.tif[1] multi.tif[2] multi.tif[7] -colorspace RGB -combine 127rgb.png
which will look dark blue because the red and the green channels are very low and only the blue channel has a large-ish value.
I am not the world's best on Python, so am uncertain of any implications/errors, but I think if you have a 600x600x40 numpy array of images, you can do what I am suggesting like this:
# Synthesize dummy array of 40 images, each 600x600
nparr = np.random.randint(0,256,(600,600,40), dtype=np.uint8)
# Make PIL/Pillow image of first
a = Image.fromarray(nparr[:,:,0])
# Save whole lot in one TIF
a.save('multi.tif', save_all=True, append_images=[Image.fromarray(nparr[:,:,x]) for x in range(1,40)])
Keywords: Multi-band, multi band, multi-spectral, multi spectral, satellite image, imagery, image processing, Python, Numpy, PIL, Pillow, TIFF, TIF, NDVI
Mark's clever answer is making a multi-page TIFF. Unfortunately, imagemagick and PIL are really MONO / RGB / RGBA / CMYK libraries and they don't have direct support for multiband images.
pyvips has true multiband support. For example:
import sys
import pyvips
import numpy as np
# make a (100, 100, 40) numpy image
array = np.zeros((100, 100, 40), dtype=sys.argv[2])
# convert to vips and save
image = numpy2vips(array)
image.write_to_file(sys.argv[1])
# read it back, convert to numpy, and show info
image2 = pyvips.Image.new_from_file(sys.argv[1])
array = vips2numpy(image2)
print("shape =", array.shape)
print("format =", array.dtype)
I can run it like this:
$ ./try284.py x.tif uint8
shape = (100, 100, 40)
format = uint8
$ vipsheader x.tif
x.tif: 100x100 uchar, 40 bands, srgb, tiffload
$ identify x.tif
x.tif TIFF 100x100 100x100+0+0 8-bit sRGB 400KB 0.000u 0:00.000
It supports other dtypes as well:
$ ./try284.py x.tif uint32
shape = (100, 100, 40)
format = uint32
$ ./try284.py x.tif float32
shape = (100, 100, 40)
format = float32
etc. etc.
You can load these TIFFs in gdal. I guess gdal can be used to write them as well, though I've not tried. Annoyingly, it moves the 40 to the outermost dimension.
$ python3
Python 3.6.7 (default, Oct 22 2018, 11:32:17)
[GCC 8.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from osgeo import gdal
>>> x = gdal.Open("x.tif")
>>> a = x.ReadAsArray()
>>> a.shape
(40, 100, 100)
vips2numpy() and numpy2vips() are defined here:
https://github.com/libvips/pyvips/blob/master/examples/pil-numpy-pyvips.py
Copy-pasted for reference:
# map vips formats to np dtypes
format_to_dtype = {
'uchar': np.uint8,
'char': np.int8,
'ushort': np.uint16,
'short': np.int16,
'uint': np.uint32,
'int': np.int32,
'float': np.float32,
'double': np.float64,
'complex': np.complex64,
'dpcomplex': np.complex128,
}
# map np dtypes to vips
dtype_to_format = {
'uint8': 'uchar',
'int8': 'char',
'uint16': 'ushort',
'int16': 'short',
'uint32': 'uint',
'int32': 'int',
'float32': 'float',
'float64': 'double',
'complex64': 'complex',
'complex128': 'dpcomplex',
}
# numpy array to vips image
def numpy2vips(a):
height, width, bands = a.shape
linear = a.reshape(width * height * bands)
vi = pyvips.Image.new_from_memory(linear.data, width, height, bands,
dtype_to_format[str(a.dtype)])
return vi
# vips image to numpy array
def vips2numpy(vi):
return np.ndarray(buffer=vi.write_to_memory(),
dtype=format_to_dtype[vi.format],
shape=[vi.height, vi.width, vi.bands])
