Add support for using pre-existing text from PDFs

.travis.yml

@@ -1,5 +1,9 @@
 language: python
 
+before_install:
+- sudo apt-get update -qq
+- sudo apt-get install -qq libpoppler-cpp-dev
+
 sudo: false
 
 matrix:

Dockerfile

@@ -18,9 +18,9 @@ ENV PAPERLESS_EXPORT_DIR=/export \
 # Install dependencies
 RUN apk --no-cache --update add \
         python3 gnupg libmagic bash \
-        sudo tesseract-ocr imagemagick ghostscript unpaper && \
+        sudo poppler tesseract-ocr imagemagick ghostscript unpaper && \
     apk --no-cache add --virtual .build-dependencies \
-        python3-dev gcc musl-dev zlib-dev jpeg-dev && \
+        python3-dev poppler-dev gcc g++ musl-dev zlib-dev jpeg-dev && \
 # Install python dependencies
     python3 -m ensurepip && \
     rm -r /usr/lib/python*/ensurepip && \

docs/changelog.rst

@@ -3,7 +3,15 @@ Changelog
 
 * 1.2.0
   * New Docker image, now based on Alpine, thanks to the efforts of `addadi`_
-  and `Pit`_.
+    and `Pit`_.
+  * `BastianPoe`_ has added the long-awaited feature to automatically skip the
+    OCR step when the PDF already contains text. This can be overridden by
+    setting ``PAPERLESS_OCR_ALWAYS=YES`` either in your ``paperless.conf`` or
+    in the environment.  Note that this also means that Paperless now requires
+    ``libpoppler-cpp-dev`` to be installed. **You'll need to run
+    ``pip install -r requirements.txt`` after the usual ``git pull`` to
+    properly update**.
+
 * 1.1.0
   * Fix for `#283`_, a redirect bug which broke interactions with
     paperless-desktop.  Thanks to `chris-aeviator`_ for reporting it.
@@ -272,6 +280,7 @@ Changelog
 .. _chris-aeviator: https://github.com/chris-aeviator
 .. _Dan Panzarella: https://github.com/pzl
 .. _addadi: https://github.com/addadi
+.. _BastianPoe: https://github.com/BastianPoe
 
 .. _#20: https://github.com/danielquinn/paperless/issues/20
 .. _#44: https://github.com/danielquinn/paperless/issues/44

docs/requirements.rst

@@ -11,24 +11,27 @@ should work) that has the following software installed:
 * `Tesseract`_, plus its language files matching your document base.
 * `Imagemagick`_ version 6.7.5 or higher
 * `unpaper`_
+* `libpoppler-cpp-dev`_ PDF rendering library
 
 .. _Python3: https://python.org/
 .. _GNU Privacy Guard: https://gnupg.org
 .. _Tesseract: https://github.com/tesseract-ocr
 .. _Imagemagick: http://imagemagick.org/
 .. _unpaper: https://www.flameeyes.eu/projects/unpaper
+.. _libpoppler-cpp-dev: https://poppler.freedesktop.org/
 
 Notably, you should confirm how you access your Python3 installation.  Many
-Linux distributions will install Python3 in parallel to Python2, using the names
+Linux distributions will install Python3 in parallel to Python2, using the
-``python3`` and ``python`` respectively.  The same goes for ``pip3`` and
+names ``python3`` and ``python`` respectively.  The same goes for ``pip3`` and
-``pip``.  Running Paperless with Python2 will likely break things, so make sure that 
+``pip``.  Running Paperless with Python2 will likely break things, so make sure
-you're using the right version.
+that you're using the right version.
 
 For the purposes of simplicity, ``python`` and ``pip`` is used everywhere to
 refer to their Python3 versions.
 
 In addition to the above, there are a number of Python requirements, all of
-which are listed in a file called ``requirements.txt`` in the project root directory.
+which are listed in a file called ``requirements.txt`` in the project root
+directory.
 
 If you're not working on a virtual environment (like Vagrant or Docker), you
 should probably be using a virtualenv, but that's your call.  The reasons why
@@ -39,12 +42,13 @@ probably figure that out before continuing.
 
 .. _requirements-apple:
 
-Apple-tastic Complications
+Problems with Imagemagick & PDFs
---------------------------
+--------------------------------
 
-Some users have `run into problems`_ with installing ImageMagick on Apple
+Some users have `run into problems`_ with getting ImageMagick to do its thing
-systems using HomeBrew.  The solution appears to be to install ghostscript as
+with PDFs.  Often this is the case with Apple systems using HomeBrew, but other
-well as ImageMagick:
+Linuxes have been a problem as well.  The solution appears to be to install
+ghostscript as well as ImageMagick:
 
 .. _run into problems: https://github.com/danielquinn/paperless/issues/25
 

requirements.txt

@@ -14,6 +14,7 @@ python-dotenv>=0.6.2
 python-gnupg>=0.3.9
 pytz>=2016.10
 gunicorn==19.7.1
+pdftotext>=2.0.1
 
 # For the tests
 factory-boy

src/paperless/settings.py

@@ -210,6 +210,9 @@ OCR_LANGUAGE = os.getenv("PAPERLESS_OCR_LANGUAGE", "eng")
 # The amount of threads to use for OCR
 OCR_THREADS = os.getenv("PAPERLESS_OCR_THREADS")
 
+# OCR all documents?
+OCR_ALWAYS = bool(os.getenv("PAPERLESS_OCR_ALWAYS", "NO").lower() in ("yes", "y", "1", "t", "true"))
+
 # If this is true, any failed attempts to OCR a PDF will result in the PDF
 # being indexed anyway, with whatever we could get.  If it's False, the file
 # will simply be left in the CONSUMPTION_DIR.

src/paperless_tesseract/parsers.py

@@ -3,6 +3,7 @@ import os
 import re
 import subprocess
 from multiprocessing.pool import Pool
+import pdftotext
 
 import langdetect
 import pyocr
@@ -31,6 +32,7 @@ class RasterisedDocumentParser(DocumentParser):
     THREADS = int(settings.OCR_THREADS) if settings.OCR_THREADS else None
     UNPAPER = settings.UNPAPER_BINARY
     DEFAULT_OCR_LANGUAGE = settings.OCR_LANGUAGE
+    OCR_ALWAYS = settings.OCR_ALWAYS
 
     def get_thumbnail(self):
         """
@@ -46,7 +48,21 @@ class RasterisedDocumentParser(DocumentParser):
 
         return os.path.join(self.tempdir, "convert-0000.png")
 
+    def _is_ocred(self):
+        # Extract text from PDF using pdftotext
+        text = get_text_from_pdf(self.document_path)
+
+        # We assume, that a PDF with at least 50 characters contains text
+        # (so no OCR required)
+        if len(text) > 50:
+            return True
+
+        return False
+
     def get_text(self):
+        if not self.OCR_ALWAYS and self._is_ocred():
+            self.log("info", "Skipping OCR, using Text from PDF")
+            return get_text_from_pdf(self.document_path)
 
         images = self._get_greyscale()
 
@@ -212,3 +228,13 @@ def image_to_string(args):
             except (TesseractError, OtherTesseractError):
                 pass
         return ocr.image_to_string(f, lang=lang)
+
+
+def get_text_from_pdf(pdf_file):
+    with open(pdf_file, "rb") as f:
+        try:
+            pdf = pdftotext.PDF(f)
+        except pdftotext.Error:
+            return False
+
+    return "\n".join(pdf)