v7.2.0 release notes update

Soft masks are only allowed to be of colorspace DeviceGray so we
shouldn't use pngquant on them. For now, avoid this exceptional
case by excluded soft masks from optimization.

.gitignore

@@ -1,47 +1,40 @@
 # Development environment
+.bash_history
+.pylintrc
+.pytest_cache/
+.ruffus_history.sqlite
+.venv/
 *.pyc
 *.sublime-*
-venv*/
-.venv/
-pyvenv.cfg
-tasks.py
-.bash_history
-.ruffus_history.sqlite
-.idea/
-.pytest_cache/
-.pylintrc
 
 # Package building
-*.egg-info/
-.cache/
 .eggs/
+*.egg-info/
 build/
 dist/
 wheelhouse/
 
 # Automatically generated files
-ocrmypdf/lib/_*.py
-ocrmypdf/version.py
 docs/_build/
 docs/_static/
 docs/_templates/
 docs/Makefile
+ocrmypdf/lib/_*.py
 
 # Code coverage
 .coverage
 htmlcov/
 
 # Testing
-log/
+.ipynb_checkpoints/
+.vscode/
+*.ipynb
+*.profile
 /*.pdf
 /*.qdf
-*.ipynb
-.ipynb_checkpoints/
+/scratch.py
+IDEAS
+log/
 tests/output/
 tests/resources/private/
 tmp/
-pdfbox-app*.jar
-.vscode/
-IDEAS
-_Dockerfile.local
-/scratch.py

.travis.yml

@@ -66,7 +66,7 @@ matrix:
           - tesseract-ocr-fra
           - unpaper
     - os: osx
-      osx_image: xcode8
+      osx_image: xcode9.2
       language: generic
 
 before_cache:
@@ -83,7 +83,7 @@ before_install: |
       sudo dpkg -i packages/unpaper_6.1-1.deb
     fi
   elif [[ "$TRAVIS_OS_NAME" == "osx" ]]; then
-    brew update && brew bundle --file=.travis/Brewfile
+    brew update --quiet && brew bundle --file=.travis/Brewfile --quiet
     pip3 install --upgrade pip
     pip3 install wheel
   fi

.travis/Brewfile

@@ -4,7 +4,6 @@ brew 'exempi'
 brew 'ghostscript'
 brew 'jbig2enc'
 brew 'leptonica'
-brew 'libtiff'
 brew 'openjpeg'
 brew 'pngquant'
 brew 'qpdf'

README.md

@@ -0,0 +1,132 @@
+OCRmyPDF
+========
+
+![image](https://travis-ci.org/jbarlow83/OCRmyPDF.svg?branch=master%0A%20:target:%20https://travis-ci.org/jbarlow83/OCRmyPDF)
+
+![image](https://img.shields.io/pypi/v/ocrmypdf.svg%0A%20:target:%20https://pypi.org/project/ocrmypdf/)
+
+![image](https://img.shields.io/homebrew/v/ocrmypdf.svg%0A%20:alt:%20homebrew%0A%20:target:%20http://brewformulas.org/Ocrmypdf)
+
+OCRmyPDF adds an OCR text layer to scanned PDF files, allowing them to be searched or copy-pasted.
+
+```bash
+ocrmypdf                      # it's a scriptable command line program
+   -l eng+fra                 # it supports multiple languages
+   --rotate-pages             # it can fix pages that are misrotated
+   --deskew                   # it can deskew crooked PDFs!
+   --title "My PDF"           # it can change output metadata
+   --jobs 4                   # it uses multiple cores by default
+   --output-type pdfa         # it produces PDF/A by default
+   input_scanned.pdf          # takes PDF input (or images)
+   output_searchable.pdf      # produces validated PDF output
+```
+
+Main features
+-------------
+
+-   Generates a searchable [PDF/A](https://en.wikipedia.org/?title=PDF/A) file from a regular PDF
+-   Places OCR text accurately below the image to ease copy / paste
+-   Keeps the exact resolution of the original embedded images
+-   When possible, inserts OCR information as a "lossless" operation without disrupting any other content
+-   Optimizes PDF images, often producing files smaller than the input file
+-   If requested deskews and/or cleans the image before performing OCR
+-   Validates input and output files
+-   Distributes work across all available CPU cores
+-   Uses [Tesseract OCR](https://github.com/tesseract-ocr/tesseract) engine
+-   Supports more than [100 languages](https://github.com/tesseract-ocr/tessdata) recognized by Tesseract
+-   Battle-tested on thousands of PDFs, a test suite and continuous integration
+
+For details: please consult the [documentation](https://ocrmypdf.readthedocs.io/en/latest/).
+
+Motivation
+----------
+
+I searched the web for a free command line tool to OCR PDF files on Linux/UNIX: I found many, but none of them were really satisfying.
+
+-   Either they produced PDF files with misplaced text under the image (making copy/paste impossible)
+-   Or they did not handle accents and multilingual characters
+-   Or they changed the resolution of the embedded images
+-   Or they generated ridiculously large PDF files
+-   Or they crashed when trying to OCR
+-   Or they did not produce valid PDF files
+-   On top of that none of them produced PDF/A files (format dedicated for long time storage)
+
+...so I decided to develop my own tool.
+
+Installation
+------------
+
+Linux, UNIX, and macOS are supported. Windows is not directly supported but there is a Docker image available that runs on Windows.
+
+Users of Debian 9 or later or Ubuntu 16.10 or later may simply
+
+```bash
+apt-get install ocrmypdf
+```
+
+and macOS users with Homebrew may simply
+
+```bash
+brew install ocrmypdf
+```
+
+For everyone else, [see our documentation](https://ocrmypdf.readthedocs.io/en/latest/installation.html) for installation steps.
+
+Languages
+---------
+
+OCRmyPDF uses Tesseract for OCR, and relies on its language packs. For Linux users, you can often find packages that provide language packs:
+
+```bash
+# Display a list of all Tesseract language packs
+apt-cache search tesseract-ocr
+
+# Debian/Ubuntu users
+apt-get install tesseract-ocr-chi-sim  # Example: Install Chinese Simplified language back
+```
+
+You can then pass the `-l LANG` argument to OCRmyPDF to give a hint as to what languages it should search for. Multiple languages can be requested.
+
+Documentation and support
+-------------------------
+
+Once ocrmypdf is installed, the built-in help which explains the command syntax and options can be accessed via:
+
+```bash
+ocrmypdf --help
+```
+
+Our [documentation is served on Read the Docs](https://ocrmypdf.readthedocs.io/en/latest/index.html).
+
+If you detect an issue, please:
+
+-   Check whether your issue is already known
+-   If no problem report exists on github, please create one here: <https://github.com/jbarlow83/OCRmyPDF/issues>
+-   Describe your problem thoroughly
+-   Append the console output of the script when running the debug mode (`-v 1` option)
+-   If possible provide your input PDF file as well as the content of the temporary folder (using a file sharing service like Dropbox)
+
+Requirements
+------------
+
+Runs on CPython 3.5, 3.6 and 3.7. Requires external program installations of Ghostscript, Tesseract OCR, QPDF, and Leptonica. ocrmypdf is pure Python, but uses CFFI to portably generate library bindings.
+
+Press & Media
+-------------
+
+-   [c't 1-2014, page 59](http://heise.de/-2279695): Detailed presentation of OCRmyPDF v1.0 in the leading German IT magazine c't
+-   [heise Open Source, 09/2014: Texterkennung mit OCRmyPDF](http://heise.de/-2356670)
+
+License
+-------
+
+The OCRmyPDF software is licensed under the GNU GPLv3. Certain files are covered by other licenses, as noted in their source files.
+
+The license for each test file varies, and is noted in tests/resources/README.rst. The documentation is licensed under Creative Commons Attribution-ShareAlike 4.0 (CC-BY-SA 4.0).
+
+OCRmyPDF versions prior to 6.0 were licensed under the MIT License.
+
+Disclaimer
+----------
+
+The software is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.

