How to find out which fonts are referenced and which are embedded in a PDF document

Embedding fonts retrospectivly which were not embedded in the original PDF does increase the file size, not decrease it.

However, there may still be a chance to reduce the overall file size by reducing the resolution of embedded images... depends on your preferences and needs.

You can try with variations of the following commandline. It will embed all fonts (even the "Base 14" ones), but embed required glyphs only (a "subset" of the original font), and also compress the fonts:

gs \
   -dCompatibilityLevel=1.4 \
   -dPDFSETTINGS=/screen \
   -dCompressFonts=true \
   -dSubsetFonts=true \
   -dNOPAUSE \
   -dBATCH \
   -sDEVICE=pdfwrite \
   -sOutputFile=output.pdf \
   -c ".setpdfwrite <</NeverEmbed [ ]>> setdistillerparams" \
   -f input.pdf

You have several options. All these methods work on Linux as well as on Windows or Mac OS X. However, be aware that most PDFs do not include to full, complete fontface when they have a font embedded. Mostly they include just the subset of glyphs used in the document.

Using pdftops

One of the most frequently used methods to do this on *nix systems consists of the following steps:

1. Convert the PDF to PostScript, for example by using XPDF's pdftops (on Windows: pdftops.exe helper program.
2. Now fonts will be embedded in .pfa (PostScript) format + you can extract them using a text editor.
3. You may need to convert the .pfa (ASCII) to a .pfb (binary) file using the t1utils and pfa2pfb.
4. In PDFs there are never .pfm or .afm files (font metric files) embedded (because PDF viewer have internal knowledge about these). Without these, font files are hardly usable in a visually pleasing way.

Using fontforge

Another method is to use the Free font editor FontForge:

1. Use the "Open Font" dialogbox used when opening files.
2. Then select "Extract from PDF" in the filter section of dialog.
3. Select the PDF file with the font to be extracted.
4. A "Pick a font" dialogbox opens -- select here which font to open.

Check the FontForge manual. You may need to follow a few specific steps which are not necessarily straightforward in order to save the extracted font data as a file which is re-usable.

Using mupdf

Next, MuPDF. This application comes with a utility called pdfextract (on Windows: pdfextract.exe) which can extract fonts and images from PDFs. (In case you don't know about MuPDF, which still is relatively unknown and new: "MuPDF is a Free lightweight PDF viewer and toolkit written in portable C.", written by Artifex Software developers, the same company that gave us Ghostscript.)
_{(Update: Newer versions of MuPDF have moved the former functionality of 'pdfextract' to the command 'mutool extract'. Download it here: mupdf.com/downloads)}

Note: pdfextract.exe is a command-line program. To use it, do the following:

c:\>  pdfextract.exe  c:\path\to\filename.pdf         # (on Windows)
$>    pdfextract  /path/tofilename.pdf                # (on Linux, Unix, Mac OS X)

This command will dump all of the extractable files from the pdf file referenced into the current directory. Generally you will see a variety of files: images as well as fonts. These include PNG, TTF, CFF, CID, etc. The image names will be like img-0412.png if the PDF object number of the image was 412. The fontnames will be like FGETYK+LinLibertineI-0966.ttf, if the font's PDF object number was 966.

CFF (Compact Font Format) files are a recognized format that can be converted to other formats via a variety of converters for use on different operating systems.

Again: be aware that most of these font files may have only a subset of characters and may not represent the complete typeface.

Update: (Jul 2013) Recent versions of mupdf have seen an internal reshuffling and renaming of their binaries, not just once, but several times. The main utility used to be a 'swiss knife'-alike binary called mubusy (name inspired by busybox?), which more recently was renamed to mutool. These support the sub-commands info, clean, extract, poster and show. Unfortunatey, the official documentation for these tools isn't up to date (yet). If you're on a Mac using 'MacPorts': then the utility was renamed in order to avoid name clashes with other utilities using identical names, and you may need to use mupdfextract.

To achieve the (roughly) equivalent results with mutool as its previous tool pdfextract did, just run mubusy extract ....*

So to extract fonts and images, you may need to run one of the following commandlines:

c:\>  mutool.exe extract filename.pdf      # (on Windows)
$>    mutool     extract filename.pdf      # (on Linux, Unix, Mac OS X)

Downloads are here: mupdf.com/downloads

Using gs (Ghostscript)

Then, Ghostscript can also extract fonts directly from PDFs. However, it needs the help of a special utility program named extractFonts.ps, written in PostScript language, which is available from the Ghostscript source code repository.

Now use it, you need to run both, this file extractFonts.ps and your PDF file. Ghostscript will then use the instructions from the PostScript program to extract the fonts from the PDF. It looks like this on Windows (yes, Ghostscript understands the 'forward slash', /, as a path separator also on Windows!):

gswin32c.exe                  ^
  -q -dNODISPLAY              ^
   c:/path/to/extractFonts.ps ^
  -c "(c:/path/to/your/PDFFile.pdf) extractFonts quit"

or on Linux, Unix or Mac OS X:

gs                          \
  -q -dNODISPLAY            \
   /path/to/extractFonts.ps \
  -c "(/path/to/your/PDFFile.pdf) extractFonts quit"

I've tested the Ghostscript method a few years ago. At the time it did extract *.ttf (TrueType) just fine. I don't know if other font types will also be extracted at all, and if so, in a re-usable way. I don't know if the utility does block extracting of fonts which are marked as protected.

Using pdf-parser.py

Finally, Didier Stevens' pdf-parser.py: this one is probably not as easy to use, because you need to have some know-how about internal PDF structures. pdf-parser.py is a Python script which can do a lot of other things too. It can also decompress and extract arbitrary streams from objects, and therefore it can extract embedded font files too.

