X and Y scale factor is being ignored during image extraction

[hdnh2006]

There are errors while reading images in pdf. As you can see in the following image. It looks like the image appears to have some distortion or stretching:

But as you can see in the pdf the original image doesn't look like that.

Environment

Which environment were you using when you encountered the problem?

$ python -m platform
Linux-6.11.0-21-generic-x86_64-with-glibc2.39

$ python -c "import pypdf;print(pypdf._debug_versions)"
pypdf==5.4.0, crypt_provider=('local_crypt_fallback', '0.0.0'), PIL=11.2.1

Code + PDF

PDF:
This link

This is a minimal, complete example that shows the issue:

from pypdf import PdfReader
import io
from PIL import Image

pdf_path = "...mypdf.pdf"

reader = PdfReader(pdf_path)
for page in reader.pages:
    for image_file_object in page.images:

        image = Image.open(io.BytesIO(image_file_object.data), formats=['JPEG', 'PNG'])
        image.save('image.png')

[pubpub-zz]

The image extraction : in the content of the page a different scale factor is applied on X and Y :

q
  1190.52 0 0 841.8 0 0 cm
  /Im0 Do
Q

this explains why the image is stretched when extracted but displayed differently in the pdf.

[hdnh2006]

Ok @pubpub-zz how do I avoid this? I want to get the image with the correct width and height

[stefan6419846]

There probably is no easy approach to avoid this for now, as the image and its extraction to not know the current transformation. This requires further processing of all transformations of the page until the image is inserted.

[pubpub-zz]

You may use the visitor function working to identify the cm matrix when the Do operator is detected
see https://pypdf.readthedocs.io/en/stable/user/extract-text.html#example-2-extract-rectangles-and-texts-into-a-svg-file
With the cm matrix you should be able the correction ratio to apply

[hdnh2006]

You may use the visitor function working to identify the cm matrix when the Do operator is detected see https://pypdf.readthedocs.io/en/stable/user/extract-text.html#example-2-extract-rectangles-and-texts-into-a-svg-file With the cm matrix you should be able the correction ratio to apply

Could you please explain a bit deeper how to use it?

I don't understand exactly why this is happening, the images should be loaded exactly as they are, am I righ?

[pubpub-zz]

Could you please explain a bit deeper how to use it?

(with a little of simplification)
PDF files can contain images. Theses are stored in /XObject files in /Resources section of a /Page object. when an image has to be displayed in the /Content you will find a Do XObjectName in the /Content section. When doing this call the current scale/rotation/translation are used : this could make the image different in size/ratio/angle different from the original image

I don't understand exactly why this is happening, the images should be loaded exactly as they are, am I righ?

No : the stored image can be different from the display.
If you use some tools such as pdfdebug you will see that the image is properly extracted

[shartzog]

Note, for the specific Image/PDF identified in the OP, the operation is pretty straightforward: the image is simply rotated 270 degrees and rescaled to the full page dimensions at display time. Wouldn't you just have to

extracted_image.rotate(270)
extracted_image.resize((3508, 4961))
extracted_image.save("/output/path/extracted_image.jpg")

where (3508, 4961) can be replaced with any tuple of integers that maintain the 'A' paper aspect ratio?

Note: (3508, 4961) corresponds to 300 dpi for an A3 portrait page. (842, 1191) would correspond to 72 dpi for an A3 portrait page and appears to be more or less exactly what the cm operands 1190.52 and 841.8 are doing in the content stream snippet shared by @pubpub-zz above.

[pubpub-zz]

This is definitely not an issue. I convert this into a disccusion