README.rst

@@ -1,150 +0,0 @@
-OCRmyPDF
-========
-
-.. image:: https://travis-ci.org/jbarlow83/OCRmyPDF.svg?branch=master
-    :target: https://travis-ci.org/jbarlow83/OCRmyPDF
-
-.. image:: https://img.shields.io/pypi/v/ocrmypdf.svg
-    :target: https://pypi.org/project/ocrmypdf/
-
-.. image:: https://img.shields.io/homebrew/v/ocrmypdf.svg
-    :alt: homebrew
-    :target: http://brewformulas.org/Ocrmypdf
-
-
-OCRmyPDF adds an OCR text layer to scanned PDF files, allowing them to
-be searched or copy-pasted.
-
-.. code-block:: bash
-
-   ocrmypdf                      # it's a scriptable command line program
-      -l eng+fra                 # it supports multiple languages
-      --rotate-pages             # it can fix pages that are misrotated
-      --deskew                   # it can deskew crooked PDFs!
-      --title "My PDF"           # it can change output metadata
-      --jobs 4                   # it uses multiple cores by default
-      --output-type pdfa         # it produces PDF/A by default
-      input_scanned.pdf          # takes PDF input (or images)
-      output_searchable.pdf      # produces validated PDF output
-
-
-Main features
--------------
-
--  Generates a searchable
-   `PDF/A <https://en.wikipedia.org/?title=PDF/A>`_ file from a regular PDF
--  Places OCR text accurately below the image to ease copy / paste
--  Keeps the exact resolution of the original embedded images
--  When possible, inserts OCR information as a "lossless" operation without disrupting any other content
--  Optimizes PDF images, often producing files smaller than the input file
--  If requested deskews and/or cleans the image before performing OCR
--  Validates input and output files
--  Distributes work across all available CPU cores
--  Uses `Tesseract OCR <https://github.com/tesseract-ocr/tesseract>`_ engine
--  Supports more than `100 languages <https://github.com/tesseract-ocr/tessdata>`_ recognized by Tesseract
--  Battle-tested on thousands of PDFs, a test suite and continuous integration
-
-For details: please consult the `documentation <https://ocrmypdf.readthedocs.io/en/latest/>`_.
-
-Motivation
-----------
-
-I searched the web for a free command line tool to OCR PDF files on
-Linux/UNIX: I found many, but none of them were really satisfying.
-
--  Either they produced PDF files with misplaced text under the image (making copy/paste impossible)
--  Or they did not handle accents and multilingual characters
--  Or they changed the resolution of the embedded images
--  Or they generated ridiculously large PDF files
--  Or they crashed when trying to OCR
--  Or they did not produce valid PDF files
--  On top of that none of them produced PDF/A files (format dedicated for long time storage)
-
-...so I decided to develop my own tool.
-
-Installation
-------------
-
-Linux, UNIX, and macOS are supported. Windows is not directly supported but there is a Docker image available that runs on Windows.
-
-Users of Debian 9 or later or Ubuntu 16.10 or later may simply
-
-.. code-block:: bash
-
-   apt-get install ocrmypdf
-
-and macOS users with Homebrew may simply
-
-.. code-block:: bash
-
-   brew install ocrmypdf
-
-For everyone else, `see our documentation <https://ocrmypdf.readthedocs.io/en/latest/installation.html>`_ for installation steps.
-
-Languages
----------
-
-OCRmyPDF uses Tesseract for OCR, and relies on its language packs. For Linux users,
-you can often find packages that provide language packs:
-
-.. code-block:: bash
-
-   # Display a list of all Tesseract language packs
-   apt-cache search tesseract-ocr
-
-   # Debian/Ubuntu users
-   apt-get install tesseract-ocr-chi-sim  # Example: Install Chinese Simplified language back
-
-You can then pass the ``-l LANG`` argument to OCRmyPDF to give a hint as to what languages it should search for. Multiple
-languages can be requested.
-
-Documentation and support
--------------------------
-
-Once ocrmypdf is installed, the built-in help which explains the command syntax and options can be accessed via:
-
-.. code-block:: bash
-
-   ocrmypdf --help
-
-Our `documentation is served on Read the Docs <https://ocrmypdf.readthedocs.io/en/latest/index.html>`_.
-
-If you detect an issue, please:
-
--  Check whether your issue is already known
--  If no problem report exists on github, please create one here:
-   https://github.com/jbarlow83/OCRmyPDF/issues
--  Describe your problem thoroughly
--  Append the console output of the script when running the debug mode
-   (``-v 1`` option)
--  If possible provide your input PDF file as well as the content of the
-   temporary folder (using a file sharing service like Dropbox)
-
-Requirements
-------------
-
-Runs on CPython 3.5, 3.6 and 3.7. Requires external program installations of Ghostscript, Tesseract OCR, QPDF, and Leptonica. ocrmypdf is pure Python, but uses CFFI to portably generate library bindings.
-
-Press & Media
--------------
-
--  `c't 1-2014, page 59 <http://heise.de/-2279695>`_:
-   Detailed presentation of OCRmyPDF v1.0 in the leading German IT
-   magazine c't
--  `heise Open Source, 09/2014: Texterkennung mit
-   OCRmyPDF <http://heise.de/-2356670>`_
-
-License
--------
-
-The OCRmyPDF software is licensed under the GNU GPLv3. Certain files are covered by other licenses, as noted in their source files.
-
-The license for each test file varies, and is noted in tests/resources/README.rst. The documentation is licensed under Creative Commons Attribution-ShareAlike 4.0 (CC-BY-SA 4.0).
-
-OCRmyPDF versions prior to 6.0 were licensed under the MIT License.
-
-Disclaimer
-----------
-
-The software is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
-CONDITIONS OF ANY KIND, either express or implied.

docs/cookbook.rst

@@ -213,4 +213,6 @@ Currently optimization attempts to find more efficient encodings for images. The
 
 .. code-block:: bash
 
-    ocrmypdf --optimize 3 in.pdf out.pdf  # Make it as small as possible
+    ocrmypdf --optimize 3 in.pdf out.pdf  # Make it small
+
+Some users may consider enabling lossy JBIG2. See: :ref:`jbig2-lossy`.

docs/jbig2.rst

@@ -20,3 +20,16 @@ For all other Linux, you must build a JBIG2 encoder from source:
     ./autogen.sh
     ./configure && make
     [sudo] make install
+
+.. _jbig2-lossy:
+
+Lossy mode JBIG2
+----------------
+
+OCRmyPDF provides lossy mode JBIG2 as an advanced feature. Users should `review the technical concerns with JBIG2 in lossy mode <https://abbyy.technology/en:kb:tip:jbig2_compression_and_ocr>`_ and decide if this feature is acceptable for their use case.
+
+JBIG2 lossy mode does achieve higher compression ratios than any other monochrome (bitonal) compression technology; for large text documents the savings are considerable. JBIG2 lossless still gives great compression ratios and is a major improvement over the older CCITT G4 standard. As explained above, there is some risk of substitution errors.
+
+To turn on JBIG2 lossy mode, add the argument ``--jbig2-lossy``. ``--optimize {1,2,3}`` are necessary for the argument to take effect also required. Also, a JBIG2 encoder must be installed as described in the previous section.
+
+*ocrmypdf v7.0 and v7.1 used lossy mode by default.*

docs/release_notes.rst

@@ -13,6 +13,45 @@ Note that it is licensed under GPLv3, so scripts that ``import ocrmypdf`` and ar
    find:    [^`]\#([0-9]{1,3})[^0-9]
    replace: `#$1 <https://github.com/jbarlow83/OCRmyPDF/issues/$1>`_
 
+
+v7.2.0
+------
+
+**Lossy JBIG2 behavior change**
+
+A user reported that ocrmypdf was in fact using JBIG2 in **lossy** compression mode. This was not the intended behavior. Users should `review the technical concerns with JBIG2 in lossy mode <https://abbyy.technology/en:kb:tip:jbig2_compression_and_ocr>`_ and decide if this is a concern for their use case.
+
+JBIG2 lossy mode does achieve higher compression ratios than any other monochrome compression technology; for large text documents the savings are considerable. JBIG2 lossless still gives great compression ratios and is a major improvement over the older CCITT G4 standard.
+
+Only users who have reviewed the concerns with JBIG2 in lossy mode should opt-in. As such, lossy mode JBIG2 is only turned on when the new argument ``--jbig2-lossy`` is issued. This is independent of the setting for ``--optimize``.
+
+Users who did not install an optional JBIG2 encoder are unaffected.
+
+(Thanks to user 'bsdice' for reporting this issue.)
+
+**Other issues**
+
+-   When the image optimizer quantizes an image to 1 bit per pixel, it will now attempt to further optimize that image as CCITT or JBIG2, instead of keeping it in the "flate" encoding which is not efficient for 1 bpp images. (`#297 <https://github.com/jbarlow83/OCRmyPDF/issues/297>`_)
+
+-   Images in PDFs that are used as soft masks (i.e. transparency masks or alpha channels) are now excluded from optimization.
+
+-   Fixed handling of Tesseract 4.0-rc1 which now accepts invalid Tesseract configuration files, which broke the test suite.
+
+v7.1.0
+------
+
+-   Improve the performance of initial text extraction, which is done to determine if a file contains existing text of some kind or not. On large files, this initial processing is now about 20x times faster. (`#299 <https://github.com/jbarlow83/OCRmyPDF/issues/299>`_)
+
+-   pikepdf 0.3.3 is now required.
+
+-   Fixed issue `#231 <https://github.com/jbarlow83/OCRmyPDF/issues/231>`_, a problem with JPEG2000 images where image metadata was only available inside the JPEG2000 file.
+
+-   Fixed some additional Ghostscript 9.25 compatibility issues.
+
+-   Improved handling of KeyboardInterrupt error messages. (`#301 <https://github.com/jbarlow83/OCRmyPDF/issues/301>`_)
+
+-   README.md is now served in GitHub markdown instead of reStructuredText.
+
 v7.0.6
 ------
 
@@ -105,6 +144,13 @@ v7.0.0
 
     +   It may be necessary to separately ``pip install pycparser`` to avoid `another Python 3.7 issue <https://github.com/eliben/pycparser/pull/135>`_.
 
