v4.1 release notes

I found this issue in ruffus 2.6.3
https://github.com/bunbun/ruffus/issues/65
also discussed here
https://github.com/bunbun/ruffus/pull/67

ruffus 2.6.3 RethrownJobError don't follow the normal conventions and
so its exception causes problems when they cross process boundaries.
This change carefully examines the various forms of ruffus exception
objects that can appear in 2.6.3 and parses them more carefully. It
also removes any direct posting of the exception to the logger because
this triggers another serializing of the exception object, mutating it
further.

.git_archival.txt

@@ -0,0 +1 @@
+ref-names: $Format:%D$

.gitattributes

@@ -5,4 +5,6 @@
 # (binary is a macro for -text -diff)
 *.jar	binary
 *.pdf	binary
-*.PDF	binary
+*.PDF	binary
+
+.git_archival.txt  export-subst

.travis.yml

@@ -5,7 +5,7 @@ cache: pip
 cache:
     directories:
         - $HOME/.ccache
-        - tarballs
+        - packages
         - tests/cache
 
 python:
@@ -21,17 +21,15 @@ before_install:
     - sudo add-apt-repository ppa:b-eltzner/qpdfview-exp -y  # for QPDF 5
     - sudo add-apt-repository ppa:itachi-san/ffmpeg -y  # for libav 11.2 (for unpaper)
     - sudo apt-get update -qq  # must go after all add-apt-repo
-    - sudo apt-get install -y ghostscript tesseract-ocr tesseract-ocr-deu tesseract-ocr-eng tesseract-ocr-fra qpdf poppler-utils gcc libavformat-dev libavcodec-dev libavutil-dev automake make pkg-config xsltproc libffi-dev
+    - sudo apt-get install -y ghostscript tesseract-ocr tesseract-ocr-deu tesseract-ocr-eng tesseract-ocr-fra qpdf poppler-utils libavformat56 libavcodec56 libavutil54 libffi-dev
 
     # pip
     - pip install --upgrade pip
 
     # Download, make and install unpaper (using ccache)
-    - mkdir -p tarballs
-    - "[ -f tarballs/unpaper-6.1.tar.xz ] || wget -q https://www.flameeyes.eu/files/unpaper-6.1.tar.xz -O tarballs/unpaper-6.1.tar.xz"
-    - tar -xvf tarballs/unpaper-6.1.tar.xz
-    - export PATH="/usr/lib/ccache:$PATH"
-    - pushd unpaper-6.1 && ./configure --prefix=/usr && make -j && sudo make install && popd
+    - mkdir -p packages
+    - "[ -f packages/unpaper_6.1-1.deb ] || wget -q https://dl.dropboxusercontent.com/u/28971240/unpaper_6.1-1.deb -O packages/unpaper_6.1-1.deb"
+    - sudo dpkg -i packages/unpaper_6.1-1.deb
 
 install:
     - pip install -r requirements.txt

Dockerfile

@@ -23,12 +23,8 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 
 RUN apt-get install -y --no-install-recommends \
   unpaper \
-  libopenjpeg-dev \
-  ghostscript \
   qpdf \
-  poppler-utils
-
-RUN apt-get install -y --no-install-recommends \
+  poppler-utils \
   tesseract-ocr \
   tesseract-ocr-deu tesseract-ocr-spa tesseract-ocr-eng tesseract-ocr-fra
 
@@ -37,6 +33,14 @@ RUN apt-get install -qy --no-install-recommends \
   libpython3-dev \
   gcc
 
+# Install Ghostscript from Debian sid to work around JPEG 2000 issue in
+# Debian stretch libgs9 or gs 9.16~dfsg-2.1
+
+COPY ./share/etc-apt-sources.list /etc/apt/sources.list
+
+RUN apt-get update && apt-get install -y ghostscript/sid
+
+
 # Enforce UTF-8
 # Borrowed from https://index.docker.io/u/crosbymichael/python/ 
 RUN dpkg-reconfigure locales && \

LICENSE.rst

@@ -18,3 +18,9 @@ IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
 CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
 TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
 SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+----
+
+sRGB_IEC61966-2-1_black_scaled.icc 
+
+To anyone who acknowledges that the file "sRGB_IEC61966-2-1_black scaled.icc" is provided "AS IS" WITH NO EXPRESS OR IMPLIED WARRANTY, permission to use, copy and distribute these file for any purpose is hereby granted without fee, provided that the file is not changed including the ICC copyright notice tag, and that the name of ICC shall not be used in advertising or publicity pertaining to distribution of the software without specific, written prior permission. ICC makes no representations about the suitability of this software for any purpose.

