Howto: Make scanned PDFs searchable (OCR) using pdfocr

[tuxcantfly]

What pdfocr is for

Suppose you have a PDF document that was made using a scanner, or otherwise consists of image data but doesn't have text data. Such a PDF can't be searched by PDF readers or desktop search applications. pdfocr is a simple utility I made that takes a PDF file, then generates a new one that has the text layer added, so it's searchable by your PDF reader and can be indexed by your desktop search application, but is still identical when printed.

What pdfocr is not for

This is only of use if your PDF was made from a scanned source; if you exported your PDF from OpenOffice or the like it already has a text layer so this is unnecessary.

If what you're looking for is to simply extract the plain text from a PDF file, but not to embed the text into the PDF file, see this guide.

Compatibility

This guide will work on Ubuntu Karmic (9.10) or Lucid (10.04); the dependencies for this software don't build on older versions.

Installing pdfocr

The easiest way to install pdfocr is to add my PPA and use apt-get. If you would instead prefer to install it manually, see here for instructions

sudo add-apt-repository ppa:gezakovacs/pdfocr
sudo apt-get update
sudo apt-get install pdfocr

Using pdfocr to add a text layer to your scanned PDF file

Open a terminal, go to the directory that has the PDF file you want to convert, and enter (substituting input.pdf with the input PDF file, and output.pdf with the output PDF file)

pdfocr -i input.pdf -o output.pdf

Now wait as OCR is performed on the PDF file page-by-page, and the output file is generated. This should take a few seconds per page, depending on the resolution of your PDF file (high-res PDF files get better accuracy, but will take longer). Once done, you should now have a searchable PDF at output.pdf.

Credits

pdfocr was written by me (Geza Kovacs). It is simply a script which automates the following process:

1. Splitting the PDF file into separate pages using pdftk
2. Extracting out the image data using pdfimages
3. Doing OCR (optical character recognition) using cuneiform
4. Embedding the detected text back into the PDF file using hocr2pdf
5. Merging together the files using pdftk.

Hence, if you want more fine-grained control than the defaults, you can just invoke these utilities manually. Source is available on github. Feedback is welcome.

[Yva]

Sounds like an awesome program, unfortunately I've got the following errors any idea why?

$pdfocr -i Acceptance.pdf -o test.pdf
Input file is /mnt/data/tmp/Acceptance.pdf
Output file is /mnt/data/tmp/test.pdf
Using working dir /tmp/d20100419-4215-ohxah0
Getting info from PDF file

Warning: no info dictionary found
NumberOfPages: 1

Converting 1 pages
==========
Extracting page 1
Converting page 1 to ppm
Running OCR on page 1
PUMA_XFinalrecognition failed.
Cuneiform for Linux 0.9.0
Error while running OCR on page 1
Merging together PDF files
Error: Failed to open PDF file:
/tmp/d20100419-4215-ohxah0/*-new.pdf
Errors encountered. No output created.
Done. Input errors, so no output created.
Updating PDF info for /mnt/data/tmp/test.pdf
Error: Failed to open PDF file:
/tmp/d20100419-4215-ohxah0/merged.pdf
Errors encountered. No output created.
Done. Input errors, so no output created.
Cleaning up temporary files

[Sub101]

Very nice. And certainly extremely useful.

Thanks.

I did get a few errors, but none of them stopped the program working.

Errors below.

Warning: Image x/y resolution not set, defaulting to: 300
Warning: tag mismatch: 'b' can not close last open: 'i'
Warning: tag mismatch: 'span' can not close last open: 'b'
Warning: tag mismatch: 'p' can not close last open: 'b'
Warning: tag mismatch: 'div' can not close last open: 'b'
Warning: tag mismatch: 'body' can not close last open: 'b'
Warning: tag mismatch: 'html' can not close last open: 'b'
Warning: unclosed tag: 'b'
Warning: unclosed tag: 'span'
Warning: unclosed tag: 'p'
Warning: unclosed tag: 'div'
Warning: unclosed tag: 'body'
Warning: unclosed tag: 'html'

[a.w.howell]

I am excited about this script, I think it will be very useful. However I ran into a different error than the one posted above:

aaron@aaron-desktop:~$ pdfocr -i 0175.pdf -o out3.pdf
Input file is /home/aaron/0175.pdf
Output file is /home/aaron/out3.pdf
Using working dir /tmp/d20100421-20950-4g11i8
Getting info from PDF file

InfoKey: Creator
InfoValue: XSane version 0.996 (sane 1.0) - by Oliver Rauch
InfoKey: Title
InfoValue: XSane scanned image
InfoKey: Producer
InfoValue: XSane 0.996
InfoKey: CreationDate
InfoValue: D:20100421215339+00'00'
NumberOfPages: 1