+v6.2.4
+------
+
+-   Backport Ghostscript 9.25 compatibility fixes, which removes support for setting Unicode metadata
+-   Backport blacklisting Ghostscript 9.24
+-   Older versions of Ghostscript are still supported
+
 v6.2.3
 ------
 

requirements.txt

@@ -3,7 +3,7 @@
 # installation
 cffi == 1.11.5
 img2pdf == 0.3.0
-pikepdf == 0.3.2
+pikepdf == 0.3.4
 Pillow >= 5.0.0, != 5.1.0 ; sys_platform == "darwin"
 pycparser == 2.18
 python-xmp-toolkit == 2.0.1

setup.py

@@ -205,13 +205,14 @@ tests_require = open('test_requirements.txt').read().splitlines()
 
 
 def readme():
-    with open('README.rst') as f:
+    with open('README.md') as f:
         return f.read()
 
 setup(
     name='ocrmypdf',
     description='OCRmyPDF adds an OCR text layer to scanned PDF files, allowing them to be searched',
     long_description=readme(),
+    long_description_content_type='text/markdown',
     url='https://github.com/jbarlow83/OCRmyPDF',
     author='James R. Barlow',
     author_email='jim@purplerock.ca',
@@ -250,7 +251,7 @@ setup(
     install_requires=[
         'cffi >= 1.9.1',          # must be a setup and install requirement
         'img2pdf >= 0.2.4, < 0.4',       # pure Python, so track HEAD closely
-        'pikepdf >= 0.3.2, < 0.4',
+        'pikepdf >= 0.3.3, < 0.4',
         'Pillow >= 4.0.0, != 5.1.0 ; sys_platform == "darwin"',
                                   # Pillow < 4 has BytesIO/TIFF bug w/img2pdf 0.2.3
                                   # block 5.1.0, broken wheels
@@ -269,4 +270,10 @@ setup(
     },
     package_data={'ocrmypdf': ['data/sRGB.icc']},
     include_package_data=True,
-    zip_safe=False)
+    zip_safe=False,
+    project_urls={
+        'Documentation': 'https://ocrmypdf.readthedocs.io/',
+        'Source': 'https://github.com/jbarlow83/ocrmypdf',
+        'Tracker': 'https://github.com/jbarlow83/ocrmypdf/issues'
+    }
+)

src/ocrmypdf/__main__.py

@@ -17,7 +17,6 @@
 # along with OCRmyPDF.  If not, see <http://www.gnu.org/licenses/>.
 
 from tempfile import mkdtemp
-from collections.abc import Sequence
 from pathlib import Path
 import sys
 import os
@@ -36,7 +35,7 @@ import ruffus.proxy_logger as proxy_logger
 from ._jobcontext import JobContext, JobContextManager, cleanup_working_files
 from ._pipeline import build_pipeline
 from .pdfa import file_claims_pdfa
-from .helpers import is_iterable_notstr, re_symlink, is_file_writable, \
+from .helpers import re_symlink, is_file_writable, \
     available_cpu_count
 from .exec import tesseract, qpdf, ghostscript
 from . import PROGRAM_NAME, VERSION
@@ -270,17 +269,17 @@ optimizing = parser.add_argument_group(
 optimizing.add_argument(
     '-O', '--optimize', type=int, choices=range(0, 4), default=1,
     help=("Control how PDF is optimized after processing:"
-        "0 - do not optimize;"
-        "1 - do safe, lossless optimizations (default);"
-        "2 - do lossy optimizations; "
-        "3 - do aggressive lossy optimizations"
+        "0 - do not optimize; "
+        "1 - do safe, lossless optimizations (default); "
+        "2 - do some lossy optimizations; "
+        "3 - do aggressive lossy optimizations (including lossy JBIG2)"
     )
 )
 optimizing.add_argument(
     '--jpeg-quality', type=numeric(int, 0, 100), default=0, metavar='Q',
     help=("Adjust JPEG quality level for JPEG optimization. "
           "100 is best quality and largest output size; "
-          "1 is lowest quality and smallest output"
+          "1 is lowest quality and smallest output; "
           "0 uses the default."
     )
 )
@@ -295,6 +294,18 @@ optimizing.add_argument(
           "Values have same meaning as with --jpeg-quality"
     )
 )
+optimizing.add_argument(
+    '--jbig2-lossy', action='store_true',
+    help=("Enable JBIG2 lossy mode (better compression, not suitable for some "
+          "use cases - see documentation)."
+    )
+)
+optimizing.add_argument(
+    '--jbig2-page-group-size', type=numeric(int, 1, 10000), default=0,
+    metavar='N',
+    # Adjust number of pages to consider at once for JBIG2 compression
+    help=argparse.SUPPRESS
+)
 
 advanced = parser.add_argument_group(
     "Advanced",
@@ -467,21 +478,34 @@ def check_options_sidecar(options, log):
         options.sidecar = options.output_file + '.txt'
 
 
-def _optional_program_check(name, version_fn, min_version, for_argument):
+def _optional_program_required(name, version_fn, min_version, for_argument):
     try:
         if version_fn() < min_version:
             raise MissingDependencyError(
                 "The installed '{}' is not supported. "
                 "Install version {} or newer.".format(name, min_version))
-    except FileNotFoundError:
+    except (FileNotFoundError, MissingDependencyError):
         raise MissingDependencyError(
             "Install the '{}' program to use {}.".format(name, for_argument))
 
 
+def _optional_program_recommended(name, version_fn, min_version, for_argument):
+    try:
+        if version_fn() < min_version:
+            raise MissingDependencyError(
+                "The installed '{}' is not supported. "
+                "Install version {} or newer.".format(name, min_version))
+    except (FileNotFoundError, MissingDependencyError):
+        complain(
+            "For best results, install the optional program '{}' to use the "
+            "argument {}.".format(name, for_argument)
+        )
+
+
 def check_options_preprocessing(options, log):
     if any((options.clean, options.clean_final)):
         from .exec import unpaper
-        _optional_program_check(
+        _optional_program_required(
             'unpaper', unpaper.version, '6.1', '--clean, --clean-final'
         )
 
@@ -496,13 +520,29 @@ def check_options_ocr_behavior(options, log):
 def check_options_optimizing(options, log):
     if options.optimize >= 2:
         from .exec import pngquant, jbig2enc
-        _optional_program_check(
+        _optional_program_required(
             'pngquant', pngquant.version, '2.0.1', '--optimize {2,3}'
         )
-        _optional_program_check(
+
+    if options.jbig2_lossy:
+        _optional_program_required(
+            'jbig2', jbig2enc.version, '0.28', '--jbig2-lossy'
+        )
+    elif options.optimize >= 2:
+        # Although we use JBIG2 for optimize=1, don't nag about it unless the
+        # user is asking for more optimization
+        _optional_program_recommended(
             'jbig2', jbig2enc.version, '0.28', '--optimize {2,3}'
         )
 
+    if options.optimize == 0 and any([
+            options.jbig2_lossy, options.png_quality, options.jpeg_quality
+            ]):
+        log.warning(
+            "The arguments --jbig2-lossy, --png-quality, and --jpeg-quality "
+            "will be ignored because --optimize=0."
+        )
+
 
 def check_options_advanced(options, log):
     if options.tesseract_oem and not tesseract.v4():
@@ -598,52 +638,45 @@ def do_ruffus_exception(ruffus_five_tuple, options, log):
     description of the error message that occurred."""
     exit_code = None
 
