PIL's Image.save('image.pdf') creates excessivly large PDF

Question

I have an low-resolution black & white image screenBWsmall.png:

I use the Python Imaging Library to convert it to PDF:

#!python
from PIL import Image 
im = Image.open('screenBWsmall.png')
im.save('screenBWsmall.pdf')

The PDF file is huge compared to one generated from ImageMagick's convert, issued from the Bash command line:

convert screenBWsmall.png screenBWsmall_IM.pdf

The file sizes are:

  11093 screenBWsmall.png
1050994 screenBWsmall.pdf
  16999 screenBWsmall_IM.pdf

While I'm puzzled by this, it is even more puzzling considering that the larger file screenBWsmall.pdf uses 1 bit per pixel (bits per component, or bpc) compared to 8 bpc for the smaller file screenBWsmall_IM.pdf:

$ pdfimages.exe -list screenBWsmall.pdf
page   num  type   width height color comp bpc  enc interp  object ID x-ppi y-ppi size ratio
--------------------------------------------------------------------------------------------
   1     0 image     960   540  gray    1   1  image  no         1  0    72    72 1025K 1621%

$ pdfimages.exe -list screenBWsmall_IM.pdf
page   num  type   width height color comp bpc  enc interp  object ID x-ppi y-ppi size ratio
-------------------------------------------------------------------------------------------- 
   1     0 image     960   540  gray    1   8  image  no         8  0    72    72 14.9K 2.9%

The Image.save documentation doesn't give much information with which I can speculate on the reason for the large file size.

Why does PIL create such a large PDF file size?

Is there any way to have it create the smaller size of ImageMagick's convert? I want to do it in Python because I will be performinng more complex steps with many files.

My Python version is:

Python 3.8.8 (default, Mar  4 2021, 21:24:42) 
[GCC 10.2.0] on cygwin

---

Investigations with ImageMagick's convert

Thanks to fmw42's suggestions, I systematically experimented with 3 ways to shrink and combine 2 JPG images to into 1 PDF file. In order of decreasing file size, the 3 methods are as follows.

Method #1: Use Python's PIL to generate IMG_077x_PIL.pdf (see jpg2pdf.py below). In the process of doing so, save shrunken versions of both images to separate PNG files for Method #3.

Method #2: Use ImageMagick's convert to generate IMG_077x_IMcvt.pdf:

convert -sample 50% -type Bilevel +dither IMG_077[45].JPG IMG_077x_IMcvt.pdf

Method #3: Apply convert to shrunken PNG files from PIL to generate IMG_077x_PIL+IMcvt.pdf:

convert IMG_077[45]small.png IMG_077x_PIL+IMcvt.pdf

Output PDF file sizes (in the same order as the methods above):

12350481 IMG_077x_PIL.pdf
 1234076 IMG_077x_IMcvt.pdf
  149782 IMG_077x_PIL+IMcvt.pdf

The 2 input JPG files sizes are a few MBs:

 2526685 IMG_0774.JPG
 2699515 IMG_0775.JPG

The 2 intermediate PNG file sizes used in Methods #1 and #3 are few dozen KBs:

   67283 IMG_0775small.png
   61968 IMG_0774small.png

Observations:

• Method #1: Great for shrinking the images down, but really bad in generating an enormous PDF file that is two orders of magnitude larger than it has to be.

• Method #2: Middle of the road, most convenient, but PDF file size is an order of magnitude larger than it has to be.

• Method #3: Requires both Python, PIL, and convert. It is the least convenient, but most byte efficient. The resulting PDF is only slightly larger than the sum of the two PNG images.

I wish there was a way to make Methods #1 and/or #2 as good as Method #3.

