PIL cannot read BigTIFF

[madelinehayes]

What did you do?

I'm trying to open an orthomosaic geotiff into Python to crop into 1000x1000 tiles. I am able to open some tif files in my notebook, but other files return an UnidentifiedImageError. I have set up my notebook using a Dockerfile that creates a conda environment.

What did you expect to happen?

I expected that my code would work the same on all of my tif files - they are all 4-band 8-bit. This code works with a tif file that is 992 MB, but it doesn't work with a file that's 691 MB.

What actually happened?

---------------------------------------------------------------------------
UnidentifiedImageError                    Traceback (most recent call last)
<ipython-input-85-c7f17ab4563e> in <module>
      6 
      7 # break it up into crops
----> 8 for k, piece in enumerate(crop(infile, tile_height, tile_width, stride, img_dict, prj_name), start_num):
      9     img=Image.new('RGB', (tile_height, tile_width), (255, 255, 255))
     10     print(img.size)

<ipython-input-83-c9b468169840> in crop(infile, tile_height, tile_width, stride, img_dict, prj_name)
      5 
      6 def crop(infile, tile_height, tile_width, stride, img_dict, prj_name):
----> 7     im = Image.open(infile)
      8     img_width, img_height = im.size
      9     print(im.size)

/opt/conda/envs/geo_env/lib/python3.8/site-packages/PIL/Image.py in open(fp, mode)
   2859     for message in accept_warnings:
   2860         warnings.warn(message)
-> 2861     raise UnidentifiedImageError(
   2862         "cannot identify image file %r" % (filename if filename else fp)
   2863     )

UnidentifiedImageError: cannot identify image file '../data/mosaics/GrandJason_SWRightThird_Nov2019_transparent_mosaic_group1.tif'

What are your OS, Python and Pillow versions?

• OS: Ubuntu 18.04
• Python: 3.8.2
• Pillow: 7.0.0

from PIL import Image
import os
import argparse
import numpy as np
import json
import csv
import rasterio
import matplotlib
import folium
from pyproj import Proj, transform


%matplotlib inline


Image.MAX_IMAGE_PIXELS = 100000000000
# ingest the image
infile = "../data/mosaics/GrandJason_SWRightThird_Nov2019_transparent_mosaic_group1.tif"

img_dir = '..' + infile.split(".")[2]
prj_name = img_dir.split("/")[-1]
dataset = rasterio.open(infile)

# what is the name of this image
img_name = dataset.name
print('Image filename: {n}\n'.format(n=img_name))

# How many bands does this image have?
num_bands = dataset.count
print('Number of bands in image: {n}\n'.format(n=num_bands))

# How many rows and columns?
rows, cols = dataset.shape
print('Image size is: {r} rows x {c} columns\n'.format(r=rows, c=cols))

# Does the raster have a description or metadata?
desc = dataset.descriptions
metadata = dataset.meta

print('Raster description: {desc}\n'.format(desc=desc))

# What driver was used to open the raster?
driver = dataset.driver
print('Raster driver: {d}\n'.format(d=driver))

# What is the raster's projection?
proj = dataset.crs
print('Image projection:')
print(proj, '\n')

# What is the raster's "geo-transform"
gt = dataset.transform

print('Image geo-transform:\n{gt}\n'.format(gt=gt))

print('All raster metadata:')
print(metadata)
print('\n')

tile_height = tile_width = 1000
overlap = 80
stride = tile_height - overlap
start_num=0

#crop image into tiles
def crop(infile, tile_height, tile_width, stride, img_dict, prj_name):
    im = Image.open(infile) 
    img_width, img_height = im.size
    print(im.size)
    print(img_width * img_height / (tile_height - stride) / (tile_width - stride))
    count = 0
    for r in range(0, img_height-tile_height+1, stride):
        for c in range(0, img_width-tile_width+1, stride):
            #tile = im[r:r+100, c:c+100]
            box = (c, r, c+tile_width, r+tile_height)
            top_pixel = [c,r]
            img_dict[prj_name + "---" + str(count) + ".png"] = top_pixel
            count += 1
            yield im.crop(box)

#split image into heightxwidth patches
img = Image

img_dict = {}

# create the dir if it doesn't already exist
if not os.path.exists(img_dir):
    os.makedirs(img_dir)

# break it up into crops
for k, piece in enumerate(crop(infile, tile_height, tile_width, stride, img_dict, prj_name), start_num):
    img=Image.new('RGB', (tile_height, tile_width), (255, 255, 255))
    print(img.size)
    print(piece.size)
    img.paste(piece)
    image_name = prj_name + "---%s.png" % k
    path=os.path.join(img_dir, image_name)
    img.save(path)

[cgohlke]

Could be a BigTIFF file, which is not supported by Pillow. The first 4 bytes in BigTIFF files are b'II\x2B\x00' or b'MM\x00\x2B'

[madelinehayes]

Could be a BigTIFF file, which is not supported by Pillow. The first 4 bytes in BigTIFF files are b'II\x2B\x00' or b'MM\x00\x2B'

None of my tif files are BigTIFF, and they are all way under 4GB size

[madelinehayes]

Here is the information of my raster:

Image size is: 18679 rows x 31880 columns

Raster description: (None, None, None, None)

Raster driver: GTiff

Image projection:
EPSG:32720 

Image geo-transform:
| 0.02, 0.00, 632879.80|
| 0.00,-0.02, 4341041.62|
| 0.00, 0.00, 1.00|

All raster metadata:
{'driver': 'GTiff', 'dtype': 'uint8', 'nodata': None, 'width': 31880, 'height': 18679, 'count': 4, 'crs': CRS.from_epsg(32720), 'transform': Affine(0.01804, 0.0, 632879.79994,
       0.0, -0.01804, 4341041.621540001)}

bytearray(b'II+\x00\x08')

[cgohlke]

bytearray(b'II+\x00\x08')

That's BigTIFF.

[hroskes]

Are there any plans to support bigtiff? I'd like to add my name to the list of people who would find it very useful.

[juliangilbey]

I've just stumbled upon this issue too. At a first glance, it seems to me that it should be pretty straightforward to modify src/libImaging/TiffDecode.c to handle the BigTIFF format, which is meant to be very closely compatible with the TIFF format.

[juliangilbey]

Ah, after working on this for about a day, and getting most of the way to a working solution, I realised that it is probably futile, as BigTIFF files are to a great extent used for medical images, and they will include multiple images within the file (for example, thumbnails). But PIL can only handle one image in a file, so this won't work very well.

A better solution is probably to identify BigTIFF files (from the b"MM\x00\x2B" or b"II\x2B\x00" header bytes) and then recommend using something like tifffile to handle the image.

If it's of interest, I could upload what I've done so far to github. (It's primarily work on TiffImagePlugin.py and TiffTags.py.)

Best wishes!

[radarhere]

But PIL can only handle one image in a file, so this won't work very well.

Pillow can load multiple images. It opens the file at one image, and then uses seek() to navigate to a different image.

If you're willing to share what you've done, and an example image, that would be interesting.

[juliangilbey]

OK, shall do when I've had a chance to get at least some of it working!

[radarhere]

I've created PR #6097 to resolve this.