-    task_name, job_name, exc_name, exc_value, exc_stack = ruffus_five_tuple
-    task_name = task_name  # unused
-    job_name = job_name  # unused
-    if exc_name == 'builtins.SystemExit':
+    _task_name, _job_name, exc_name, exc_value, exc_stack = ruffus_five_tuple
+
+    if isinstance(exc_name, type):
+        # ruffus is full of mystery... sometimes (probably when the process
+        # group leader is killed) exc_name is the class object of the exception,
+        # rather than a str. So reach into the object and get its name.
+        exc_name = exc_name.__name__
+
+    if exc_name.startswith('ocrmypdf.exceptions.'):
+        base_exc_name = exc_name.replace('ocrmypdf.exceptions.', '')
+        exc_class = getattr(ocrmypdf_exceptions, base_exc_name)
+        exit_code = getattr(exc_class, 'exit_code', ExitCode.other_error)
+        try:
+            if isinstance(exc_value, exc_class):
+                exc_msg = str(exc_value)
+            else:
+                exc_msg = str(exc_class())
+        except Exception:
+            exc_msg = "Unknown"
+
+    if exc_name in ('builtins.SystemExit', 'SystemExit'):
         match = re.search(r"\.(.+?)\)", exc_value)
         exit_code_name = match.groups()[0]
         exit_code = getattr(ExitCode, exit_code_name, 'other_error')
     elif exc_name == 'ruffus.ruffus_exceptions.MissingInputFileError':
         log.error(cleanup_ruffus_error_message(exc_value))
         exit_code = ExitCode.input_file
-    elif exc_name == 'builtins.KeyboardInterrupt':
-        log.error("Interrupted by user")
+    elif exc_name in ('builtins.KeyboardInterrupt', 'KeyboardInterrupt'):
+        # We have to print in this case because the log daemon might be toast
+        print("Interrupted by user", file=sys.stderr)
         exit_code = ExitCode.ctrl_c
     elif exc_name == 'subprocess.CalledProcessError':
         # It's up to the subprocess handler to report something useful
         msg = "Error occurred while running this command:"
         log.error(msg + '\n' + exc_value)
         exit_code = ExitCode.child_process_error
-    elif exc_name == 'ocrmypdf.exceptions.EncryptedPdfError':
-        log.error(textwrap.dedent("""\
-            Input PDF is encrypted. The encryption must be removed to
-            perform OCR.
-
-            For information about this PDF's security use
-                qpdf --show-encryption infilename
-
-            You can remove the encryption using
-                qpdf --decrypt [--password=[password]] infilename
-
-            """))
-        exit_code = ExitCode.encrypted_pdf
-    elif exc_name == 'ocrmypdf.exceptions.PdfMergeFailedError':
-        log.error(textwrap.dedent("""\
-            Failed to merge PDF image layer with OCR layer
-
-            Usually this happens because the input PDF file is mal-formed and
-            ocrmypdf cannot automatically correct the problem on its own.
-
-            Try using
-                ocrmypdf --pdf-renderer sandwich  [..other args..]
-            """))
-        exit_code = ExitCode.input_file
     elif exc_name.startswith('ocrmypdf.exceptions.'):
-        base_exc_name = exc_name.replace('ocrmypdf.exceptions.', '')
-        exc_class = getattr(ocrmypdf_exceptions, base_exc_name)
-        exit_code = exc_class.exit_code
+        if exc_msg:
+            log.error(exc_msg)
     elif exc_name == 'PIL.Image.DecompressionBombError':
         msg = cleanup_ruffus_error_message(exc_value)
         msg += ("\nUse the --max-image-mpixels argument to set increase the "
@@ -751,6 +784,7 @@ def preamble(_log):
     _log.debug('ocrmypdf ' + VERSION)
     _log.debug('tesseract ' + tesseract.version())
     _log.debug('qpdf ' + qpdf.version())
+    _log.debug('gs ' + ghostscript.version())
 
 
 def check_environ(options, _log):
@@ -835,9 +869,11 @@ def report_output_file_size(options, _log, input_file, output_file):
         """.format(ratio, explanation)))
 
 
-def run_pipeline():
-    options = parser.parse_args()
+def run_pipeline(args=None):
+    options = parser.parse_args(args=args)
     options.verbose_abbreviated_path = 1
+    if os.environ.get('_OCRMYPDF_THREADS'):
+        options.use_threads = True
 
     if not check_closed_streams(options):
         return ExitCode.bad_args

src/ocrmypdf/_pipeline.py

@@ -163,7 +163,7 @@ def repair_and_parse_pdf(
     copyfile(input_file, output_file)
 
     try:
-        pdfinfo = PdfInfo(output_file)
+        pdfinfo = PdfInfo(output_file, log=log)
     except pikepdf.PasswordError as e:
         raise EncryptedPdfError()
     except pikepdf.PdfError as e:

src/ocrmypdf/_weave.py

@@ -169,17 +169,16 @@ def _traverse_toc(pdf_base, visitor_fn, log):
                 continue
             item = node[key]
             if not item.is_indirect:
-                #  or not isinstance(item, pikepdf.Dictionary):
-                # # If there is garbage data, replace the key with an indirect
-                # # ref to None. Kodak Capture Desktop produces keys like these.
-                # log.error('Removing invalid reference from TOC: %s', repr(item))
+                # Direct references are not allowed here, but it's not clear
+                # what we should do if we find any. Removing them is an option:
                 # node[key] = pdf_base.make_indirect(None)
                 continue
             objgen = item.objgen
             if objgen not in visited:
                 queue.add(objgen)
 
-        visitor_fn(pdf_base, node, log)
+        if visitor_fn:
+            visitor_fn(pdf_base, node, log)
 
 
 def _fix_toc(pdf_base, pageref_remap, log):
@@ -282,7 +281,7 @@ def weave_layers(
     # page references in the table of contents. Some PDF generators put invalid
     # references in the ToC, so we want to resolve them to null before we
     # create any references, or the ToC will be corrupted
-    _traverse_toc(pdf_base, lambda *args: None, log)
+    _traverse_toc(pdf_base, None, log)
 
     procset = pdf_base.make_indirect(
         pikepdf.Object.parse(b'[ /PDF /Text /ImageB /ImageC /ImageI ]'))

src/ocrmypdf/exceptions.py

@@ -17,6 +17,7 @@
 
 
 from enum import IntEnum
+from textwrap import dedent
 
 class ExitCode(IntEnum):
     ok = 0
@@ -36,6 +37,13 @@ class ExitCode(IntEnum):
 
 class ExitCodeException(Exception):
     exit_code = ExitCode.other_error
+    message = ""
+
+    def __str__(self):
+        super_msg = super().__str__()  # Don't do str(super())
+        if self.message:
+            return self.message.format(super_msg)
+        return super_msg
 
 
 class BadArgsError(ExitCodeException):
@@ -44,7 +52,15 @@ class BadArgsError(ExitCodeException):
 
 class PdfMergeFailedError(ExitCodeException):
     exit_code = ExitCode.input_file
+    message = dedent('''\
+        Failed to merge PDF image layer with OCR layer
 
+        Usually this happens because the input PDF file is malformed and
+        ocrmypdf cannot automatically correct the problem on its own.
+
+        Try using
+            ocrmypdf --pdf-renderer sandwich  [..other args..]
+        ''')
 
 class MissingDependencyError(ExitCodeException):
     exit_code = ExitCode.missing_dependency
@@ -76,7 +92,18 @@ class SubprocessOutputError(ExitCodeException):
 
 class EncryptedPdfError(ExitCodeException):
     exit_code = ExitCode.encrypted_pdf
+    message = dedent('''\
+        Input PDF is encrypted. The encryption must be removed to
+        perform OCR.
+
+        For information about this PDF's security use
+            qpdf --show-encryption infilename
+
+        You can remove the encryption using
+            qpdf --decrypt [--password=[password]] infilename
+        ''')
 
 
 class TesseractConfigError(ExitCodeException):
     exit_code = ExitCode.invalid_config
+    message = "Error occurred while parsing a Tesseract configuration file"

src/ocrmypdf/exec/__init__.py

@@ -24,7 +24,6 @@ from subprocess import run, STDOUT, PIPE, CalledProcessError
 from ..exceptions import MissingDependencyError
 