But you need to know what to look for. Let's see it with an example. I have a file named big.pdf. As a first step I use the -s parameter to search the PDF for any occurrence of the keyword FontFile (pdf-parser.py does not require a case sensitive search):

pdf-parser.py -s fontfile big.pdf

In my case, for my big1.pdf, I get this result:

obj 9 0
 Type: /FontDescriptor
 Referencing: 15 0 R
  <<   
    /Ascent 728
    /CapHeight 716
    /Descent -210 
    /Flags 32
    /FontBBox [ -665 -325 2000 1006 ]
    /FontFile2 15 0 R
    /FontName /ArialMT
    /ItalicAngle 0
    /StemV 87
    /Type /FontDescriptor
    /XHeight 519
  >>   

obj 11 0 
 Type: /FontDescriptor
 Referencing: 16 0 R
  <<   
    /Ascent 728
    /CapHeight 716
    /Descent -210 
    /Flags 262176
    /FontBBox [ -628 -376 2000 1018 ]
    /FontFile2 16 0 R
    /FontName /Arial-BoldMT
    /ItalicAngle 0
    /StemV 165
    /Type /FontDescriptor
    /XHeight 519
  >>   

It tells me that there are two instances of FontFile2 inside the PDF, and these are in PDF objects no. 15 and no. 16, respectively. Object no. 15 holds the /FontFile2 for font /ArialMT, object no. 16 holds the /FontFile2 for font /Arial-BoldMT.

To show this more clearly:

pdf-parser.py -s fontfile big1.pdf | grep -i fontfile
  /FontFile2 15 0 R
  /FontFile2 16 0 R

A quick peeking into the PDF specification reveals the the keyword /FontFile2 relates to a 'stream containing a TrueType font program' (/FontFile would relate to a 'stream containing a Type 1 font program' and /FontFile3 would relate to a 'stream containing a font program whose format is specified by the Subtype entry in the stream dictionary' {hence being either a Type1C or a CIDFontType0C subtype}.)

To look specifically at PDF object no. 15 (which holds the font /ArialMT), one can use the -o 15 parameter:

pdf-parser.py -o 15 big1.pdf

 obj 15 0
  Type: 
  Referencing: 
  Contains stream
   <<
     /Length1 778552
     /Length 1581435
     /Filter /ASCIIHexDecode
   >>

This pdf-parser.py output tells us that this object contains a stream (which it will not directly display) that has a length of 1.581.435 Bytes and is encoded ( == "compressed") with ASCIIHexEncode and needs to be decoded ( == "de-compressed" or "filtered") with the help of the standard /ASCIIHexDecode filter.

To dump any stream from an object, pdf-parser.py can be called with the -d dumpname parameter. Let's do it:

pdf-parser.py -o 15 -d dumped-data.ext big1.pdf

Our extracted data dump will be in the file named dumped-data.ext. Let's see how big it is:

ls -l dumped-data.ext
  -rw-r--r--  1 kurtpfeifle  staff  1581435 Apr 11 00:29 dumped-data.ext

Oh look, it is 1.581.435 Bytes. We saw this figure in the previous command's output. Opening this file with a text editor confirms that its content is ASCII hex encoded data.

Opening the file with a font reading tool like otfinfo (this is a part of the lcdf-typetools package) will lead to some disappointment at first:

otfinfo -i dumped-data.ext
  otfinfo: dumped-data.ext: not an OpenType font (bad magic number)

OK, this is because we did not (yet) let pdf-parser.py make use of its full magic: to dump a filtered, decoded stream. For this we have to add the -f parameter:

pdf-parser.py -o 15 -f -d dumped-data-decoded.ext big1.pdf

What's the size is this new file?

ls -l dumped-data-decoded.ext
  -rw-r--r--  1 kurtpfeifle  staff  778552 Apr 11 00:39 dumped-data-decoded.ext

Oh, look: that exact number was also already stored in the PDF object no. 15 dictionary as the value for key /Length1...

What does file think it is?

file dumped-data-decoded.ext
  dumped-data-decoded.ext: TrueType font data

What does otfinfo tell us about it?

otfinfo -i dumped-data-decoded.ext
  Family:              Arial
  Subfamily:           Regular
  Full name:           Arial
  PostScript name:     ArialMT
  Version:             Version 5.10
  Unique ID:           Monotype:Arial Regular:Version 5.10 (Microsoft)
  Designer:            Monotype Type Drawing Office - Robin Nicholas, Patricia Saunders 1982
  Manufacturer:        The Monotype Corporation
  Trademark:           Arial is a trademark of The Monotype Corporation.
  Copyright:           © 2011 The Monotype Corporation. All Rights Reserved.
  License Description: You may use this font to display and print content as permitted by
                       the license terms for the product in which this font is included.
                       You may only (i) embed this font in content as permitted by the 
                       embedding restrictions included in this font; and (ii) temporarily 
                       download this font to a printer or other output device to help
                       print content.
  Vendor ID:           TMC

So Bingo!, we have a winner: pdf-parser.py did indeed extract a valid font file for us. Given the size of this file (778.552 Bytes), it looks like this font had been embedded even completely in the PDF...

We could rename it to arial-regular.ttf and install it as such and happily make use of it.

Caveats:

• In any case you need to follow the license that applies to the font. Some font licences do not allow free use and/or distribution. Pirating fonts is like pirating any software or other copyrighted material.

• Most PDFs which are in the wild out there do not embed the full font anyway, but only subsets. Extracting a subset of a font is only useful in a very limited scope, if at all.

Please do also read the following about Pros and (more) Cons regarding font extraction efforts:

• http://typophile.com/node/34377 — not available anymore, but can bee seen on Wayback Machine at https://web.archive.org/web/20110717120241/typophile.com/node/34377