README.rst

@@ -48,6 +48,12 @@ Download OCRmyPDF here: https://github.com/jbarlow83/OCRmyPDF/releases
 
 You can install it to a Python virtual environment or system-wide. 
 
+Debian and Ubuntu
+~~~~~~~~~~~~~~~~~
+
+Users of Debian 9 or later or Ubuntu 16.10 or later may simply
+``apt-get install ocrmypdf``.
+
 Installing the Docker image
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
@@ -116,8 +122,8 @@ Install or upgrade the required Homebrew packages, if any are missing::
    brew install qpdf
    brew install ghostscript
    brew install python3
-   brew install libxml2
-   brew install leptonica
+   brew install libxml2 libffi leptonica
+   brew install unpaper    # optional
    brew install tesseract
    
 Update the homebrew pip and install Pillow::
@@ -148,6 +154,7 @@ Install system dependencies::
    sudo apt-get install \
       zlib1g-dev \
       libjpeg-dev \
+      libffi-dev \
       ghostscript \
       tesseract-ocr \
       qpdf \
@@ -252,11 +259,11 @@ In case you detect an issue, please:
 Press & Media
 -------------
 
--  `c't 1-2014, page 59 <http://www.heise.de/ct/inhalt/2014/1/58/>`__:
+-  `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://www.heise.de/-2356670>`__
+   OCRmyPDF <http://heise.de/-2356670>`__
 
 Disclaimer
 ----------

RELEASE_NOTES.rst

@@ -6,6 +6,62 @@ Please always read this file before installing the package
 Download software here: https://github.com/jbarlow83/OCRmyPDF/tags
 
 
+v4.1:
+=====
+
+-  ``--rotate-pages`` now only rotates pages when reasonably confidence in the orientation. This behavior can be adjusted with the new argument ``--rotate-pages-threshold``
+-  Fixed problems in error checking if ``unpaper`` is uninstalled or missing at run-time
+-  Fixed problems with "RethrownJobError" errors during error handling that suppressed the useful error messages
+
+
+v4.0.7:
+=======
+
+-  Minor correction to Ghostscript output settings
+
+
+v4.0.6:
+=======
+
+-  Update install instructions
+-  Provide a sRGB profile instead of using Ghostscript's
+
+
+v4.0.5:
+=======
+
+Fixes
+-----
+
+-  Remove some verbose debug messages from v4.0.4
+-  Fixed temporary that wasn't being deleted
+-  DPI is now calculated correctly for cropped images, along with other image transformations
+-  Inline images are now checked during DPI calculation instead of rejecting the image
+
+v4.0.4:
+=======
+
+Released with verbose debug message turned on. Do not use. Skip to v4.0.5.
+
+
+v4.0.3:
+=======
+
+New features
+------------
+
+-  Page orientations detected are now reported in a summary comment
+
+
+Fixes
+-----
+
+-  Show stack trace if unexpect errors occur
+-  Treat "too few characters" error message from Tesseract as a reason to skip that page rather than
+   abort the file
+-  Docker: fix blank JPEG2000 issue by insisting on Ghostscript versions that have this fixed
+
+
 v4.0.2:
 =======
 

ocrmypdf/ghostscript.py

@@ -5,6 +5,7 @@ from tempfile import NamedTemporaryFile
 from subprocess import Popen, PIPE, check_call
 from shutil import copy
 from . import get_program
+from .pdfa import SRGB_ICC_PROFILE
 
 
 def rasterize_pdf(input_file, output_file, xres, yres, raster_device, log,
@@ -52,7 +53,6 @@ def generate_pdfa(pdf_pages, output_file, threads=1):
             "-dJPEGQ=95",
             "-dPDFA=2",
             "-sPDFACompatibilityPolicy=2",
-            "-sOutputICCProfile=srgb.icc",
             "-sOutputFile=" + gs_pdf.name,
         ]
         args_gs.extend(pdf_pages)

ocrmypdf/leptonica.py

@@ -110,8 +110,8 @@ class Pix:
             return "<leptonica.Pix image NULL>"
 
     def __getstate__(self):
-        data = ffi.new('l_uint32 *[]', 1)
-        size = ffi.new('size_t *', 0)
+        data = ffi.new('l_uint32 **')
+        size = ffi.new('size_t *')
 
         err = lept.pixSerializeToMemory(self.cpix, data, size)
         if err != 0:
@@ -195,16 +195,40 @@ class Pix:
             else:
                 return (None, None)
 