 
-
 def get_version(program, *,
         version_arg='--version', regex=r'(\d+(\.\d+)*)'):
     "Get the version of the specified program"

src/ocrmypdf/exec/ghostscript.py

@@ -32,8 +32,7 @@ def version():
 
 
 def jpeg_passthrough_available():
-    """
-    Returns True if the installed version of Ghostscript supports JPEG passthru
+    """Returns True if the installed version of Ghostscript supports JPEG passthru
 
     Prior to 9.23, Ghostscript decode and re-encoded JPEGs internally. In 9.23
     it gained the ability to keep JPEGs unmodified. However, the 9.23
@@ -54,8 +53,7 @@ def _gs_error_reported(stream):
 
 
 def extract_text(input_file, pageno=1):
-    """
-    Use the txtwrite device to get text layout information out
+    """Use the txtwrite device to get text layout information out
 
     For details on options of -dTextFormat see
     https://www.ghostscript.com/doc/current/VectorDevices.htm#TXT
@@ -66,10 +64,18 @@ def extract_text(input_file, pageno=1):
     <span bbox="left top right bottom" font="..." size="...">
     <char bbox="...." c="X"/>
 
+    :param pageno: number of page to extract, or all pages if None
     :return: XML-ish text representation in bytes
-
     """
 
+    if pageno is not None:
+        pages = [
+            '-dFirstPage=%i' % pageno,
+            '-dLastPage=%i' % pageno
+        ]
+    else:
+        pages = []
+
     args_gs = [
         'gs',
         '-dQUIET',
@@ -78,10 +84,9 @@ def extract_text(input_file, pageno=1):
         '-dNOPAUSE',
         '-sDEVICE=txtwrite',
         '-dTextFormat=0',
-        '-dFirstPage=%i' % pageno,
-        '-dLastPage=%i' % pageno,
+    ] + pages + [
         '-o', '-',
-        input_file
+        fspath(input_file)
     ]
 
     p = run(args_gs, stdout=PIPE, stderr=PIPE)
@@ -97,8 +102,7 @@ def extract_text(input_file, pageno=1):
 
 def rasterize_pdf(input_file, output_file, xres, yres, raster_device, log,
                   pageno=1, page_dpi=None, rotation=None):
-    """
-    Rasterize one page of a PDF at resolution (xres, yres) in canvas units.
+    """Rasterize one page of a PDF at resolution (xres, yres) in canvas units.
 
     The image is sized to match the integer pixels dimensions implied by
     (xres, yres) even if those numbers are noninteger. The image's DPI will
@@ -180,6 +184,23 @@ def rasterize_pdf(input_file, output_file, xres, yres, raster_device, log,
 
 def generate_pdfa(pdf_pages, output_file, compression, log,
                   threads=1, pdf_version='1.5', pdfa_part='2'):
+    """Generate a PDF/A.
+
+    The pdf_pages, a list files, will be merged into output_file. One or more
+    PDF files may be merged. One of the files in this list must be a pdfmark
+    file that provides Ghostscript with details on how to perform the PDF/A
+    conversion. By default with we pick PDF/A-2b, but this works for 1 or 3.
+
+    compression can be 'jpeg', 'lossless', or an empty string. In 'jpeg',
+    Ghostscript is instructed to convert color and grayscale images to DCT
+    (JPEG encoding). In 'lossless' Ghostscript is told to convert images to
+    Flate (lossless/PNG). If the parameter is omitted Ghostscript is left to
+    make its own decisions about how to encode images; it appears to use a
+    heuristic to decide how to encode images. As of Ghostscript 9.25, we
+    support passthrough JPEG which allows Ghostscript to avoid transcoding
+    images entirely. (The feature was added in 9.23 but broken, and the 9.24
+    release of Ghostscript had regressions, so we don't support it until 9.25.)
+    """
     compression_args = []
     if compression == 'jpeg':
         compression_args = [

src/ocrmypdf/exec/jbig2enc.py

@@ -30,7 +30,6 @@ def version():
     return get_version('jbig2', regex=r'jbig2enc (\d+(\.\d+)*).*')
 
 
-@lru_cache(maxsize=1)
 def available():
     try:
         version()
@@ -44,10 +43,24 @@ def convert_group(*, cwd, infiles, out_prefix):
         'jbig2',
         '-b',
         out_prefix,
-        '-s',
+        '-s',  # symbol mode (lossy)
+        # '-r', # refinement mode (lossless symbol mode, currently disabled in
+                # jbig2)
         '-p',
     ]
     args.extend(infiles)
     proc = run(args, cwd=cwd, stdout=PIPE, stderr=PIPE)
     proc.check_returncode()
-    return proc
+    return proc
+
+
+def convert_single(*, cwd, infile, outfile):
+    args = [
+        'jbig2',
+        '-p',
+        infile
+    ]
+    with open(outfile, 'wb') as fstdout:
+        proc = run(args, cwd=cwd, stdout=fstdout, stderr=PIPE)
+    proc.check_returncode()
+    return proc

src/ocrmypdf/exec/pngquant.py

@@ -23,7 +23,7 @@ import os
 import shutil
 
 from . import get_version
-from ..exceptions import ExitCode
+from ..exceptions import ExitCode, MissingDependencyError
 
 
 @lru_cache(maxsize=1)
@@ -31,6 +31,14 @@ def version():
     return get_version('pngquant', regex=r'(\d+(\.\d+)*).*')
 
 
+def available():
+    try:
+        version()
+    except MissingDependencyError:
+        return False
+    return True
+
+
 def quantize(input_file, output_file, quality_min, quality_max):
     args = [
         'pngquant',
@@ -42,4 +50,4 @@ def quantize(input_file, output_file, quality_min, quality_max):
         input_file
     ]
     proc = run(args)
-    proc.check_returncode()
+    proc.check_returncode()

src/ocrmypdf/exec/tesseract.py

@@ -190,6 +190,10 @@ def tesseract_log_output(log, stdout, input_file):
             pass  # Appears to be spurious/problem with nonwhite borders
         elif 'Error in boxClipToRectangle' in line:
             pass  # Always appears with pixScanForForeground message
+        elif 'parameter not found: ' in line.lower():
+            log.error(prefix + line.strip())
+            problem = line.split('found: ')[1]
+            raise TesseractConfigError(problem)
         elif 'error' in line.lower() or 'exception' in line.lower():
             log.error(prefix + line.strip())
         elif 'warning' in line.lower():
@@ -263,8 +267,6 @@ def generate_hocr(input_file, output_files, language: list, engine_mode,
         _generate_null_hocr(output_hocr, output_sidecar, input_file)
     except CalledProcessError as e:
         tesseract_log_output(log, e.output, input_file)
-        if b'read_params_file: parameter not found' in e.output:
-            raise TesseractConfigError() from e
         if b'Image too large' in e.output:
             _generate_null_hocr(output_hocr, output_sidecar, input_file)
             return
@@ -350,9 +352,6 @@ def generate_pdf(*, input_image, skip_pdf=None, output_pdf, output_text,
         use_skip_page(text_only, skip_pdf, output_pdf, output_text)
     except CalledProcessError as e:
         tesseract_log_output(log, e.output, input_image)
-        if b'read_params_file: parameter not found' in e.output:
-            raise TesseractConfigError() from e
-
         if b'Image too large' in e.output:
             use_skip_page(text_only, skip_pdf, output_pdf, output_text)
             return

src/ocrmypdf/optimize.py

@@ -30,7 +30,6 @@ from . import leptonica
 from .helpers import re_symlink, fspath
 from .exec import pngquant, jbig2enc
 
-PAGE_GROUP_SIZE = 10
 DEFAULT_JPEG_QUALITY = 75
 DEFAULT_PNG_QUALITY = 70
 
@@ -51,26 +50,34 @@ def tif_name(root, xref):
     return img_name(root, xref, '.tif')
 
