I have installed on my RedHat machine:
(py36_maw) [rvp#lib-archcoll box]$ tesseract -v
tesseract 4.1.0
leptonica-1.78.0
libjpeg 6b (libjpeg-turbo 1.2.90) : libpng 1.5.13 : libtiff 4.0.3 : zlib 1.2.7 : libopenjp2 2.3.1
Found SSE
I try to run, per what docs I can find, to produce pdf output:
(py36_maw) [rvp#lib-archcoll box]$ time tesseract test.jp2 out -l eng PDF
read_params_file: Can't open PDF
Tesseract Open Source OCR Engine v4.1.0 with Leptonica
Warning: Invalid resolution 0 dpi. Using 70 instead.
Estimating resolution as 275
That takes 10 seconds and produces file out.txt with fine OCR to text conversion evident.
However, it tries to read a file called PDF, but I cannot figure how to get PDF output.
I have read various docs, the most promising seeming to be advising to edit the config file, but the only docs I can guess are relevant, by googling 'tesseract 4.1 config', list many 'config' variable names, for older versions of tesseract, but none of which seems to indicate I can specify producing pdf output, much less specifically for tesseract 4.1.
How can I invoke tesseract 4.1 (using libopenjp2 2.3.1) via CLI to produce pdf output from my jp2 input file? Bonus question: how can I get it to produce both txt and pdf output in one run?
Robert
After more surfing and digging, assuming the reader also has done some and knows what TESSDATA_PREFIX is used for by tesseract, here are the steps that worked for me:
Download the pdf.ttf file from: https://github.com/tesseract-ocr/tesseract/blob/master/tessdata/pdf.ttf
Copy pdf.ttf to your directory $TESSDATA_PREFIX and make sure that variable is exported to your shell.
TIP: Use command: tesseract --print-parameters # to discover defined variable names you can use in your own config file
Go to your dir with the test.jp2 file and create file config with these lines.
tessedit_create_pdf 1 Write .pdf output file
tessedit_create txt 1 Write .txt output file
(Note: or you may be able to put the config file in the TESSDATA_PREFIX directory as well and let it always be the default. Not tested.)
Run in that dir:
$ tesseract test.jp2 outputbase -l eng config
Verify your success: it runs and produces files outputbase.txt and outputbase.pdf. The txt file looks good and the searchable pdf looks and works OK in a pdf viewer, that is, you can search and find text strings.
Hope this helps someone else!
Related
I am using Spatie/pdfToImage that builds on ghost script and imagemagick to on my server:
Take a multiple page pdf from an email using mailgun routing.
Save the pdf in folder /docs_pdf like file.pdf
Use a foreach to loop through each page and save each page as a png to /docs like file_#.png
locally where I use laravel -> valet everything works fine.
On my server using digital ocean through laravel forge the language in a multipaged pdf that is in swedish transforms from normal swedish to a bunch of random letters and signs.
The left is correct (yes, its true. Its Swedish) and the right is wrong:
Someone suggested to me that this is probably a matter of the font missing on the server. The fonts used in the pdf:
<</StemV 68/FontName/PSQHMO+FoundrySans-Normal/FontFile2 216 0 R/FontStretch/Normal/FontWeight 400/Flags 32/Descent -240/FontBBox[-40 -240 960 916]/Ascent 916/FontFamily(FoundrySans-Normal)/CapHeight 667/XHeight 465/Type/FontDescriptor/ItalicAngle 0>>
<</StemV 100/FontName/MLHPWU+FoundrySans-Medium/FontFile2 217 0 R/FontStretch/Normal/FontWeight 400/Flags 32/Descent -241/FontBBox[-42 -241 1008 916]/Ascent 916/FontFamily(FoundrySans-Medium)/CapHeight 667/XHeight 470/Type/FontDescriptor/ItalicAngle 0>>
<</StemV 68/FontName/SUEECI+FoundrySans-Normal/FontFile2 218 0 R/FontStretch/Normal/FontWeight 400/Flags 4/Descent -240/FontBBox[-40 -240 960 916]/Ascent 916/FontFamily(FoundrySans-Normal)/CapHeight 667/XHeight 465/Type/FontDescriptor/ItalicAngle 0>>
<</StemV 48/FontName/KIDDUY+FoundrySans-Light/FontFile2 9 0 R/FontStretch/Normal/FontWeight 400/Flags 32/Descent -248/FontBBox[-28 -248 978 924]/Ascent 924/FontFamily(FoundrySans-Light)/CapHeight 667/XHeight 458/Type/FontDescriptor/ItalicAngle 0>>
Here is configuration of fonts in imagemagick and ghostscript:
https://www.imagemagick.org/script/resources.php
how can this be solved?
