Converts a scanned PDF into an OCR'ed pdf using Tesseract-OCR and Ghostscript
PyPDFOCR - Tesseract-OCR based PDF filing
This program will help manage your scanned PDFs by doing the following:
- Take a scanned PDF file and run OCR on it (using the Tesseract OCR software from Google), generating a searchable PDF
- Optionally, watch a folder for incoming scanned PDFs and automatically run OCR on them
- Optionally, file the scanned PDFs into directories based on simple keyword matching that you specify
- Evernote auto-upload and filing based on keyword search
- Email status when it files your PDF
pypdfocr filename.pdf --> filename_ocr.pdf will be generated
If you have a language pack installed, then you can specify it with the -l option:
pypdfocr -l spa filename.pdf
pypdfocr -w watch_directory --> Every time a pdf file is added to `watch_directory` it will be OCR'ed
To automatically move the OCR’ed pdf to a directory based on a keyword, use the -f option and specify a configuration file (described below):
pypdfocr filename.pdf -f -c config.yaml
You can also do this in folder monitoring mode:
pypdfocr -w watch_directory -f -c config.yaml
Filing based on filename match:
If no keywords match the contents of the filename, you can optionally allow it to fallback to trying to find keyword matches with the PDF filename using the -n option. For example, you may have receipts always named as receipt_2013_12_2.pdf by your scanner, and you want to move this to a folder called ‘receipts’. Assuming you have a keyword receipt matching to folder receipts in your configuration file as described below, you can run the following and have this filed even if the content of the pdf does not contain the text ‘receipt’:
pypdfocr filename.pdf -f -c config.yaml -n
Configuration file for automatic PDF filing
The config.yaml file above is a simple folder to keyword matching text file. It determines where your OCR’ed PDFs (and optionally, the original scanned PDF) are placed after processing. An example is given below:
target_folder: "docs/filed" default_folder: "docs/filed/manual_sort" original_move_folder: "docs/originals" folders: finances: - american express - chase card - internal revenue service travel: - boarding pass - airlines - expedia - orbitz receipts: - receipt
The target_folder is the root of your filing cabinet. Any PDF moving will happen in sub-directories under this directory.
The folders section defines your filing directories and the keywords associated with them. In this example, we have three filing directories (finances, travl, receipts), and some associated keywords for each filing directory. For example, if your OCR’ed PDF contains the phrase “american express” (in any upper/lower case), it will be filed into docs/filed/finances
The default_folder is where the OCR’ed PDF is moved to if there is no keyword match.
The original_move_folder is optional (you can comment it out with # in front of that line), but if specified, the original scanned PDF is moved into this directory after OCR is done. Otherwise, if this field is not present or commented out, your original PDF will stay where it was found.
If there is any naming conflict during filing, the program will add an underscore followed by a number to each filename, in order to avoid overwriting files that may already be present.
Evernote authentication token
To enable Evernote support, you will need to get a developer token for your Evernote account.. You should note that this script will never delete or modify existing notes in your account, and limits itself to creating new Notebooks and Notes. Once you get that token, you copy and paste it into your configuration file as shown below
Evernote filing usage
To automatically upload the OCR’ed pdf to a folder based on a keyword, use the -e option instead of the -f auto filing option.
pypdfocr filename.pdf -e -c config.yaml
Similarly, you can also do this in folder monitoring mode:
pypdfocr -w watch_directory -e -c config.yaml
Evernote filing configuration file
The config file shown above only needs to change slightly. The folders section is completely unchanged, but note that target_folder is the name of your “Notebook stack” in Evernote, and the default_folder should just be the default Evernote upload notebook name.
target_folder: "evernote_stack" default_folder: "default" original_move_folder: "docs/originals" evernote_developer_token: "YOUR_TOKEN" folders: finances: - american express - chase card - internal revenue service travel: - boarding pass - airlines - expedia - orbitz receipts: - receipt
You can have PyPDFOCR email you everytime it converts a file and files it. You need to first specify the following lines in the configuration file and then use the -m option when invoking pypdfocr:
mail_smtp_server: "smtp.gmail.com:587" mail_smtp_login: "firstname.lastname@example.org" mail_smtp_password: "PASSWORD" mail_from_addr: "email@example.com" mail_to_list: - "firstname.lastname@example.org" - "email@example.com"
You can specify Tesseract and Ghostscript executable locations manually, as well as the number of concurrent processes allowed during preprocessing and tesseract. Use the following in your configuration file:
tesseract: binary: "/usr/bin/tesseract" threads: 8 ghostscript: binary: "/usr/local/bin/gs" preprocess: threads: 8
Handling disk time-outs
If you need to increase the time interval (default 3 seconds) between new document scans when pypdfocr is watching a directory, you can specify the following option in the configuration file:
watch: scan_interval: 6
PyPDFOCR is available in PyPI, so you can just run:
pip install pypdfocr
Please note that some of the 3rd-party libraries required by PyPDFOCR wiill require some build tools, especially on a default Ubuntu system. If you run into any issues using pip install, you may want to install the following packages on Ubuntu and try again:
For those on Windows, because it’s such a pain to get all the PIL and PDF dependencies installed, I’ve gone ahead and made an executable called pypdfocr.exe
You still need to install Tesseract, GhostScript, etc. as detailed below in the external dependencies list.