 
-def extract_image(*, pike, root, log, image, xref, jbig2s,
-                  pngs, jpegs, options):
+def extract_image_filter(pike, root, log, image, xref):
     if image.Subtype != '/Image':
-        return False
+        return None
     if image.Length < 100:
-        log.debug("Skipping small image, xref {}".format(xref))
-        return False
+        log.debug("Skipping small image, xref %s", xref)
+        return None
 
     pim = pikepdf.PdfImage(image)
 
     if len(pim.filter_decodeparms) > 1:
-        log.debug("Skipping multiply filtered, xref {}".format(xref))
-        return False
+        log.debug("Skipping multiply filtered, xref %s", xref)
+        return None
     filtdp = pim.filter_decodeparms[0]
 
     if pim.bits_per_component > 8:
-        return False  # Don't mess with wide gamut images
+        return None  # Don't mess with wide gamut images
 
     if filtdp[0] == '/JPXDecode':
-        return False  # Don't do JPEG2000
+        return None  # Don't do JPEG2000
+
+    return pim, filtdp
+
+
+def extract_image_jbig2(*, pike, root, log, image, xref, options):
+    result = extract_image_filter(pike, root, log, image, xref)
+    if result is None:
+        return None
+    pim, filtdp = result
 
     if pim.bits_per_component == 1 \
             and filtdp != '/JBIG2Decode' \
@@ -81,9 +88,18 @@ def extract_image(*, pike, root, log, image, xref, jbig2s,
                 ext = pim.extract_to(stream=f)
             imgname.rename(imgname.with_suffix(ext))
         except pikepdf.UnsupportedImageTypeError:
-            return False
-        jbig2s.append((xref, ext))
-    elif filtdp[0] == '/DCTDecode' \
+            return None
+        return xref, ext
+    return None
+
+
+def extract_image_generic(*, pike, root, log, image, xref, options):
+    result = extract_image_filter(pike, root, log, image, xref)
+    if result is None:
+        return None
+    pim, filtdp = result
+
+    if filtdp[0] == '/DCTDecode' \
             and options.optimize >= 2:
         # This is a simple heuristic derived from some training data, that has
         # about a 70% chance of guessing whether the JPEG is high quality,
@@ -92,7 +108,7 @@ def extract_image(*, pike, root, log, image, xref, jbig2s,
         # bytes_per_pixel = int(raw_jpeg.Length) / (w * h)
         # jpeg_quality_estimate = 117.0 * (bytes_per_pixel ** 0.213)
         # if jpeg_quality_estimate < 65:
-        #     return False
+        #     return None
 
         # We could get the ICC profile here, but there's no need to look at it
         # for quality transcoding
@@ -107,88 +123,111 @@ def extract_image(*, pike, root, log, image, xref, jbig2s,
                 ext = pim.extract_to(stream=f)
             imgname.rename(imgname.with_suffix(ext))
         except pikepdf.UnsupportedImageTypeError:
-            return False
-        jpegs.append(xref)
+            return None
+        return xref, ext
     elif pim.indexed \
             and pim.colorspace in pim.SIMPLE_COLORSPACES \
             and options.optimize >= 3:
         # Try to improve on indexed images - these are far from low hanging
         # fruit in most cases
         pim.as_pil_image().save(png_name(root, xref))
-        pngs.append(xref)
+        return xref, '.png'
     elif not pim.indexed and pim.colorspace in pim.SIMPLE_COLORSPACES:
         # An optimization opportunity here, not currently taken, is directly
         # generating a PNG from compressed data
         pim.as_pil_image().save(png_name(root, xref))
-        pngs.append(xref)
+        return xref, '.png'
     else:
-        return False
+        return None
 
     return True
 
 
-def extract_images(pike, root, log, options):
-    # Extract images we can improve
-    changed_xrefs = set()
-    jbig2_groups = defaultdict(lambda: [])
-    jpegs = []
-    pngs = []
+def extract_images(pike, root, log, options, extract_fn):
+    """Extract image using extract_fn
+
+    extract_fn decides where the image is interesting in this case
+    """
+
+    include_xrefs = set()
+    exclude_xrefs = set()
     errors = 0
     for pageno, page in enumerate(pike.pages):
-        group, _ = divmod(pageno, PAGE_GROUP_SIZE)
         try:
             xobjs = page.Resources.XObject
         except AttributeError:
             continue
-        for imname, image in dict(xobjs).items():
+        for _imname, image in dict(xobjs).items():
             if image.objgen[1] != 0:
                 continue  # Ignore images in an incremental PDF
             xref = image.objgen[0]
-            if xref in changed_xrefs:
-                continue  # Don't improve same image twice
-            try:
-                result = extract_image(
-                    pike=pike, root=root, log=log, image=image,
-                    xref=xref, jbig2s=jbig2_groups[group], pngs=pngs,
-                    jpegs=jpegs, options=options
-                )
-                if result:
-                    changed_xrefs.add(xref)
-            except Exception as e:
-                log.debug("Image {} xref {}".format(imname, xref))
-                log.debug(repr(e))
-                errors += 1
+            if hasattr(image, 'SMask'):
+                # Ignore soft masks
+                smask_xref = image.SMask.objgen[0]
+                exclude_xrefs.add(smask_xref)
+            include_xrefs.add(xref)
+
+    working_xrefs = include_xrefs - exclude_xrefs
+    for xref in working_xrefs:
+        image = pike.get_object((xref, 0))
+        try:
+            result = extract_fn(
+                pike=pike, root=root, log=log, image=image,
+                xref=xref, options=options
+            )
+        except Exception as e:
+            log.debug("Image xref %s", xref)
+            log.debug(repr(e))
+            errors += 1
+        else:
+            if result:
+                _, ext = result
+                yield pageno, xref, ext
+
+
+def extract_images_generic(pike, root, log, options):
+    """Extract any >=2bpp image we think we can improve"""
+
+    jpegs = []
+    pngs = []
+    for _, xref, ext in extract_images(
+            pike, root, log, options, extract_image_generic):
+        log.debug('xref = %s ext = %s', xref, ext)
+        if ext == '.png':
+            pngs.append(xref)
+        elif ext == '.jpg':
+            jpegs.append(xref)
+    log.debug(
+        "Optimizable images: "
+        "JPEGs: %s PNGs: %s", len(jpegs), len(pngs)
+    )
+    return jpegs, pngs
+
+
+def extract_images_jbig2(pike, root, log, options):
+    """Extract any bitonal image that we think we can improve as JBIG2"""
+
+    jbig2_groups = defaultdict(list)
+    for pageno, xref, ext in extract_images(
+            pike, root, log, options, extract_image_jbig2):
+        group = pageno // options.jbig2_page_group_size
+        jbig2_groups[group].append((xref, ext))
 
     # Elide empty groups
     jbig2_groups = {group: xrefs for group, xrefs in jbig2_groups.items()
                     if len(xrefs) > 0}
     log.debug(
         "Optimizable images: "
-        "JBIG2 groups: {} JPEGs: {} PNGs: {} Errors: {}".format(
-            len(jbig2_groups), len(jpegs), len(pngs), errors
-    ))
-
-    return jbig2_groups, jpegs, pngs
+        "JBIG2 groups: %s", (len(jbig2_groups),)
+    )
+    return jbig2_groups
 
 
-def convert_to_jbig2(pike, jbig2_groups, root, log, options):
-    """
-    Convert a group of JBIG2 images and insert into PDF.
+def _produce_jbig2_images(jbig2_groups, root, log, options):
+    """Produce JBIG2 images from their groups"""
 
-    We use a group because JBIG2 works best with a symbol dictionary that spans
-    multiple pages. When inserted back into the PDF, each JBIG2 must reference
-    the symbol dictionary it is associated with. So convert a group at a time,
-    and replace their streams with a parameter set that points to the
-    appropriate dictionary.
-
-    If too many pages shared the same dictionary JBIG2 encoding becomes more
-    expensive and less efficient.
-
-    """
-    with concurrent.futures.ThreadPoolExecutor(
-            max_workers=options.jobs) as executor:
-        futures = []
-        for group, xref_exts in jbig2_groups.items():
+    def jbig2_group_futures(executor, root, groups):
+        for group, xref_exts in groups.items():
             prefix = 'group{:08d}'.format(group)
             future = executor.submit(
                 jbig2enc.convert_group,
@@ -196,15 +235,64 @@ def convert_to_jbig2(pike, jbig2_groups, root, log, options):
                 infiles=(img_name(root, xref, ext) for xref, ext in xref_exts),
                 out_prefix=prefix
             )
-            futures.append(future)
+            yield future
+
+    def jbig2_single_futures(executor, root, groups):
+        for group, xref_exts in groups.items():
+            prefix = 'group{:08d}'.format(group)
+            # Second loop is to ensure multiple images per page are unpacked
+            for n, xref_ext in enumerate(xref_exts):
+                xref, ext = xref_ext
+                future = executor.submit(
+                    jbig2enc.convert_single,
+                    cwd=fspath(root),
+                    infile=img_name(root, xref, ext),
+                    outfile=root / ('{}.{:04d}'.format(prefix, n))
+                )
+                yield future
+
+    if options.jbig2_page_group_size > 1:
+        jbig2_futures = jbig2_group_futures
+    else:
+        jbig2_futures = jbig2_single_futures
+
+    with concurrent.futures.ThreadPoolExecutor(
+            max_workers=options.jobs) as executor:
+        futures = jbig2_futures(executor, root, jbig2_groups)
         for future in concurrent.futures.as_completed(futures):
             proc = future.result()
             log.debug(proc.stderr.decode())
 