Update:
I have now made a clean install on a new server.
Installed Imagick and spatie/pdfToImage
As suggested by KenS I ran
gs -sDEVICE=png16m -o out%d.png
terminal output
forge#Server:~/app/storage/app/public/files$ gs -sDEVICE=png16m -o test_out%d.png file.pdf
GPL Ghostscript 9.22 (2017-10-04)
Copyright (C) 2017 Artifex Software, Inc. All rights reserved.
This software comes with NO WARRANTY: see the file PUBLIC for details.
Processing pages 1 through 2.
Page 1
Page 2
the document rendered the same = wrong.
I am at a complete loss.. Don't know what next step might be..
Update2:
I also run the convert imagemagick commando and the img rendered the same way also.
So even if I do it with ghostscript solo, imagemagick or spatie/pdfToImage it gives me the same output
Well, the current version of Ghostscript (9.25) renders this acceptably for me; that is the text appears to be correct. All the fonts are embedded, so there shouldn't be any problems.
And this means that even if you did replace the default font substitution, it wouldn't help, because Ghostscript shouldn't be using the default font, it will be using the fonts embedded in the PDF file.
Without knowing what version of Ghostscript you are using (I see from a later comment that its 9.25), or the command line that is used to start it, I can't really do a like-for-like comparison. Its hard for me to see how you could be getting such a different result though. That looks like Ghostscript has failed to find the embedded fonts.
Its possible that whatever package you are using has done something 'unfortunate'. The various package maintainers on Linux add their own patches, and sometimes modify the way that Ghostscript is built. Possibly that has broken something.
If you are able to build Ghostscript yourself you could try cloning our Git repository and doing that. You could also try downloading the Linux binaries off our website. They won't work with every Linux distribution (different ABI) but you can try, you might be lucky.
You could also try running Ghostscript directly on the PDF file. Something like:
gs -sDEVICE=png16m -o out%d.png
should produce 2 PNG files, out1.png and out2.png. It will also produce a bunch of stuff on the terminal. That back channel output is valuable information for me so if you can reproduce the problem, I'd like to see that too.
One last thought; its possible to have more than one version of Ghostscript installed simultaneously, its possible that your current setup is using an old version of Ghostscript.
I can't help you with ImageMagick or Spatie, but if you can debug those to the point where you can reproduce the problem with a plain Ghostscript command line then I can look further at it.
Finally got it to work. I want to first give kudos to KenS that really helped me, and without him it would not have worked.
This is what I did:
1 - I removed Ghostscript:
sudo apt-get purge --auto-remove ghostscript
then
wget https://github.com/ArtifexSoftware/ghostpdl-downloads/releases/download/gs925/ghostscript-9.25.tar.gz
tar xvf ghostscript-9.25.tar.gz
Enter the unpacked folder and do
./configure
make
make install
then
sudo ln -s /usr/local/bin/gs /usr/bin/gs
On top of the above I did:
sudo add-apt-repository ppa:glasen/freetype2
and then:
sudo apt update && sudo apt install freetype2-demos
i'm running Tesseract 4.0.0 and i tried the following command in order to create a searchable pdf but it doesn't seem to work :
tesseract input output pdf
It gives an error :
can't open file "\Program Files\...//pdf.ttf"!
error during processing
The pdf file gets created but it cannot be open.
I tried it on different image formats : jpg, tif, png with no success.
It does work, not sure which os you are using, but I realised that to make it work on Linux a full install was necessary
sudo apt install tesseract-ocr
sudo apt install tesseract-ocr-all
then, for a German document for example, originally a multipage tif:
tesseract multipage-tiff.tif out pdf -l deu
the manual is useful - https://github.com/tesseract-ocr/tesseract/wiki
I have much trouble to have a code to convert pdf file to png on python 3.6, windows 10.
I know what you are going to say : google it !
But barely everything I've found was on python 2.7. And some packages haven't been updated.