Converting 1 pages
==========
Extracting page 1
Converting page 1 to ppm
Running OCR on page 1
1.ppm is not a BMP file.
Cuneiform for Linux 0.9.0
Error while running OCR on page 1
Merging together PDF files
Error: Failed to open PDF file: 
   /tmp/d20100421-20950-4g11i8/*-new.pdf
Errors encountered.  No output created.
Done.  Input errors, so no output created.
Updating PDF info for /home/aaron/out3.pdf
Error: Failed to open PDF file: 
   /tmp/d20100421-20950-4g11i8/merged.pdf
Errors encountered.  No output created.
Done.  Input errors, so no output created.
Cleaning up temporary files

Notice the "1.ppm is not a BMP file." line. I get a similar error if I run cuneiform by itself:

aaron@aaron-desktop:~$ cuneiform -f hocr -o out.hocr 0175.pdf
Cuneiform for Linux 0.9.0
0175.pdf is not a BMP file.

It's as if cuneiform defaults to assuming the input file is a bmp. Is there a way to change this?

I should also mention that I am on amd64, and I initially had problems with cuneiform throwing an error. I had to add /usr/local/lib64 to a .conf file in /etc/ld.so.conf.d, and running ldconfig (per https://answers.launchpad.net/cuneif...uestion/100695). After making this change cuneiform seems to work, but there still could be other unseen issues.

Thanks.

[meho_r]

Thanks a lot, great utility.

[clasikowski]

Hi!

It's as if cuneiform defaults to assuming the input file is a bmp. Is there a way to change this?

Yes; you just have to install

libmagick++-dev

before compiling cuneiform, then it'll be able to recognize just about every input format (all imagemagick is able to use, to be precise)

so long
clasikowski aka hank

[the_summer]

Its a great helper. I tried it on some files. The ocr seem to work, but when i search for words, the marks are often quite far away from the searched word.
Maybe it's because there is always the warning:

Warning: Image x/y resolution not set, defaulting to: 300

Is there a way to manually set different values for the resolution.
I didnt find something about it in the man-page.

Best wishes

Jan

[mdwy62]

Great utility! Just used in on a 60 page document. Many Thanks. In evince, I also find that the placement of the search term results is off. Any suggestions how to improve this would be appreciated.

[nanonils]

Hi!



Yes; you just have to install

libmagick++-dev

before compiling cuneiform, then it'll be able to recognize just about every input format (all imagemagick is able to use, to be precise)

so long
clasikowski aka hank

Hank,
Does that mean install libmagick++-dev before doing "sudo add-apt-repository ppa:gezakovacs/pdfocr
sudo apt-get update
sudo apt-get install pdfocr"?

I do have libmagick++-dev installed and did a reinstall of Geza's pdfocr but I still get the following error output:
pdfocr -i 1.pdf -o 2.pdf
Input file is /home/nils/1.pdf
Output file is /home/nils/2.pdf
Using working dir /tmp/d20100520-15606-f6vqow
Getting info from PDF file

InfoKey: Title
InfoValue: /tmp/simple-scan-RDLXCV.pdf
InfoKey: Producer
InfoValue: ImageMagick 6.5.7-8 2009-11-26 Q16 http://www.imagemagick.org
InfoKey: ModDate
InfoValue: D:20100520074947
InfoKey: CreationDate
InfoValue: D:20100520074947
NumberOfPages: 1