+
+def convert_to_jbig2(pike, jbig2_groups, root, log, options):
+    """Convert images to JBIG2 and insert into PDF.
+
+    When the JBIG2 page group size is > 1 we do several JBIG2 images at once
+    and build a symbol dictionary that will span several pages. Each JBIG2
+    image must reference to its symbol dictionary. If too many pages shared the
+    same dictionary JBIG2 encoding becomes more expensive and less efficient.
+    The default value of 10 was determined through testing. Currently this
+    must be lossy encoding since jbig2enc does not support refinement coding.
+
+    When the JBIG2 symbolic coder is not used, each JBIG2 stands on its own
+    and needs no dictionary. Currently this is must be lossless JBIG2.
+    """
+
+    _produce_jbig2_images(jbig2_groups, root, log, options)
+
     for group, xref_exts in jbig2_groups.items():
         prefix = 'group{:08d}'.format(group)
-        jbig2_globals_data = (root / (prefix + '.sym')).read_bytes()
-        jbig2_globals = pikepdf.Stream(pike, jbig2_globals_data)
+        jbig2_symfile = root / (prefix + '.sym')
+        if jbig2_symfile.exists():
+            jbig2_globals_data = jbig2_symfile.read_bytes()
+            jbig2_globals = pikepdf.Stream(pike, jbig2_globals_data)
+            jbig2_globals_dict = pikepdf.Dictionary({
+                    '/JBIG2Globals': jbig2_globals
+            })
+        elif options.jbig2_page_group_size == 1:
+            jbig2_globals_dict = None
+        else:
+            raise FileNotFoundError(jbig2_symfile)
 
         for n, xref_ext in enumerate(xref_exts):
             xref, _ = xref_ext
@@ -213,9 +301,7 @@ def convert_to_jbig2(pike, jbig2_groups, root, log, options):
             im_obj = pike.get_object(xref, 0)
             im_obj.write(
                 jbig2_im_data, pikepdf.Name('/JBIG2Decode'),
-                pikepdf.Dictionary({
-                    '/JBIG2Globals': jbig2_globals
-                })
+                jbig2_globals_dict
             )
 
 
@@ -232,9 +318,9 @@ def transcode_jpegs(pike, jpegs, root, log, options):
             im.save(fspath(opt_jpg),
                     optimize=True,
                     quality=options.jpeg_quality)
-        # pylint: disable=E1101
+        # pylint: disable=no-member
         if opt_jpg.stat().st_size > in_jpg.stat().st_size:
-            log.debug("xref {}, jpeg, made larger - skip".format(xref))
+            log.debug("xref %s, jpeg, made larger - skip", xref)
             continue
 
         compdata = leptonica.CompressedData.open(opt_jpg)
@@ -326,21 +412,25 @@ def optimize(
     if options.png_quality == 0:
         options.png_quality = \
                 DEFAULT_PNG_QUALITY if options.optimize < 3 else 30
+    if options.jbig2_page_group_size == 0:
+        options.jbig2_page_group_size = \
+                10 if options.jbig2_lossy else 1
 
     pike = pikepdf.Pdf.open(input_file)
 
     root = Path(output_file).parent / 'images'
-    root.mkdir(exist_ok=True)  # pylint: disable=E1101
-    jbig2_groups, jpegs, pngs = extract_images(
-        pike, root, log, options)
+    root.mkdir(exist_ok=True)  # pylint: disable=no-member
 
-    convert_to_jbig2(pike, jbig2_groups, root, log, options)
+    jpegs, pngs = extract_images_generic(pike, root, log, options)
     transcode_jpegs(pike, jpegs, root, log, options)
     transcode_pngs(pike, pngs, root, log, options)
 
-    # Not object_stream_mode + preserve_pdfa generates noncompliant PDFs
+    jbig2_groups = extract_images_jbig2(pike, root, log, options)
+    convert_to_jbig2(pike, jbig2_groups, root, log, options)
+
     target_file = Path(output_file).with_suffix('.opt.pdf')
-    pike.save(target_file, preserve_pdfa=True)
+    pike.save(target_file, preserve_pdfa=True,
+              object_stream_mode=pikepdf.ObjectStreamMode.generate)
 
     input_size = Path(input_file).stat().st_size
     output_size = Path(target_file).stat().st_size
@@ -363,11 +453,14 @@ def main(infile, outfile, level, jobs=1):
     class OptimizeOptions:
         """Emulate ocrmypdf's options"""
 
-        def __init__(self, jobs, optimize, jpeg_quality, png_quality):
+        def __init__(
+                self, jobs, optimize, jpeg_quality, png_quality, jb2lossy):
             self.jobs = jobs
             self.optimize = optimize
             self.jpeg_quality = jpeg_quality
             self.png_quality = png_quality
+            self.jbig2_page_group_size = 0
+            self.jbig2_lossy = jb2lossy
 
     logging.basicConfig(level=logging.DEBUG)
     log = logging.getLogger()
@@ -377,7 +470,8 @@ def main(infile, outfile, level, jobs=1):
         jobs=jobs,
         optimize=int(level),
         jpeg_quality=0,  # Use default
-        png_quality=0
+        png_quality=0,
+        jb2lossy=False
     )
     ctx.set_options(options)
 

src/ocrmypdf/pdfa.py

@@ -131,6 +131,7 @@ def _encode_ascii(s: str) -> str:
         '(': '',
         ')': '',
         '\\': '',
+        '\0': ''
     })
     return s.translate(trans).encode('ascii', errors='replace').decode()
 
@@ -284,7 +285,7 @@ def generate_pdfa_ps(target_filename, pdfmark, icc='sRGB', ascii_docinfo=False):
     hex_icc_profile = hexlify(bytes_icc_profile)
     icc_profile = '<' + hex_icc_profile.decode('ascii') + '>'
 
-    ps = _get_pdfa_def(icc_profile, icc, pdfmark)
+    ps = _get_pdfa_def(icc_profile, icc, pdfmark, ascii_docinfo=ascii_docinfo)
 
     # We should have encoded everything to pure ASCII by this point, and
     # to be safe, only allow ASCII in PostScript

src/ocrmypdf/pdfinfo.py

@@ -16,12 +16,14 @@
 # You should have received a copy of the GNU General Public License
 # along with OCRmyPDF.  If not, see <http://www.gnu.org/licenses/>.
 
-from decimal import Decimal
-from math import hypot, isclose
-import re
 from collections import namedtuple
-from pathlib import Path
+from decimal import Decimal
 from enum import Enum
+from math import hypot, isclose
+from pathlib import Path
+from unittest.mock import Mock
+import re
+import xml.etree.ElementTree as ET
 
 from .exec import ghostscript
 from .helpers import fspath
@@ -486,27 +488,13 @@ def _find_images(*, pdf, container, shorthand=None):
     yield from _find_form_xobject_images(pdf, container, contentsinfo)
 
 
-def _page_get_textblocks(infile, pageno):
+def _page_get_textblocks(infile, pageno, xmltext):
     """Smarter text detection"""
-    import xml.etree.ElementTree as ET
 
-    gstext = ghostscript.extract_text(infile, pageno+1)
-
-    # Remove all <char /> tags, because they might contain invalid XML entities
-    # like <char bbox="348 596 348 596" c="&#x1;"/> which chokes on the
-    # inclusion of U+0001. Understandably.
-    # Just remove the whole <char /> tag since we don't use it at all, and they
-    # are only generated as innermost self-closing tags.
-    gstext = regex_remove_char_tags.sub(b' ', gstext)
-
-    if gstext.strip() == '':
+    root = xmltext
+    if not hasattr(xmltext, 'findall'):
         return []
 
-    try:
-        root = ET.fromstring(gstext)
-    except ET.ParseError as e:
-        return []  # If we can't parse, assume none...
-
     def blocks():
         for span in root.findall('.//span'):
             bbox_str = span.attrib['bbox']
@@ -565,14 +553,15 @@ def _page_has_text(text_blocks, page_width, page_height):
     return has_text
 