What I've seen so far it's that the best way to do it is using Wand, right ? (I have installed ImageMagick before )
from wand.image import Image
# Converting first page into JPG
with Image(filename='0.pdf') as img:
img.save(filename="/temp.jpg")
# Resizing this image
Here was my second error :
wand.exceptions.DelegateError: PDFDelegateFailed
`The system cannot find the file specified.' # error/pdf.c/ReadPDFImage/809
So i read i need ghostscript. I installed it. But the package is for python 2.7 and it doesn't work. I found python3-ghostscript 0.5.0. https://pypi.python.org/pypi/python3-ghostscript/0.5.0
New error :
RuntimeError: Can not find Ghostscript DLL in registry
So here I needed to install Ghostscript 9 :
https://www.ghostscript.com/download/gsdnld.html
First of all it's not a GPL license ... That's not even a package but a program. I don't know how I can use it in my futures python codes...
and there is still an error :
RuntimeError: Can not find Ghostscript DLL in registry
and i can't find anything for it.
Ghostscript is licensed under the AGPL, the licence can be found in /Program Files (x86)/gs/gs9.21/doc if you want sources then they are available from the Ghostscript Git repository. Note I'm assuming you are running on Windows since you refer to the Registry.
If you install the prebuilt binary then it will create an entry in the Windows Registry, I assume that's what your Python code is looking for but I can't be sure. You should make sure you install the correct word size (32 or 64) version required by Python, if it cares.
You can, of course, simply run Ghostscript to render a PDF file and produce PNG output.
gswin32c -sDEVICE=png16m -sOutputFile=out%d.png input.pdf
This will create one file per page of the input PDF file, use gswin64c for the 64-bit version...
You can alter the resolution of the output with the -r switch, eg -r300
I presume you can simply fork a process from Python. Otherwise you'll have to get someone to tell you what the Python script is looking for in the Registry. Perhaps its looking for a specific version of Ghostscript, or the 32-bit version or something.
When I run my app, that converts pdf to png, from django server, the conversion works fine. But when I run this from an apache server, I am getting this error: GhoscriptError: Fatal. Reading from the sterr of ghostscript, it says
Initialization file gs_init.ps does
not begin with an integer.
It seems an initialization error for me, but I have no idea how to fix this.
Using Ubuntu by the way. gs folder is in the path, so Im not sure if that is causing the problem.
Here's my code that generates the images
def PDF_to_png(input,output):
args = [
"-dSAFER",
"-dBATCH", "-dNOPAUSE", "-sDEVICE=png16m",
"-r300",
"-sOutputFile=" + os.path.join(output,input.file_name_without_extension)+"_%d.png",
input
]
ghostscript.Ghostscript(*args)
The error is telling you that the file gs_init.ps which is normally found in gs/Resource/Init/ is not valid. From the header of the file:
------------------------------------------------------------------------
% Interpreter library version number
% NOTE: the interpreter code requires that the first non-comment token
% in this file be an integer, and that it match the compiled-in version!
902
------------------------------------------------------------------------
You can build GS with the resources built-in or on disk, I don't know which build you get with Ubuntu but it sounds like either there is a gs_init.ps in the GS path which has been damaged. This probably means you are using a version with the resources on disk.
You should first try just starting up Ghostscript. If that works then it's something to do with the environment which is different when you run the failing instance. Look for environment variables which begin GS_ (especially *GS_LIB*). You should also try actually defining where GS should look on the command line by including something like :
-I/usr/src/gs/Resource
This I ncludes the specified directory as a search path for Ghostscript (NB GS does not use the PATH environment variable). GS will search here for initialisation files first before proceeding on its fall back mechanism.
If I have 10,000 PDFs, some of which have been OCRed, some of which have 1 page that has been OCRed but the rest of the pages have not, how can I go through all the PDFs and only OCR the pages that haven't already been done?
This is exactly what I was looking for, I have thousands of scanned PDF files, where some were already OCR'ed and some are not.
So, I combined information I found on fora and Stack Overflow, and made my own solution that does EXACTLY that, which I have summarized for you here:
scan through all subdirectories recursively for PDF files;
check if the PDF was already OCR'ed, and if not, process the PDF with OCR with high quality, in the language(s) you can specify;
save the OCR PDF in-place, as PDF/A, and overwriting the old (not-OCR'ed) one.