+    def otsu_adaptive_threshold(
+            self, tile_size=(300, 300), kernel_size=(4, 4), scorefract=0.1):
+        with LeptonicaErrorTrap():
+            sx, sy = tile_size
+            smoothx, smoothy = kernel_size
+            p_cpix = ffi.new('PIX **')
+
+            result = lept.pixOtsuAdaptiveThreshold(
+                self.cpix,
+                sx, sy,
+                smoothx, smoothy,
+                scorefract,
+                ffi.NULL,
+                p_cpix)
+            if result == 0:
+                return Pix(p_cpix[0])
+            else:
+                return None
+
+    @staticmethod
+    @lru_cache(maxsize=1)
+    def make_pixel_sum_tab8():
+        return lept.makePixelSumTab8()
+
     @staticmethod
     def correlation_binary(pix1, pix2):
         if get_leptonica_version() < 'leptonica-1.72':
             # Older versions of Leptonica (pre-1.72) have a buggy
             # implementation of pixCorrelationBinary that overflows on larger
             # images.
-            pix1_count = ffi.new('l_int32 *', 0)
-            pix2_count = ffi.new('l_int32 *', 0)
-            pixn_count = ffi.new('l_int32 *', 0)
-            tab8 = lept.makePixelSumTab8()  # Small memory leak on each call
+            pix1_count = ffi.new('l_int32 *')
+            pix2_count = ffi.new('l_int32 *')
+            pixn_count = ffi.new('l_int32 *')
+            tab8 = Pix.make_pixel_sum_tab8()
 
             lept.pixCountPixels(pix1.cpix, pix1_count, tab8)
             lept.pixCountPixels(pix2.cpix, pix2_count, tab8)

ocrmypdf/lib/compile_leptonica.py

@@ -62,6 +62,16 @@ l_int32 * makePixelSumTab8 ( void );
 PIX * pixDeserializeFromMemory ( const l_uint32 *data, size_t nbytes );
 l_int32 pixSerializeToMemory ( PIX *pixs, l_uint32 **pdata, size_t *pnbytes );
 
+l_int32
+pixOtsuAdaptiveThreshold(PIX       *pixs,
+                         l_int32    sx,
+                         l_int32    sy,
+                         l_int32    smoothx,
+                         l_int32    smoothy,
+                         l_float32  scorefract,
+                         PIX      **ppixth,
+                         PIX      **ppixd);
+
 void lept_free(void *ptr);
 """)
 

ocrmypdf/main.py

@@ -188,6 +188,10 @@ advanced.add_argument(
     '--tesseract-timeout', default=180.0, type=float, metavar='SECONDS',
     help='give up on OCR after the timeout, but copy the preprocessed page '
          'into the final output')
+advanced.add_argument(
+    '--rotate-pages-threshold', default=14.0, type=float, metavar='CONFIDENCE',
+    help="only rotate pages when confidence is above this value (arbitrary "
+         "units reported by tesseract)")
 
 debugging = parser.add_argument_group(
     "Debugging",
@@ -234,13 +238,18 @@ if options.pdf_renderer == 'tesseract' and tesseract.version() < '3.04.01' \
         "Some PDF viewers will fail to find searchable text.\n"
         "--pdf-renderer=tesseract is not recommended.")
 
-if any((options.deskew, options.clean, options.clean_final)):
+if any((options.clean, options.clean_final)):
     try:
         from . import unpaper
-    except ImportError:
+        if unpaper.version() < '6.1':
+            complain(
+                "The installed 'unpaper' is not supported. "
+                "Install version 6.1 or newer.")
+            sys.exit(ExitCode.missing_dependency)
+    except FileNotFoundError:
         complain(
             "Install the 'unpaper' program to use --deskew or --clean.")
-        sys.exit(ExitCode.bad_args)
+        sys.exit(ExitCode.missing_dependency)
 else:
     unpaper = None
 
@@ -378,7 +387,7 @@ def cleanup_working_files(*args):
 @transform(
     input=options.input_file,
     filter=formatter('(?i)\.pdf'),
-    output=work_folder + '{basename[0]}.repaired.pdf',
+    output=os.path.join(work_folder, '{basename[0]}.repaired.pdf'),
     extras=[_log, _pdfinfo, _pdfinfo_lock])
 def repair_pdf(
         input_file,
@@ -529,15 +538,29 @@ def orient_page(
         270: '⇦'
     }
 
