Can pdflatex (or any tex package) automatically rescale included images which have been reduced in size?

Question

I'm writing my thesis in LaTeX, generating it with pdflatex. I have a large number of figures, many of which are bitmaps (as opposed to SVG) in PNG/JPEG format. I've generally created them to be fairly high resolution (say 1600x1200-ish) to ensure that whatever size they end up in the document, they'll be at least 300dpi when printed.

As I'm writing/laying out the document, I'm including graphics (using \includegraphics from the graphicx package) and setting widths/heights as appropriate (e.g. subfigures are quite small). I don't need the images to be any more than about 300 dpi at best, so where I have shrunk a 1600x1200 image down to say 5cm, the image is now at 800 dpi. So despite including some very small (on the page) images, the PDF is becoming quite large.

Is there a way to tell pdflatex or graphicx (or something else involved?) to convert all images to a maximum of 300 dpi, based on the dimensions I'm setting with say \includegraphics[width=2in]{filename}? i.e. so it scales the image to a max of 600x600 pixels as it includes it in the PDF (leaving the original file untouched).

I know I can resize the original images with various command line applications, and include the pre-resized versions, but given the images vary in size considerably, it wouldn't be as simple as making sure they're all 300dpi for a constant printed size. It'd also be nice to be able to easily create different versions of PDFs (web vs final print) without resizing images manually, so that the 'web' PDF capped images at say 72-100 dpi while the final print one could cap at 600 (if at all).

Update: I've found a poor solution to the problem in some cases (quite long so I've added it as an answer myself, see below). By reprocessing the PDF after creation, you can convert all images to JPEGs at a specific resolution (in dpi). This is good for docs with only JPEGs, but sucks if you have PNG (or similar non-lossy-compression) diagrams which are PNG for a reason (e.g. simple diagrams with hard edges).

So I'm still after a solution that can do this without forcing the same compression algorithm on all images (whether it's a LaTeX macro/package, or a script that parses the LaTeX and produces appropriately sized files, or a post-processor that doesn't take a one-compression-algorithm-fits-all approach).

Answer 1

I've tried a little black magic with LaTeX, and found out how to do it without external scripts ─ though it still is somewhat dirty and not that easy to use. Basically, you would need to call ImageMagick from inside LaTeX.

To do this, you need to call it with --shell-escape parameter. This allows for using a command named \write18, that executes a parameter passed to it. Then you could add the following code to the preamble:

\newcommand{\doConvert}[2]{\immediate\write18{convert #1 -resample %
  300x300 -resize #2 #1.rs}}
\newcommand{\imgRs}[4]{\doConvert{#1.#2}{#3x#4}%
  \includegraphics[width=#3, height=#4, type=#2, ext=.#2.rs, read=.#2.rs]{#1}%
}

This enables you to call \imgRs{file}{extension}{width_in_pts}{height_in_pts} in order to convert the image to 300 dpi and include it in the document.

Of course you could always specify the dimensions in inches or any other units: you would just have to change the call to \doConvert in imgRs to account for scaling from unit used to points -- and you would have to use the dimensions consistently. It could be possible to actually find out what were the units used, or make some parameters optional, but it's independent of this solution.

Answer 2

You can post-process the PDF and still use lossless compression. I have also had to learn a lot about manipulating PDF files for / from LaTeX when writing my thesis. Basically, you can use GhostScript and tell it what compression method you want for, say, color images (Win32 example, but you get the idea):

gswin32c.exe -sDEVICE=pdfwrite -dMaxSubsetPct=100 -dPDFSETTINGS=/prepress -dAutoFilterColorImages=false -dColorImageFilter=/FlateEncode -sOutputFile="outfile.pdf" -dNOPAUSE -dBATCH "infile.pdf"

You need to specify both -dAutoFilterColorImages=false (don't automatically choose the compression method) and -dColorImageFilter=/FlateEncode (use the Flate encoder).

You need to specify the corresponding parameters for Gray images as well if you want to control those.

