TIFF UnidentifiedImageError

[welrbraga]

Hi team,

I have some images in tiff format that came from many different people and that I need to convert do JPG.

Both images are ok and visible without problems on all image viewer I could check, and using Image Magick I can sucessfully convert them, but, when I try to use PIL to do the same result in Python, some images are not converted, raising the "UnidentifiedImageError: cannot identify image file".

Using convert

#IMAGE 1 TEST OK
convert -verbose FLOR0017314_02.tif FLOR0017314_02.tif.jpg
FLOR0017314_02.tif TIFF 3664x2748 3664x2748+0+0 8-bit TrueColor sRGB 28.8278MiB 0.050u 0:00.039
FLOR0017314_02.tif=>FLOR0017314_02.tif.jpg TIFF 3664x2748 3664x2748+0+0 8-bit TrueColor sRGB 1.72944MiB 0.140u 0:00.150

#IMAGE 2 TEST OK
convert -verbose ../rb-back/so_tif/00082851.tif 00082851.tif.jpg
../rb-back/so_tif/00082851.tif TIFF 6412x10261 6412x10261+0+0 8-bit TrueColor sRGB 137.123MiB 1.510u 0:01.350
../rb-back/so_tif/00082851.tif=>00082851.tif.jpg TIFF 6412x10261 6412x10261+0+0 8-bit TrueColor sRGB 21.7932MiB 1.280u 0:01.280

Using Python + Pillow

My env test is:

• Ubuntu 20.04
• Python 3.8.5
• Pillow-8.0.1

To open my images I tried this minimal piece of code:

#IMAGE 1 TEST OK
In [1]: from PIL import Image
In [2]: filename = '../rb-back/so_tif/00082851.tif'                                                                                                                               
In [3]: image_tif = Image.open(filename)                                                                                                                                          
In [4]: image_tif                                                                                                                                                                 
Out[4]: <PIL.TiffImagePlugin.TiffImageFile image mode=RGB size=6412x10261 at 0x7FAAED2CE310>

#IMAGE 2 BUGGED TEST 
In [1]: from PIL import Image
In [2]: filename = 'FLOR0017314_02.tif'
In [3]: image_tif = Image.open(filename)
---------------------------------------------------------------------------
UnidentifiedImageError                    Traceback (most recent call last)
<ipython-input-3-49b4a2ebb92e> in <module>
----> 1 image_tif = Image.open(filename)

/usr/lib/python3/dist-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 'FLOR0017314_02.tif'

In [4]: image_tif                                                                                                                                                                 
Out[4]: <PIL.TiffImagePlugin.TiffImageFile image mode=RGB size=6412x10261 at 0x7FAAED7ADA30>

I've tried to compare the difference between them but I couldn't note anything that is causing this. Maybe it is a bug in Pillow or my brain, I sincerely don't known. Can you help me, please?

I'm attached the returned exit of "identify -verbose" for each image file: identify-verbose-bugged-image.txt identify-verbose-ok-image.txt

[radarhere]

You're unable to attach the images themselves?

[welrbraga]

One image is approximately 30MB (the problematic) and another is more than 140MB but Github limit the size of attachment to 10MB, only.

Is there another way to share them?

[radarhere]

I wouldn't necessarily worry about the ok image, just the broken one. You could create a GitHub repo and use the command line to commit and push the 30mb image. If you use Google Drive or Dropbox you could upload them and then create a shareable link. There are likely other ways as well, but those are my thoughts.

[welrbraga]

Thanks your attention. I've put three of them in a Google Drive shared folder.

I hope it can help you to help me.

[radarhere]

Debugging further, I get

Traceback (most recent call last):
  File "PIL/TiffImagePlugin.py", line 1263, in _setup
KeyError: (b'II', 2, (2, 2, 2), 1, (8, 8, 8), ())

All three images have the same information. So it is an image mode (RGB;S ?) that we don't support yet.

For posterity, here is a zip of one of the images - FLOR0017314_02.tif.zip

[welrbraga]

Thanks @radarhere , but I didn't understand how they could have the same informations. If you open in a image viewer you see they are different pics. What happen to cause this, do you know? This information could help me to minimize this trouble in future.

[radarhere]

Oh, I didn't mean the images were exactly the same. I meant they all have (b'II', 2, (2, 2, 2), 1, (8, 8, 8), ()). And I don't want you to think the images are broken. There are just many ways to save images, and these have been saved in a way that Pillow hasn't added support for - they have 3 samples per pixel of two's complement signed integer data, and Pillow doesn't support that yet.

[welrbraga]

Hi @radarhere , Thanks your answer. This is one of my thousands curious cases in a set with with 5.5 milions images. I'd like to use just one lib to process it, but I couldn't and adjust my tools to use "Wand" (https://pypi.org/project/Wand/ - a ImageMagick binding for Python) as a fallback when Pillow fail and it solved for me.

[cipri-tom]

I also encounter this with some satellite images that I want to process. These have (16 16 16) bits per sample, according to exiftool.

As a workaround, I have to resort to using tifffile.imread(), but I would also like to use Pillow for all my image operations.

Here's the output of exiftool

ExifTool Version Number         : 10.80
File Name                       : mexico-earthquake_00000000_post_disaster.tif
Directory                       : /home/data/xbd/geotiffs/tier1/images
File Size                       : 6.0 MB
File Modification Date/Time     : 2020:08:03 22:15:48+00:00
File Access Date/Time           : 2021:06:08 15:31:03+00:00
File Inode Change Date/Time     : 2021:06:08 15:30:57+00:00
File Permissions                : rw-r--r--
File Type                       : TIFF
File Type Extension             : tif
MIME Type                       : image/tiff
Exif Byte Order                 : Little-endian (Intel, II)
Image Width                     : 1024
Image Height                    : 1024
Bits Per Sample                 : 16 16 16
Compression                     : Uncompressed
Photometric Interpretation      : BlackIsZero
Strip Offsets                   : (Binary data 8012 bytes, use -b option to extract)
Samples Per Pixel               : 3
Rows Per Strip                  : 1
Strip Byte Counts               : (Binary data 5119 bytes, use -b option to extract)
Planar Configuration            : Chunky
Extra Samples                   : Unknown (0 0)
Sample Format                   : Signed; Signed; Signed
Pixel Scale                     : 4.49466028913751e-06 4.49466028913751e-06 0
Model Tie Point                 : 0 0 0 -99.228625156624 19.326981105618 0
GDAL No Data                    : -99
Geo Tiff Version                : 1.1.0
GT Model Type                   : Geographic
GT Raster Type                  : Pixel Is Area
Geographic Type                 : WGS 84
Geog Citation                   : WGS 84
Geog Angular Units              : Angular Degree
Geog Semi Major Axis            : 6378137
Geog Inv Flattening             : 298.257223563
Image Size                      : 1024x1024
Megapixels                      : 1.0

And the image (zipped):

mexico-earthquake_00000000_post_disaster.tif.zip

[radarhere]

@cipri-tom if I deduct the 2 extra samples from the 3 samples per pixel, then I get one channel left. Is the image grayscale? I tried loading it as I;16S, but it was still black. Could you attach a copy of what the image should look like?

[cipri-tom]

@radarhere No, the image is colour, RGB, 1024 x 1024 x 3.

Here's what is looks like

Obtained by simply resaving it as png (all values were in 0-255, although original had 16 bits per sample).

[radarhere]

I've come to the conclusion that these images fall into the category of #1888.