+    apply_correction = False
+    description = ''
+    if orient_conf.confidence >= options.rotate_pages_threshold:
+        if orient_conf.angle != 0:
+            apply_correction = True
+            description = ' - will rotate'
+        else:
+            description = ' - rotation appears correct'
+    else:
+        if orient_conf.angle != 0:
+            description = ' - confidence too low to rotate'
+        else:
+            description = ' - no change'
+
     log.info(
         '{0:4d}: page is facing {1}, confidence {2:.2f}{3}'.format(
             page_number(preview),
             direction.get(orient_conf.angle, '?'),
             orient_conf.confidence,
-            ' - correcting rotation' if orient_conf.angle != 0 else '')
+            description)
     )
 
-    if orient_conf.angle == 0:
+    if not apply_correction:
         re_symlink(page_pdf, output_file)
     else:
         writer = pypdf.PdfFileWriter()
@@ -550,6 +573,12 @@ def orient_page(
         with open(output_file, 'wb') as out:
             writer.write(out)
 
+        with pdfinfo_lock:
+            pageno = int(os.path.basename(page_pdf)[0:6]) - 1
+            pageinfo = pdfinfo[pageno].copy()
+            pageinfo['rotated'] = orient_conf.angle
+            pdfinfo[pageno] = pageinfo
+
 
 @transform(
     input=orient_page,
@@ -1004,49 +1033,82 @@ def cleanup_ruffus_error_message(msg):
     return msg
 
 
+def do_ruffus_exception(ruffus_five_tuple):
+    """Replace the elaborate ruffus stack trace with a user friendly
+    description of the error message that occurred."""
+
+    task_name, job_name, exc_name, exc_value, exc_stack = ruffus_five_tuple
+    if exc_name == 'builtins.SystemExit':
+        match = re.search(r"\.(.+?)\)", exc_value)
+        exit_code_name = match.groups()[0]
+        exit_code = getattr(ExitCode, exit_code_name, 'other_error')
+        return exit_code
+    elif exc_name == 'ruffus.ruffus_exceptions.MissingInputFileError':
+        _log.error(cleanup_ruffus_error_message(exc_value))
+        return ExitCode.input_file
+    elif exc_name == 'builtins.TypeError':
+        # Even though repair_pdf will fail, ruffus will still try
+        # to call split_pages with no input files, likely due to a bug
+        if task_name == 'split_pages':
+            _log.error("Input file '{0}' is not a valid PDF".format(
+                options.input_file))
+            return ExitCode.input_file
+    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)
+        return ExitCode.child_process_error
+    elif not options.verbose:
+        _log.error(exc_stack)
+        return ExitCode.other_error
+
+
+def traverse_ruffus_exception(e):
+    """Walk through a RethrownJobError and find the first exception.
+
+    The exit code will be based on this, even if multiple exceptions occurred
+    at the same time."""
+
+    if isinstance(e[0], str) and len(e) == 5:
+        return do_ruffus_exception(e)
+    elif hasattr(e, '__iter__'):
+        for exc in e:
+            return traverse_ruffus_exception(exc)
+
+
 def run_pipeline():
     if not options.jobs:
         options.jobs = available_cpu_count()
     try:
-        options.history_file = os.path.join(work_folder, 'ruffus_history.sqlite')
+        options.history_file = os.path.join(
+            work_folder, 'ruffus_history.sqlite')
         cmdline.run(options)
     except ruffus_exceptions.RethrownJobError as e:
         if options.verbose:
-            _log.debug(e)
+            _log.debug(str(e))  # stringify exception so logger doesn't have to
 
-        # Yuck. Hunt through the ruffus exception to find out what the
-        # return code is supposed to be.
-        # Ruffus flattens the exception to a string, throwing away all kinds
-        # of helpful details
-        # task_name, job_name - ruffus status
-        # exc_name - class name of exception
-        # exc_value - irritating string that makes impossible to recover
-        #   exception object
-        # exc_stack - string that contains traceback of exception
-        for exc in e.args:
-            task_name, job_name, exc_name, exc_value, exc_stack = exc
-            if exc_name == 'builtins.SystemExit':
-                match = re.search(r"\.(.+?)\)", exc_value)
-                exit_code_name = match.groups()[0]
-                exit_code = getattr(ExitCode, exit_code_name, 'other_error')
-                return exit_code
-            elif exc_name == 'ruffus.ruffus_exceptions.MissingInputFileError':
-                _log.error(cleanup_ruffus_error_message(exc_value))
-                return ExitCode.input_file
-            elif exc_name == 'builtins.TypeError':
-                # Even though repair_pdf will fail, ruffus will still try
-                # to call split_pages with no input files, likely due to a bug
-                if task_name == 'split_pages':
-                    _log.error("Input file '{0}' is not a valid PDF".format(
-                        options.input_file))
-                    return ExitCode.input_file
-            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)
-                return ExitCode.child_process_error
+        # Ruffus flattens exception to 5 element tuples. Because of a bug
+        # in <= 2.6.3 it may present either the single:
+        #   (task, job, exc, value, stack)
+        # or something like:
+        #   [[(task, job, exc, value, stack)]]
+        #
+        # Generally cross-process exception marshalling doesn't work well
+        # and ruffus doesn't support because BaseException has its own
+        # implementation of __reduce__ that attempts to reconstruct the
+        # exception based on e.__init__(e.args).
+        #
+        # Attempting to log the exception directly marshalls it to the logger
+        # which is probably in another process, so it's better to log only
+        # data from the exception at this point.
 