Converting 1 pages
==========
Extracting page 1
Converting page 1 to ppm
Running OCR on page 1
*** buffer overflow detected ***: cuneiform terminated
======= Backtrace: =========
/lib/libc.so.6(__fortify_fail+0x37)[0x7f742e9131a7]
/lib/libc.so.6(+0xfe060)[0x7f742e912060]
/usr/lib/cuneiform/libfon32.so.0.9.0(+0x21a5c)[0x7f7429d93a5c]
/usr/lib/cuneiform/libfon32.so.0.9.0(+0x2226a)[0x7f7429d9426a]
/usr/lib/cuneiform/libfon32.so.0.9.0(FONRecog2Glue+0x1e7)[0x7f7429d81f47]
/usr/lib/cuneiform/libpass2.so.0.9.0(+0x73b3)[0x7f742a8793b3]
/usr/lib/cuneiform/libpass2.so.0.9.0(+0x7592)[0x7f742a879592]
/usr/lib/cuneiform/libpass2.so.0.9.0(+0xab63)[0x7f742a87cb63]
/usr/lib/cuneiform/libpass2.so.0.9.0(p2_proc+0xa7a)[0x7f742a87d85a]
/usr/lib/cuneiform/librstr.so.0.9.0(+0x98e01)[0x7f742ad4de01]
/usr/lib/cuneiform/librstr.so.0.9.0(RSTRRecognizeMain+0x224)[0x7f742ad60184]
/usr/lib/cuneiform/librstr.so.0.9.0(RSTRRecognize+0x19)[0x7f742ad60dc9]
/usr/lib/cuneiform/libcuneiform.so.0.9.0(+0xcafe)[0x7f742f343afe]
/usr/lib/cuneiform/libcuneiform.so.0.9.0(PUMA_XFinalRecognition+0xd1)[0x7f742f345091]
cuneiform[0x403beb]
/lib/libc.so.6(__libc_start_main+0xfd)[0x7f742e832c4d]
cuneiform[0x402a19]
======= Memory map: ========
00400000-00405000 r-xp 00000000 08:01 59229185 /usr/bin/cuneiform
00604000-00605000 r--p 00004000 08:01 59229185 /usr/bin/cuneiform
00605000-00606000 rw-p 00005000 08:01 59229185 /usr/bin/cuneiform
010d0000-04c92000 rw-p 00000000 00:00 0 [heap]
7f741ef2d000-7f741ef31000 r-xp 00000000 08:01 59232850 /usr/lib/ImageMagick-6.5.7/modules-Q16/coders/dib.so
7f741ef31000-7f741f130000 ---p 00004000 08:01 59232850 /usr/lib/ImageMagick-6.5.7/modules-Q16/coders/dib.so
7f741f130000-7f741f131000 r--p 00003000 08:01 59232850 /usr/lib/ImageMagick-6.5.7/modules-Q16/coders/dib.so
7f741f131000-7f741f132000 rw-p 00004000 08:01 59232850 /usr/lib/ImageMagick-6.5.7/modules-Q16/coders/dib.so
7f74236b5000-7f74236b6000 ---p 00000000 00:00 0
7f74236b6000-7f7423eb6000 rw-p 00000000 00:00 0
7f7423eb6000-7f7423eb7000 ---p 00000000 00:00 0
7f7423eb7000-7f74246b7000 rw-p 00000000 00:00 0
7f74246b7000-7f74246b8000 ---p 00000000 00:00 0
7f74246b8000-7f7424eb8000 rw-p 00000000 00:00 0
7f7424eb8000-7f7424ec1000 r-xp 00000000 08:01 59232948 /usr/lib/ImageMagick-6.5.7/modules-Q16/coders/pnm.so
7f7424ec1000-7f74250c0000 ---p 00009000 08:01 59232948 /usr/lib/ImageMagick-6.5.7/modules-Q16/coders/pnm.so
7f74250c0000-7f74250c1000 r--p 00008000 08:01 59232948 /usr/lib/ImageMagick-6.5.7/modules-Q16/coders/pnm.so
7f74250c1000-7f74250c2000 rw-p 00009000 08:01 59232948 /usr/lib/ImageMagick-6.5.7/modules-Q16/coders/pnm.so
7f74250c2000-7f74250c7000 r-xp 00000000 08:01 59231793 /usr/lib/libXdmcp.so.6.0.0
7f74250c7000-7f74252c6000 ---p 00005000 08:01 59231793 /usr/lib/libXdmcp.so.6.0.0
7f74252c6000-7f74252c7000 r--p 00004000 08:01 59231793 /usr/lib/libXdmcp.so.6.0.0
7f74252c7000-7f74252c8000 rw-p 00005000 08:01 59231793 /usr/lib/libXdmcp.so.6.0.0
7f74252c8000-7f74252ca000 r-xp 00000000 08:01 59231782 /usr/lib/libXau.so.6.0.0
7f74252ca000-7f74254ca000 ---p 00002000 08:01 59231782 /usr/lib/libXau.so.6.0.0
7f74254ca000-7f74254cb000 r--p 00002000 08:01 59231782 /usr/lib/libXau.so.6.0.0
7f74254cb000-7f74254cc000 rw-p 00003000 08:01 59231782 /usr/lib/libXau.so.6.0.0
7f74254cc000-7f74254d3000 r-xp 00000000 08:01 50824095 /lib/librt-2.11.1.so
7f74254d3000-7f74256d2000 ---p 00007000 08:01 50824095 /lib/librt-2.11.1.so
7f74256d2000-7f74256d3000 r--p 00006000 08:01 50824095 /lib/librt-2.11.1.so
7f74256d3000-7f74256d4000 rw-p 00007000 08:01 50824095 /lib/librt-2.11.1.so
7f74256d4000-7f74256ef000 r-xp 00000000 08:01 59232788 /usr/lib/libxcb.so.1.1.0