I am on Windows 10, and could not find the definitive answer. I tried doing this with Acrobat Pro, but that gave me many errors, and Acrobat's batch processing stops on every error or password-protected file. I also tried many other batch-OCR tools on Windows, but none worked well.
I spent countless hours manually checking which files already had a text-layer "under" the image.
UNTIL! Microsoft announced that it was now very easy to run Linux under Windows, on the same machine, on the same filesystem.
There are many more tools and utilities available on Linux than Windows, so I thought I would give that a try.
So, here it is, step by step:
Enable the Windows subsystem for Linux in the Windows Control Panel; there are many guides. Google it. It's a couple of minutes.
Install Linux from the Windows Store. Open the Windows Store, search for Ubuntu, and install. Takes around 5 minutes.
Now you have the "Ubuntu app". Run it. It shows you the linux bash, and with file access to your Windows files through /mnt/c. It's magic!
You need some Linux "apps", namely pdffonts and ocrmypdf; which you can install by using the command sudo apt install pdffonts and sudo apt install ocrmypdf. We will use these apps to check if there is an embedded font in a PDF, and if not, OCR the PDF. (see note below).
Install the very small bash script (below) to your home directory ~.
Go to (cd) the directory where all your PDF's are saved. For example: /mnt/c/Users/name/OneDrive/Documents.
Run the command: find . -type f -name "*.pdf" -exec /your/homedir/pdf-ocr.sh '{}' \;
Done!
Running this might, of course, take a long time, depending on how many PDF's you have, and how many of those are not OCR'ed yet.
Here is the sh-script. You should save it somewhere in your home folder so that it is easy to call from anywhere. Like so:
type cd ~. This will bring you to your home folder.
type pico pdf-ocr.sh. This will bring up an editor. Paste the below script code. Then press Ctrl+X, and press Y. Your file is now saved.
type sudo chmod +x pdf-ocr.sh. This will give the script permission to be run.
MYFONTS=$(pdffonts -l 5 "$1" | tail -n +3 | cut -d' ' -f1 | sort | uniq)
if [ "$MYFONTS" = '' ] || [ "$MYFONTS" = '[none]' ]; then
echo "Not yet OCR'ed: $1 -------- Processing...."
echo " "
ocrmypdf -l eng+deu+nld -s "$1" "$1"
echo " "
else
echo "Already OCR'ed: $1"
echo " "
fi
What does this do?
Well, the find command looks up all PDF files in the current directory including subdirectories. It then "sends" these files to the script, in which pdffonts checks if there are embedded fonts. If so, skip the file and try the next one. If no embedded fonts are found, use ocrmypdf to do the OCR-ing.
I found the quality of OCR from ocrmypdf VERY good, even better than Acrobat's. You can of course tweak the settings. I can imagine for example that you might want to use other languages for OCR than eng+deu+nld. You can look up all options here: https://ocrmypdf.readthedocs.io/en/latest/
Note: I am making the assumption here that if a PDF file has no embedded fonts (so it's basically an image (scan) in a PDF-file), that it has not OCR'ed. I know that this might not always be accurate and/or true, but for me that is enough to determine which files to put through OCR. So that it is not neccesary to re-do hundreds or thousands of PDF files....
I know that it is a bit more hassle to install Linux under Windows, but as it is very easy to do if you have basic Linux skills. For me it was worth the effort because I now have made "one click" batch processor that works. I could not find a solution for that with Windows-tools.
I hope someone finds this and finds this useful. If anyone has improvements, please post them here.
Thanks.
Jos Jonkeren
Why don't you re-OCR everything? The amount of time you spend agonizing over repeated work probably exceeds the time taken for the work itself.
If by OCRed you mean that they contain the text in machine-readable form, you could use a library like Apache PDFBox to try to extract the text from the second page of the document. If it throws an error or returns garbage, it's most likely not OCRed.
Unburying this thread.
You can know which PDF files have already been OCRed by testing them with pdffonts. If there are embedded fonts, it's very probable that the PDF is already OCRed.
As for the batch processing, I wrote a little script that can batch OCR to pdf/word/excel/csv output format.
You may find it at https://github.com/deajan/pmOCR
pmOCR (poor man's OCR is a wrapper for Abbyy OCR CLI for linux or Tesseract 3 open source solution).