Pymupdf: Question / Comment: Error when get bbox with page.getImageBbox()

[dothinking]

Hi JorjMcKie,

I got problems when extracting image bbox with page.getImageBbox() from this pdf.

File "D:\89_Program_Files\Python368\lib\site-packages\fitz\fitz.py", line 4791, in get ImageBbox
    raise ValueError("unsupported image item")
ValueError: unsupported image item

What I did:

import fitz

doc = fitz.open("test.pdf")
page = doc[0]

# I can get the image from page
items = page.getImageList(full=True)
print(items)
# [(8, 0, 1600, 939, 8, 'DeviceRGB', '', 'Im0', 'FlateDecode', 7)]

# but failed to get the bbox
bbox = page.getImageBbox(items[0])
print(bbox)
# ValueError: unsupported image item

Check fitz.py line 4791:

if item[-1] != 0:
    raise ValueError("unsupported image item")

In this case, Seems the last xref item[-1]==7 != 0 caused the error.

So, how to get bbox of image with item[-1] != 0? Thanks in advance.

[dothinking]

I tried to cross check the bbox with page.getText(), but unfortunately I can't get any image blocks, seems it's due to this image is partly outside the page.

print(page.getText('rawdict'))
# {'width': 612.0, 'height': 792.0, 'blocks': []}

[JorjMcKie]

As your analysis correctly showed:
Image bboxes can only be determined if the page directly displays the image. In your case, a so-called "Form XObject" is invoked (xref 7), which in turn displays an image.
You can only find out the bbox occupied by the XObject doc.getPageXObjectList(page.number), which contains a tuple for the bbox.
The reason for this restriction mainly is that the resp. MuPDF algorithm is not bug-free. It is indeed a highly complex topic, because XObjects can be nested inside each other at arbitrary levels, each with its own transformation matrix, bbox and what not.

[dothinking]

Thanks for the explanation.

You can only find out the bbox occupied by the XObject doc.getPageXObjectList(page.number)

I still can't get the right bbox with getPageXObjectList, considered also page.transformationMatrix, as documented in the Docs. Indeed, "It is indeed a highly complex topic".

Then I try to bypass item[-1] != 0, and see what happen from MuPDF. Luckily I get the result. Maybe "MuPDF algorithm is not bug-free", but at least works for this case. So a workaround for my case, bypass item[-1] != 0 to give MuPDF a chance. Of course, wrap in try/except clause for safe.

import fitz

doc = fitz.open("test.pdf")
page = doc[0]
items = page.getImageList(full=True)

# bypass `item[-1] != 0`
item = list(items[0])
item[-1]=0

bbox = page.getImageBbox(item)
print(bbox)
# Rect(57.900001525878906, 129.5078125, 688.8927612304688, 582.1100463867188)

[JorjMcKie]

Maybe I should follow your implicit recommendation, and no longer refuse to do it like that.
The only thing I do not want to see happening is having to deal with issues, when it didn' twork correctly ... 😎.
BTW, the risk to see exception is minor - which is unfortunate in this case: you will get an incorrect bbox without necessarily noticing it.
Maybe I let it go through and warn if I detect that the image is embedded in an XObject.