7f74256ef000-7f74258ee000 ---p 0001b000 08:01 59232788 /usr/lib/libxcb.so.1.1.0
7f74258ee000-7f74258ef000 r--p 0001a000 08:01 59232788 /usr/lib/libxcb.so.1.1.0
7f74258ef000-7f74258f0000 rw-p 0001b000 08:01 59232788 /usr/lib/libxcb.so.1.1.0
7f74258f0000-7f74258f4000 r-xp 00000000 08:01 50824123 /lib/libuuid.so.1.3.0
7f74258f4000-7f7425af3000 ---p 00004000 08:01 50824123 /lib/libuuid.so.1.3.0
7f7425af3000-7f7425af4000 r--p 00003000 08:01 50824123 /lib/libuuid.so.1.3.0
7f7425af4000-7f7425af5000 rw-p 00004000 08:01 50824123 /lib/libuuid.so.1.3.0
7f7425af5000-7f7425b02000 r-xp 00000000 08:01 59232191 /usr/lib/libgomp.so.1.0.0
7f7425b02000-7f7425d01000 ---p 0000d000 08:01 59232191 /usr/lib/libgomp.so.1.0.0
7f7425d01000-7f7425d02000 r--p 0000c000 08:01 59232191 /usr/lib/libgomp.so.1.0.0
7f7425d02000-7f7425d03000 rw-p 0000d000 08:01 59232191 /usr/lib/libgomp.so.1.0.0
7f7425d03000-7f7425e34000 r-xp 00000000 08:01 59231778 /usr/lib/libX11.so.6.3.0
7f7425e34000-7f7426034000 ---p 00131000 08:01 59231778 /usr/lib/libX11.so.6.3.0
7f7426034000-7f7426035000 r--p 00131000 08:01 59231778 /usr/lib/libX11.so.6.3.0
7f7426035000-7f7426039000 rw-p 00132000 08:01 59231778 /usr/lib/libX11.so.6.3.0
7f7426039000-7f7426050000 r-xp 00000000 08:01 59231747 /usr/lib/libICE.so.6.3.0
7f7426050000-7f742624f000 ---p 00017000 08:01 59231747 /usr/lib/libICE.so.6.3.0
7f742624f000-7f7426250000 r--p 00016000 08:01 59231747 /usr/lib/libICE.so.6.3.0
7f7426250000-7f7426251000 rw-p 00017000 08:01 59231747 /usr/lib/libICE.so.6.3.0
7f7426251000-7f7426254000 rw-p 00000000 00:00 0
7f7426254000-7f742625c000 r-xp 00000000 08:01 59231776 /usr/lib/libSM.so.6.0.1
7f742625c000-7f742645b000 ---p 00008000 08:01 59231776 /usr/lib/libSM.so.6.0.1
7f742645b000-7f742645c000 r--p 00007000 08:01 59231776 /usr/lib/libSM.so.6.0.1
7f742645c000-7f742645d000 rw-p 00008000 08:01 59231776 /usr/lib/libSM.so.6.0.1
7f742645d000-7f7426465000 r-xp 00000000 08:01 59232423 /usr/lib/libltdl.so.7.2.1
7f7426465000-7f7426665000 ---p 00008000 08:01 59232423 /usr/lib/libltdl.so.7.2.1Cuneiform for Linux 0.9.0
Error while running OCR on page 1
Merging together PDF files
/tmp/d20100520-15606-f6vqow/*-new.pdf not found as file or resource.
Error: Failed to open PDF file:
/tmp/d20100520-15606-f6vqow/*-new.pdf
Errors encountered. No output created.
Done. Input errors, so no output created.
Updating PDF info for /home/nils/2.pdf
/tmp/d20100520-15606-f6vqow/merged.pdf not found as file or resource.
Error: Failed to open PDF file:
/tmp/d20100520-15606-f6vqow/merged.pdf
Errors encountered. No output created.
Done. Input errors, so no output created.
Cleaning up temporary files

[texla]

What pdfocr is for

Installing pdfocr

The easiest way to install pdfocr is to add my PPA and use apt-get. If you would instead prefer to install it manually, see here for instructions

sudo add-apt-repository ppa:gezakovacs/pdfocr
sudo apt-get update
sudo apt-get install pdfocr

Using pdfocr to add a text layer to your scanned PDF file

Open a terminal, go to the directory that has the PDF file you want to convert, and enter (substituting input.pdf with the input PDF file, and output.pdf with the output PDF file)

pdfocr -i input.pdf -o output.pdf

Now wait as OCR is performed on the PDF file page-by-page, and the output file is generated. This should take a few seconds per page, depending on the resolution of your PDF file (high-res PDF files get better accuracy, but will take longer). Once done, you should now have a searchable PDF at output.pdf.

Worked great for me - thanks! I had to go in and fix about 20% of the text but this helped a lot - will be using this more...