How can I re-compress an image within a PDF?

Question

I have a PDF document which I want to reduce in size. I've already tried PyPDF2 File Size reductions:

from PyPDF2 import PdfReader, PdfWriter

reader = PdfReader("example.pdf")
writer = PdfWriter()

for page in reader.pages:
    page.compress_content_streams()  # This is CPU intensive!
    writer.add_page(page)
writer.add_metadata(reader.metadata)

with open("out.pdf", "wb") as f:
    writer.write(f)

and I used

mutool clean -czggggfDi input.pdf output.pdf

but that didn't really change the size.

Using mutool extract input.pdf I can see that the 7.3 MB of the PDF are likely dominated by 4 images (img-0317.png: 6.4 MB and img-0201.jpg: 3.5 MB and img-0095.png: 1.3 MB and img-0133.png: 1.1 MB). Just storing those PNG images as JPG with gimp reduced the size from 6.4 MB to 1.4 MB.

How can I exchange the embedded images without changing anything else?

Answer 1

You can try Ghostscript (see here and, for optimizations, this German article) – or, for even more compression, pdfsizeopt (see here). The latter takes a lot longer, but gets better results on scanned text pages.

Ghostscript

Should be provided by your package management. On Debian/Ubuntu/Mint etc, you can install it using sudo apt install ghostscript. You then can use a shell script wrapper like the one I compiled going by the above sources:

#!/bin/bash
# Compress/Optimize PDF files
# see: https://askubuntu.com/a/256449/68291
# and: https://www.linux-community.de/ausgaben/linuxuser/2019/04/optimal-kombiniert/

[[ -z "$1" ]] && {
  echo
  echo "Compress/Optimize PDF files"
  echo "Syntax:"
  echo "  ${0} <InFile> [OutFile] [level]"
  echo ""
  echo "Levels:"
  echo "- screen:    72dpi (low res)"
  echo "- ebook:    150dpi (OK; default if not specified otherwise)"
  echo "- prepress: 300dpi (Acrobat Distiller 'Prepress Optimized')"
  echo "- printer:  300dpi (Acrobat Distiller 'Print Optimized')"
  echo "- default:  ???dpi (auto-select, usually results in bigger file)"
  exit 1
}

[[ ! -f "$1" ]] && {
  echo
  echo "Input File '${1}' not found, nothing to do."
  exit 5
}

outfile="$2"
[[ -z "$2" ]] && outfile=${1%\.pdf}_comp.pdf

[[ -f "$outfile" ]] && {
  echo
  echo "The specified output file '$outfile' does already exist, aborting."
  exit 3
}

case "$3" in
  screen)   level="-dPDFSETTINGS=/screen" ;;      #  72dpi
  prepress) level="-dPDFSETTINGS=/prepress" ;;    # 300dpi
  printer)  level="-dPDFSETTINGS=/printer" ;;     # 300dpi
  default)  level="-dPDFSETTINGS=/default" ;;     # auto-estimate
  *)        level="-dPDFSETTINGS=/ebook -dColorImageResolution=150 -dGrayImageResolution=150 -dMonoImageResolution=150 -dMonoImageDownsampleType=/Subsample" ;;       # 150dpi
esac;

gs -sDEVICE=pdfwrite \
-dColorImageDownsampleType=/Bicubic \
-dGrayImageDownsampleType=/Bicubic \
-dMonoImageDownsampleType=/Bicubic \
-dCompatibilityLevel=1.4 $level -dNOPAUSE -dQUIET -dBATCH -sOutputFile=$outfile $1

Called without parameters it tells you its syntax. With just the input PDF as parameter, it compresses images to 150dpi (ebook), writing output to a file having _comp (aka "compressed") appended to its basename.

PdfSizeOpt

This requires quite some binaries downloaded from different places (mostly Github). To make it easier, here's a script performing the "installation" steps – just run it in an empty directory:

wget https://github.com/pts/tif22pnm/releases/download/2014-01-09/png22pnm.xstatic
wget https://github.com/pts/sam2p/releases/download/v0.49.4/sam2p.xstatic
wget http://www.jonof.id.au/files/kenutils/pngout-20200115-linux.tar.gz
wget https://github.com/pts/pdfsizeopt-jbig2/releases/download/2017-01-24/jbig2.xstatic
wget https://raw.githubusercontent.com/pts/pdfsizeopt/master/pdfsizeopt.single

mv png22pnm.xstatic png22pnm
mv sam2p.xstatic sam2p
mv jbig2.xstatic jbig2
tar xf pngout-20200115-linux.tar.gz --strip-components=2 pngout-20200115-linux/amd64/pngout
chmod u+x pdfsizeopt.single png22pnm sam2p

Then you can use the following wrapper script to compress your PDF files (works very good on scanned text pages; not sure if it's really suited for graphical stuff):

#!/bin/bash
# see: https://askubuntu.com/a/1011292
# takes quite long but gets good results on scanned text pages, even leaving OCR intact

[[ -z "$1" ]] && {
  echo
  echo "Compress PDF consisting of scanned text pages."
  echo
  echo "Usage: $0 <infile>"
  echo
  echo "Run 'pdfsizeopt.single --help' for more details."
  echo
  exit
}

./pdfsizeopt.single --v=30 --do-fast-bilevel-images=yes $1 ${1%\.pdf}_opt.pdf

(if you want to call it from anywhere, adjust the path of ./pdfsizeopt.single to point to where the tools have been placed, and make an alias to the script)

I decided to include the --do-fast-bilevel-images=yes parameter as it speeds up processing a lot (on a test PDF: about 2min without the parameter, less than a second with the parameter), plus --v=30 to reduce output to a usable level. There are many more parameters you can play with to optimize results to your liking.