Ghostscript resamples to a given dpi based on dPDFSETTINGS:

prepress = 300 dpi

printer = 300 dpi

ebook = 150 dpi

screen = 72 dpi

I have written a short tutorial on my website about manipulating and compressing PDF files, to explain this in more detail.

Answer 3

Sorry to use up space for yet another answer - just wanted to have the nice formatting :)

Well, thanks to answers by @finrod and @Ken Bloom; and additionally How to execute shell script from LaTeX?; How can I save shell output to a variable in LaTeX? and Scale (resize) large images (graphics) that exceed page margins; I managed to cook up a "script" (well, rather a macro), such that:

• if pdflatex is called with -shell-escape; then \includegraphics is redefined to call the resampling process through the shell, and include the downsampled image;
• else, \includegraphics stays as is (i.e. it will include the large images)

Here's the code (note that even within \write18, the percent sign '%' still functions as a comment):

% for downsampling large images

\ifnum\pdfshellescape=1
% Yes, enabled
% #2 is the filename - assumed relative
% IFP - image file path
% IFN - image filename only
% here resizing at 800 px wide
\let\Oldincludegraphics\includegraphics
\newread\myinput
% must add default argument [] so typeout works
\renewcommand\includegraphics[2][\@empty]{%
% debug - print arguments to stdout:
\immediate\message{== #1 ==}
\immediate\typeout{== #2 ==}
% run downsampling bash script
\immediate\write18{mkdir -p downsampled; IFP=#2 ; %
echo "Processing $IFP" ; %
IFN=$(basename $IFP) ; %
echo "downsampled/$IFN" > tmpname ; %
if [ ! -f downsampled/$IFN ]; then %
echo "File downsampled/$IFN not found! Converting..." ; %
% convert $IFP -resize 50\% downsampled/$IFN ; %
convert $IFP -resize 800x downsampled/$IFN ; %
else %
echo "Found downsampled/$IFN - reusing." ; %
fi ; %
} % end write18
% here we should have downsampled file
% retrieve downsample filename first
\immediate\openin\myinput=tmpname
% The group localizes the change to \endlinechar
\bgroup
  \endlinechar=-1
  \immediate\read\myinput to \localline
  % Since everything in the group is local, we have to explicitly make the
  % assignment global
  \global\let\myTmpFileName\localline
\egroup
\immediate\closein\myinput
\immediate\typeout{== \myTmpFileName .. #1 ==}
% Clean up after ourselves
% \immediate\write18{rm -f -- tmpname}
% finally, include downsampled image
\Oldincludegraphics[#1]{\myTmpFileName}
} % end renewcommand
\else
% No, disabled
\fi

EDIT: and here is an example where instead of redefining, a new \includegraphicsRS is defined (so it can be mixed with usual \includegraphics) with the same behavior as above:

% for downsampling large images: \includegraphicsRS
% (\includegraphics "resize")

\newcommand{\includegraphicsRS}{}

\ifnum\pdfshellescape=1
% Yes, enabled
% #2 is the filename - assumed relative
% IFP - image file path
% IFN - image filename only
% here resizing at 800 px wide
\newread\myinput
% must add default argument [] so typeout works
\renewcommand\includegraphicsRS[2][\@empty]{%
% run downsampling bash script
\immediate\write18{mkdir -p downsampled; IFP=#2 ; %
  echo "Processing $IFP" ; %
  IFN=$(basename $IFP) ; %
  echo "downsampled/$IFN" > tmpname ; %
  if [ ! -f downsampled/$IFN ]; then %
  echo "File downsampled/$IFN not found! Converting..." ; %
  % convert $IFP -resize 50\% downsampled/$IFN ; %
  convert $IFP -resize 800x downsampled/$IFN ; %
  else %
  echo "Found downsampled/$IFN - reusing." ; %
  fi ; %
} % end write18
% here we should have downsampled file
% retrieve downsample filename first
\immediate\openin\myinput=tmpname
% The group localizes the change to \endlinechar
\bgroup
  \endlinechar=-1
  \immediate\read\myinput to \localline
  % Since everything in the group is local, we have to explicitly make the
  % assignment global
  \global\let\myTmpFileName\localline
\egroup
\immediate\closein\myinput
% Clean up after ourselves
% \immediate\write18{rm -f -- tmpname}
% finally, include downsampled image
\includegraphics[#1]{\myTmpFileName}
} % end renewcommand
\else
% No, disabled
% in this case, \includegraphicsRS is just the usual \includegraphics
\renewcommand\includegraphicsRS[2][\@empty]{%
\includegraphics[#1]{#2}
}
\fi

