Struggling with fitz.Pixmap (RuntimeError: unknown image file format)

[axlgit]

Hi,

I've got a PNG image described as numpy ndarray.
I think the ndarray is well shaped because when I passed it to Pytesseract it is accepted and Tesseract gets the text out of it.
Now I would like the same PNG image described as Numpy ndarray to be passed to fitz.Pixmap so that (I guess) it can be inserted into a PDF page by using to .InsertImage and the pixmap input.

To try to understand how to proceed I used this article and the PyMuPDF doc

https://code.activestate.com/recipes/580630-saving-a-numpy-array-to-png-image-using-fitz-mupdf/

https://documentation.help/PyMuPDF/page.html

but when I run this code I get the error about "image format not recognized".
(The module tools is just a module of mine with a set of procedure to select which images to load from a directory.)

>>> import numpy, fitz
>>> from tools import *
>>>
>>> image=select_images('TEST_page0001_trimmed2_rotated_whitened.png')
>>> image = load_images(image)
**>>> image_np_arr = np.ndarray(buffer=image[0].image.write_to_memory(),dtype=np.uint8,shape=[image[0].h,image[0].w,image[0].image.bands])**
>>> out_pdf = fitz.open()
>>> fmt = fitz.PaperRect("a4")
>>> out_page = out_pdf.newPage(width = fmt.width, height = fmt.height)
>>> pix = fitz.Pixmap((fitz.csRGB, image[0].w, image[0].h, image[0].image.bands, bytearray(image_np_arr.tobytes())))
mupdf: unknown image file format
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Users\axl\AppData\Local\Programs\Python\Python39\lib\site-packages\fitz\fitz.py", line 6389, in __init__
    _fitz.Pixmap_swiginit(self, _fitz.new_Pixmap(*args))
RuntimeError: unknown image file format

The assignement to the variable image_np_arr doesn't throw any error (I had just to switch to .tobytes() because the code told me .tostring() was deprecated)

As you can see the image is a PNG and there are 4 bands

>>> image[0].image.bands
4

How can I fix this? Or how to troubleshoot it more?

Thanks in advance,

Alex

[JorjMcKie]

You are using the wrong object constructor pattern when defining the pixmap: review the brackets!
You are passing one argument to the constructor, in which case it must be one of

• a string that defines a filename
• a pathlib.Path object
• a bytes object that contains a valid image file content

Your argument falls in none of these categories.

[JorjMcKie]

And because the pixmap constructor correctly excludes the first 2 possibilities, it treats the argument as a bytes object and compares its first dozen or so bytes with the valid image signatures ... and fails of course!

[axlgit]

Hi Jori,

sorry for such a basic mistake!
I corrected the number of brackets but I still receive an error.

C:\Users\axl\Scripts\>python
Python 3.9.1 (tags/v3.9.1:1e5d33e, Dec  7 2020, 17:08:21) [MSC v.1927 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>>
>>> import numpy, fitz
>>> from tools import *
>>>
>>> image=select_images('TEST_page0001_trimmed2_rotated_whitened.png')
>>> image = load_images(image)
>>> image_np_arr = np.ndarray(buffer=image[0].image.write_to_memory(),dtype=np.uint8,shape=[image[0].h,image[0].w,image[0].image.bands])
>>> out_pdf = fitz.open()
>>> fmt = fitz.PaperRect("a4")
>>> out_page = out_pdf.newPage(width = fmt.width, height = fmt.height)
>>>
>>> pix = fitz.Pixmap(fitz.csRGB, image[0].w, image[0].h, bytearray(image_np_arr.tobytes()))
mupdf: bad samples length
Exception ignored in: <function Pixmap.__del__ at 0x000002417FB13820>
Traceback (most recent call last):
  File "C:\Users\axl\AppData\Local\Programs\Python\Python39\lib\site-packages\fitz\fitz.py", line 6691, in __del__
    self.__swig_destroy__(self)
TypeError: in method 'delete_Pixmap', argument 1 of type 'struct Pixmap *'
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Users\axl\AppData\Local\Programs\Python\Python39\lib\site-packages\fitz\fitz.py", line 6389, in __init__
    _fitz.Pixmap_swiginit(self, _fitz.new_Pixmap(*args))
RuntimeError: bad samples length
>>>
>>> fitz.version
('1.18.9', '1.18.0', '20210226134632')
>>>

Alex

[JorjMcKie]

bad samples length

This error means width * height * (n + alpha) does not equal the length of the bytes object, which is used as samples in the constructor.
The factor (n + alpha) means: n = 3 for RGB, 1 for GRAY, 4 for CMYK, alpha is 0 or 1 depending on whether an alpha channel is present.

[JorjMcKie]

It is a little confusing - but only at first sight:
Both, classes Pixmap and Colorspace have an attribute named n. There relationship in a concrete pixmap is pix.n = colorspace.n + pix.alpha. So if you for an RGB image, you always have colorspace.n = 3, but its pix.n may be 3 or 4, depending on whether it is transparent or not.
Similarly for the subtractive colorspace CMYK, which can have pix.n 4 or 5.

[JorjMcKie]

All questions answered?

[axlgit]

Hi Jorj,

sorry for the late reply but I could resume work on this only today.

>>> import fitz
>>> import pyvips as pv
>>> import numpy as np
>>>
>>> FILENAME_PNG = 'Screenshot.png'
>>> FILENAME_PDF = 'TEST_PAGE_TEST.pdf'
>>>
>>> out_pdf = fitz.open()
>>> rect = fitz.PaperRect("a4")
>>> out_page = out_pdf.newPage(width = rect.width, height = rect.height)
>>>
>>> a = pv.Image.new_from_file(FILENAME_PNG)
>>> np_arr = np.ndarray(buffer=a.write_to_memory(), dtype=np.uint8,shape=[a.height, a.width, a.bands])
>>> a.hasalpha()
1
>>> a.height
1080
>>> a.width
1920
>>> a.bands
4
>>> a.format
'uchar'

I refer to https://pymupdf.readthedocs.io/en/latest/pixmap.html
From the output above I understand my image is standard RGB (colorspace.n = 3) and has the alpha channel, hence in calling fitz.Pixmap I have to pass values as specified by init(self, colorspace, width, height, samples, alpha) being colorspace = csRGB and alpha = True but the option alpha is not recognized.

>>> pix = fitz.Pixmap(fitz.csRGB, a.height, a.width, bytearray(np_arr.tobytes()), alpha = True)
Exception ignored in: <function Pixmap.__del__ at 0x000002A5A9C961F0>
Traceback (most recent call last):
  File "C:\Users\axl\AppData\Local\Programs\Python\Python39\lib\site-packages\fitz\fitz.py", line 6691, in __del__
    self.__swig_destroy__(self)
TypeError: in method 'delete_Pixmap', argument 1 of type 'struct Pixmap *'
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: __init__() got an unexpected keyword argument 'alpha'

ImageMagick Identify also says the image is sRGB with alpha channel, see attachment
Screenshot.txt
What I'm doing wrong? And what would be the right way to call Pixmap?

TIA, Alex

[JorjMcKie]

alpha is not a keyword! Simply put in 0 (no alpha) or 1.

[axlgit]

Great, now it all works, thanks for your patience with me!