 
-def _pdf_get_pageinfo(pdf, pageno: int, infile):
+def _pdf_get_pageinfo(pdf, pageno: int, infile, xmltext):
     pageinfo = {}
     pageinfo['pageno'] = pageno
     pageinfo['images'] = []
 
     page = pdf.pages[pageno]
 
-    pageinfo['textinfo'] = _page_get_textblocks(fspath(infile), pageno)
+    pageinfo['textinfo'] = _page_get_textblocks(
+        fspath(infile), pageno, xmltext=xmltext)
 
     mediabox = [Decimal(d) for d in page.MediaBox.as_list()]
     width_pt = mediabox[2] - mediabox[0]
@@ -609,16 +598,47 @@ def _pdf_get_pageinfo(pdf, pageno: int, infile):
     return pageinfo
 
 
-def _pdf_get_all_pageinfo(infile):
+def _pdf_get_all_pageinfo(infile, log=None):
+    if not log:
+        log = Mock()
+
     pdf = pikepdf.open(infile)
-    return [PageInfo(pdf, n, infile) for n in range(len(pdf.pages))], pdf
+
+    existing_text = ghostscript.extract_text(infile, pageno=None)
+    existing_text = regex_remove_char_tags.sub(b' ', existing_text)
+
+    try:
+        root = ET.fromstringlist([
+            b'<document>\n', existing_text, b'</document>\n'
+        ])
+        page_xml = root.findall('page')
+    except ET.ParseError as e:
+        log.error(
+            "An error occurred while attempting to retrieve existing text in "
+            "the input file. Will attempt to continue assuming that there is "
+            "no existing text in the file. The error was:")
+        log.error(e)
+        page_xml = [None] * len(pdf.pages)
+
+    page_count_difference = len(pdf.pages) - len(page_xml)
+    if page_count_difference != 0:
+        log.error("The number of pages in the input file is inconsistent.")
+        if page_count_difference > 0:
+            page_xml.extend([None] * page_count_difference)
+
+    pages = []
+    for n in range(len(pdf.pages)):
+        page = PageInfo(pdf, n, infile, page_xml[n])
+        pages.append(page)
+
+    return pages, pdf
 
 
 class PageInfo:
-    def __init__(self, pdf, pageno, infile):
+    def __init__(self, pdf, pageno, infile, xmltext):
         self._pageno = pageno
         self._infile = infile
-        self._pageinfo = _pdf_get_pageinfo(pdf, pageno, infile)
+        self._pageinfo = _pdf_get_pageinfo(pdf, pageno, infile, xmltext)
 
     @property
     def pageno(self):
@@ -695,9 +715,9 @@ class PdfInfo:
     """Get summary information about a PDF
 
     """
-    def __init__(self, infile):
+    def __init__(self, infile, log=None):
         self._infile = infile
-        self._pages, pdf = _pdf_get_all_pageinfo(infile)
+        self._pages, pdf = _pdf_get_all_pageinfo(infile, log=log)
         self._needs_rendering = pdf.root.get('/NeedsRendering', False)
 
     @property

tests/conftest.py

@@ -134,7 +134,8 @@ def check_ocrmypdf(input_file, output_file, *args, env=None):
     "Run ocrmypdf and confirmed that a valid file was created"
 
     p, out, err = run_ocrmypdf(input_file, output_file, *args, env=env)
-    print(err)  # ensure py.test collects the output, use -s to view
+    # ensure py.test collects the output, use -s to view
+    print(err, file=sys.stderr)
     assert p.returncode == 0
     assert os.path.exists(str(output_file)), "Output file not created"
     assert os.stat(str(output_file)).st_size > 100, "PDF too small or empty"

tests/spoof/gs_render_failure.py

@@ -20,11 +20,11 @@
 # TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
 # SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
+"""Replicate Ghostscript render failure while allowing rasterizing"""
 
 import sys
 import os
 
-"""Replicate Ghostscript render failure while allowing rasterizing"""
 
 
 def real_ghostscript(argv):

tests/test_main.py

@@ -595,7 +595,7 @@ THIS FILE IS INVALID
         resources / 'ccitt.pdf', outdir / 'out.pdf',
         '--pdf-renderer', renderer,
         '--tesseract-config', cfg_file)
-    assert "parameter not found" in err, "No error message"
+    assert "parameter not found" in err.lower(), "No error message"
     assert p.returncode == ExitCode.invalid_config
 
 

tests/test_optimize.py

@@ -22,9 +22,11 @@ import logging
 
 from PIL import Image
 
+import pikepdf
+
 from ocrmypdf import optimize as opt
 from ocrmypdf.exec.ghostscript import rasterize_pdf
-from ocrmypdf.exec import jbig2enc
+from ocrmypdf.exec import jbig2enc, pngquant
 from ocrmypdf.helpers import fspath
 
 
@@ -53,10 +55,53 @@ def test_mono_not_inverted(resources, outdir):
     assert im.getpixel((0, 0)) == 255, "Expected white background"
 
 
-@pytest.mark.skipif(not jbig2enc.available(), reason='need jbig2enc')
 def test_jpg_png_params(resources, outpdf, spoof_tesseract_noop):
     check_ocrmypdf(
         resources / 'crom.png', outpdf, '--image-dpi', '200',
-        '--optimize', '2', '--jpg-quality', '50', '--png-quality', '20',
+        '--optimize', '3', '--jpg-quality', '50', '--png-quality', '20',
         env=spoof_tesseract_noop
     )
+
+
+@pytest.mark.skipif(not jbig2enc.available(), reason='need jbig2enc')
+@pytest.mark.parametrize('lossy', [False, True])
+def test_jbig2_lossy(lossy, resources, outpdf, spoof_tesseract_noop):
+    args = [
+        resources / 'ccitt.pdf', outpdf, '--image-dpi', '200',
+        '--optimize', 3, '--jpg-quality', '50', '--png-quality', '20'
+    ]
+    if lossy:
+        args.append('--jbig2-lossy')
+
+    check_ocrmypdf(*args, env=spoof_tesseract_noop)
+
+    pdf = pikepdf.open(outpdf)
+    pim = pikepdf.PdfImage(next(iter(pdf.pages[0].images.values())))
+    assert pim.filters[0] == '/JBIG2Decode'
+
+    if lossy:
+        assert '/JBIG2Globals' in pim.decode_parms[0]
+    else:
+        assert len(pim.decode_parms) == 0
+
+
+@pytest.mark.skipif(not jbig2enc.available() or not pngquant.available(),
+                    reason='need jbig2enc and pngquant')
+def test_flate_to_jbig2(resources, outdir, spoof_tesseract_noop):
+    # This test requires an image that pngquant is capable of converting to
+    # to 1bpp - so use an existing 1bpp image, convert up, confirm it can
+    # convert down
+    im = Image.open(fspath(resources / 'typewriter.png'))
+    assert im.mode in ('1', 'P')
+    im = im.convert('L')
+    im.save(fspath(outdir / 'type8.png'))
+
+    check_ocrmypdf(
+        outdir / 'type8.png', outdir / 'out.pdf',
+        '--image-dpi', '100', '--png-quality', '10', '--optimize', '3',
+        env=spoof_tesseract_noop
+    )
+
+    pdf = pikepdf.open(outdir / 'out.pdf')
+    pim = pikepdf.PdfImage(next(iter(pdf.pages[0].images.values())))
+    assert pim.filters[0] == '/JBIG2Decode'

tests/test_tess4.py

@@ -29,8 +29,7 @@ from pathlib import Path
 spoof = pytest.helpers.spoof
 
 
-@pytest.fixture
-def ensure_tess4():
+def _ensure_tess4():
     if tesseract.v4():
         # "tesseract" on $PATH is already v4
         return os.environ.copy()
@@ -49,6 +48,11 @@ def ensure_tess4():
     raise EnvironmentError("Can't find Tesseract 4")
 
 
+@pytest.fixture
+def ensure_tess4():
+    return _ensure_tess4()
+
+
 @contextmanager
 def modified_os_environ(env):
     old_env = os.environ.copy()
@@ -63,8 +67,8 @@ def tess4_available():
 
     """
     try:
-        # ensure_tess4 locates the tess4 binary we are going to check
-        env = ensure_tess4()
+        # _ensure_tess4 locates the tess4 binary we are going to check
+        env = _ensure_tess4()
         with modified_os_environ(env):
             # Now jump into this environment and make sure it really is Tess4
             return tesseract.v4() and tesseract.has_textonly_pdf()