Clone the source directly from github (you need to have git installed):
git clone https://github.com/virantha/pypdfocr.git
Then, install the following third-party python libraries:
- Pillow (Python Imaging Library) https://pillow.readthedocs.org/en/3.1.x/
- ReportLab (PDF generation library) http://www.reportlab.com/opensource/
- Watchdog (Cross-platform fhlesystem events monitoring) https://pypi.python.org/pypi/watchdog
- PyPDF2 (Pure python pdf library)
These can all be installed via pip:
pip install Pillow pip install reportlab pip install watchdog pip install pypdf2
You will also need to install the external dependencies listed below.
PyPDFOCR relies on the following (free) programs being installed and in the path:
- Tesseract OCR software https://code.google.com/p/tesseract-ocr/
- GhostScript http://www.ghostscript.com/
- ImageMagick http://www.imagemagick.org/
- Poppler http://poppler.freedesktop.org/ (Windows)
Poppler is only required if you want pypdfocr to figure out the original PDF resolution automatically; just make sure you have pdfimages in your path. Note that the xpdf provided pdfimages does not work for this, because it does not support the -list option to list the table of images in a PDF file.
On Mac OS X, you can install these using homebrew:
brew install tesseract brew install ghostscript brew install poppler brew install imagemagick
On Windows, please use the installers provided on their download pages.
** Important ** Tesseract version 3.02.02 or newer required (apparently 3.02.01-6 and possibly others do not work due to a hocr output format change that I’m not planning to address). On Ubuntu, you may need to compile and install it manually by following these instructions
Also note that if you want Tesseract to recognize rotated documents (upside down, or rotated 90 degrees) then you need to find your tessdata directory and do the following:
cd /usr/local/share/tessdata cp eng.traineddata osd.traineddata
osd stands for Orientation and Script Detection, so you need to copy the .traineddata for whatever language you want to scan in as osd.traineddata. If you don’t do this step, then any landscape document will produce garbage
While test coverage is at 84% right now, Sphinx docs generation is at an early stage. The software is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|v0.9.1||10/11/16||Fixes (#43, #41)|
|v0.9.0||2/29/16||Fixed rotated page text, Mac OS X invisible fonts, and pdf merge slowdown|
|v0.8.5||2/21/16||Better ctrl-c and cleanup behavior|
|v0.8.3||2/18/16||Bug fix for multiprocessing on windows, ctrl-c interrupt, and integer keywords|
|v0.8.2||12/8/14||Fixed imagemagick invocation on windows. Parallelized preprocessing and tesseract execution|
|v0.8.1||12/5/14||Added –skip-preprocess option, scan_interval option, and fixed too many open files bug during page overlay|
|v0.8.0||10/27/14||Added preprocessing to clean up prior to tesseract, bug fixes on file names with spaces/dots|
|v0.7.6||9/10/14||Fixed issue 17 rotation bug|
|v0.7.5||8/18/14||Update for Tesseract 3.03 .hocr filename change|
|v0.7.4||3/28/14||Bug fix on pdf assembly|
|v0.7.3||3/27/14||Modified internals to use single image per page (instead of multipage tiff). Also enabled orientation detection|
|v0.7.2||3/26/14||Switched from Pil to Pillow. Now uses original images from PDF in output pdf (no dpi/color/quality changes!)|
|v0.7.1||3/25/14||OCR Language is now an option|
|v0.7.0||3/25/14||Now honors original pdf resolution|
|v0.6.1||2/16/14||Bug fix for pdfs with only numbers in the filename|
|v0.6.0||1/16/14||Added filing based on filename match as fallback, added tesseract version check|
|v0.5.4||1/12/14||Fixed bug with reordering of text pages on certain platforms(glob)|
|v0.5.3||12/12/13||Fix to evernote server specification|
|v0.5.2||12/08/13||Fix to lowercase keywords|
|v0.5.1||11/02/13||Fixed a bunch of windows critical path handling issues|
|v0.5.0||10/30/13||Email status added, 90% test coverage|
|v0.4.1||10/28/13||Made HOCR parsing more robust|
|v0.4.0||10/28/13||Added early Evernote upload support|
|v0.3.1||10/24/13||Path fix on windows|
|v0.3.0||10/23/13||Added filing of converted pdfs using a configuration file to specify target directories based on keyword matches in the pdf text|
|v0.2.2||10/22/13||Added a console script to put the pypdfocr script into your bin|
|v0.2.1||10/22/13||Fix to initial packaging problem.|
- #43 version check for tesseract
- On windows, search for pdfimages and imagemagick instead of relying on path
- Split up into flow steps
- Run more robustness tests for watching networked shares
- Add more docstrings
- Add more option specifiers to tesseract and ghostscript
Release history Release notifications | RSS feed
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.