Answer 4

Try the degrade package, which performs the ImageMagick trick that others have written about.

Answer 5

You may change resolution automatically using ImageMagick's convert -resample. Here is an awk script that parses the tex file, finds \includegraphics statements and then converts images embedded there to a required resolution:

(/\\includegraphics/&&(/{.*\.png}/||/{.*\.jpg}/)){
 if(match($0,"width=[0123456789\.]*in")==0){
    print "%Cannot convert, no width in inch!\n" $0;
 }else{
    wid=substr($0,RSTART+6,RLENGTH-8);  
    pixSize=wid*reso;  
    match($0,"{.*}");file=substr($0,RSTART+1,RLENGTH-2);  
    print "%convert " file " -resample " reso "x" reso " -resize " pixSize " RESAMPLED_" file;   
    system("convert " file " -resample " reso "x" reso " -resize " pixSize " RESAMPLED_" file);  
    gsub("{.*}","{RESAMPLED_" file "}",$0); print $0;
 }
}

!(/\\includegraphics/&&(/{.*\.png}/||/{.*\.jpg}/)){print $0}

you may use it like so (assuming it is in file resample.awk):

$ awk -f resample.awk -v reso=300 latexfile.tex > resampledlatexfile.tex
$ pdflatex resampledlatexfile.tex ...

Note that to work properly it needs \includegraphics to have width parameter specified, and specified in inch; if this is not satisfied, it will skip such picture.
It will create a series of resampled images (with RESAMPLED_ prefix), embedded in the resampledlatexfile.tex.

Answer 6

I've found one solution is to post-process the PDF, re-saving it in an application which will resize the images in the PDF to a maximum resolution.

Adobe Acrobat Pro (not just Acrobat Reader) can do this via File / Save As / Optimized PDF. Have recently tested this on Mac OS X (and so assume it works on Windows too) with Acrobat Pro X. You can specify what resolution to reprocess to, and there's a few options for compression formats, including Retain Existing which keeps JPG/PNGs as they were. Probably the easiest way for a one-off conversion, e.g. for a final manuscript for mobile device/web download... assuming you have access to Acrobat Pro.

EDIT: Latest Mac OS X (10.8) seems to include a filter to reduce PDF file size already; simply go to File / Export and select the Reduce File Size Quartz Filter. Unfortunately this seems to, like the method below, convert ALL images to JPEG, introducing unnecessary noise in files that were PNG.

For older versions: On Mac OS X, I've found this post on the Apple Support forums which details a method of using Quartz filters and the Preview app (PDF/image viewer) to let you save a PDF with a reduced maximum image resolution for the whole document (75,150,300,600 dpi). The filter converts ALL images in the document to JPEG with the specified resolution (dpi) and specified quality (low/average).

So you can't leave certain images alone (e.g. diagrams in GIF/PNG form that are already small in file size, which may look awful as JPEGs), which would be the only real complaint using this method. There doesn't seem to be a way of preserving the original image format.

Set-up (required once only of course, instructions copied from the original post):

1. get the filters from here in a .zip file
2. unzip the archive, it'll create a 'Filters' folder
3. move the 'Filters' folder into your ~/Library folder

After compiling a PDF, reduce unnecessarily high image resolutions:

1. open the PDF in Preview.app
2. select Save As... from the file menu
3. Select the appropriate resolution in the 'Quartz Filter' drop-down (e.g. "Reduce to 300 dpi average quality")