Characteristics of the output PDF files

page   num  type   width height color comp bpc  enc interp  object ID x-ppi y-ppi size ratio
--------------------------------------------------------------------------------------------
$pdfimages -list IMG_077x_PIL.pdf
   1     0 image    1512  2016  gray    1   1  image  no         1  0    72    72 6030K 1621%
   2     1 image    1512  2016  gray    1   1  image  no         4  0    72    72 6030K 1621%

$pdfimages -list IMG_077x_IMcvt.pdf
   1     0 image    2016  1512  gray    1   8  jpeg   no         8  0    72    72  576K  19% 
   2     1 image    2016  1512  gray    1   8  jpeg   no        22  0    72    72  626K  21% 

$pdfimages -list IMG_077x_PIL+IMcvt.pdf
   1     0 image    1512  2016  gray    1   8  image  no         8  0    72    72 68.9K 2.3% 
   2     1 image    1512  2016  gray    1   8  image  no        22  0    72    72 74.6K 2.5% 

jpg2pdf.py

#!python

# jpg2pdf.py
#-----------
# Use PIL to subsample, rotate, and convert 2 JPGs to B&W.
# Save each to small PNGs.
# Combine both into a PDF.

import os
from PIL import Image

ims = [] # Stores the 2 images
fns=('IMG_0774.JPG','IMG_0775.JPG') # Filenames of the 2 images

for fn in fns:
   
   # Read, resize, rotate, convert to B&W, add to list `fns`
   im = Image.open(fn)
   im = im.resize((im.width//2, im.height//2))
   im = im.rotate(-90,expand=True)
   im = im.convert(mode="1", dither=Image.NONE)
   ims.append(im)

   # Write IMG_077[45]small.png
   fnBase = os.path.splitext(fn)[0]
   im.save( fnBase+'small.png' )

# Write both to a single PDF
ims[0].save( 'IMG_077x_PIL.pdf' , save_all=True , append_images=ims[1:] )

A test input file

This dummy JPEG image file should be save-able as both IMG_0774.JPGand IMG_0775.JPG. Methods #1 through #3 should then work exactly as described with the code posted above. Using this JPG image, I confirmed that the 3 output file sizes are almost the same as reported in my question. Being just over 2MB, unfortunately, it can't be uploaded to this posted question.

Answer 1

Why does PIL create such a large PDF file size?

Unfortunately, Pillow re-encodes black & white image by DCTDecode (that is used for JPEG):
https://github.com/python-pillow/Pillow/blob/0f44136e720cd3b2db72bdf29614897b7aa3e868/src/PIL/PdfImagePlugin.py#L127

$ pdfimages -list screenBWsmall.pdf
page   num  type   width height color comp bpc  enc interp  object ID x-ppi y-ppi size ratio
--------------------------------------------------------------------------------------------
   1     0 image     960   540  gray    1   8  jpeg   no         1  0    72    72 93.1K  18%

You can use img2pdf without re-encode:

from pathlib import Path

import img2pdf

Path("screenBWsmall.pdf").write_bytes(img2pdf.convert("screenBWsmall.png"))

$ pdfimages -list screenBWsmall.pdf 
page   num  type   width height color comp bpc  enc interp  object ID x-ppi y-ppi size ratio
--------------------------------------------------------------------------------------------
   1     0 image     960   540  gray    1   1  image  no         7  0    96    96 10.8K  17%

I tested by:

• Python: 3.10.4
• Pillow: 9.1.1
• img2pdf: 0.4.4
• pdfimages: 20.09.0

Answer 2

I think it's due to the mode of your image and PIL feeling constrained to retaining it as a bi-level image.

If you do this:

im = Image.open('lorem.png')

# Check type of image - it is bi-level, i.e. mode=1
print(im)
<PIL.PngImagePlugin.PngImageFile image mode=1 size=960x540 at 0x7F9E08A65100>


# Save and check size
im.save('lorem.pdf')

# -rw-r--r--     1 mark  staff  1050978 18 Apr 10:01 lorem.pdf   <--- YIKES

If you tell PIL it's ok to treat it as colour, it works fine:

im = Image.open('lorem.png').convert('RGB')
im.save('lorem.pdf')

# -rw-r--r--     1 mark  staff    99556 18 Apr 10:02 lorem.pdf  <--- THAT'S BETTER

---

There is a clue in the documentation here where it says the way it is written depends on the mode and the availability of the JPEG encoder.