-        return ExitCode.other_error
+        exitcode = traverse_ruffus_exception(e.args)
+        if exitcode is None:
+            _log.error("Unexpected ruffus exception: " + str(e))
+            _log.error(repr(e))
+            return ExitCode.other_error
+        else:
+            return exitcode
     except Exception as e:
         _log.error(e)
         return ExitCode.other_error
@@ -1055,6 +1117,20 @@ def run_pipeline():
         _log.warning('Output file: The generated PDF/A file is INVALID')
         return ExitCode.invalid_output_pdfa
 
+    with _pdfinfo_lock:
+        _log.debug(_pdfinfo)
+        direction = {0: 'n', 90: 'e',
+                     180: 's', 270: 'w'}
+        orientations = []
+        for n, page in enumerate(_pdfinfo):
+            angle = _pdfinfo[n].get('rotated', 0)
+            if angle != 0:
+                orientations.append('{0}{1}'.format(
+                    n + 1,
+                    direction.get(angle, '')))
+        if orientations:
+            _log.info('Page orientations detected: ' + ' '.join(orientations))
+
     return ExitCode.ok
 
 

ocrmypdf/pageinfo.py

@@ -3,10 +3,13 @@
 
 from subprocess import Popen, PIPE
 from decimal import Decimal, getcontext
+from math import hypot
 import re
 import sys
 import PyPDF2 as pypdf
+from collections import namedtuple
 
+matrix_mult = pypdf.pdf.utils.matrixMultiply
 
 FRIENDLY_COLORSPACE = {
     '/DeviceGray': 'gray',
@@ -19,7 +22,11 @@ FRIENDLY_COLORSPACE = {
     '/Indexed': 'index',
     '/Separation': 'sep',
     '/DeviceN': 'devn',
-    '/Pattern': '-'
+    '/Pattern': '-',
+    '/G': 'gray',  # Abbreviations permitted in inline images
+    '/RGB': 'rgb',
+    '/CMYK': 'cmyk',
+    '/I': 'index',
 }
 
 FRIENDLY_ENCODING = {
@@ -27,6 +34,13 @@ FRIENDLY_ENCODING = {
     '/DCTDecode': 'jpeg',
     '/JPXDecode': 'jpx',
     '/JBIG2Decode': 'jbig2',
+    '/CCF': 'ccitt',  # Abbreviations permitted in inline images
+    '/DCT': 'jpeg',
+    '/AHx': 'asciihex',
+    '/A85': 'ascii85',
+    '/LZW': 'lzw',
+    '/Fl': 'flate',
+    '/RL': 'runlength'
 }
 
 FRIENDLY_COMP = {
@@ -38,47 +52,165 @@ FRIENDLY_COMP = {
 }
 
 
-def _page_has_inline_images(page):
-    # PDF always uses \r\n for separator regardless of platform
-    # Really basic heuristic that might trigger the odd false positive
-    # This is only finds the first image and is not quite spec compliant
-    try:
-        contents = page.getContents()
-        data = contents.getData()
-    except AttributeError:
-        # If we can't access the contents or data (empty page?) then there
-        # are no inline images
-        return False
+def _matrix_from_shorthand(shorthand):
+    """Convert from PDF matrix shorthand to full matrix
 
-    begin_image, image_data, end_image = False, False, False
-    for data in re.split(b'\s+', data):
-        if data == b'BI':
-            begin_image = True
-        elif data == b'ID':
-            image_data = True
-        elif data == b'EI':
-            end_image = True
-        if all((begin_image, image_data, end_image)):
-            return True
-    return False
+    PDF 1.7 spec defines a shorthand for describing the entries of a matrix
+    since the last column is always (0, 0, 1).
+    """
+
+    a, b, c, d, e, f = map(float, shorthand)
+    return ((a, b, 0),
+            (c, d, 0),
+            (e, f, 1))
 
 
-def _find_page_images(page, pageinfo):
+def _shorthand_from_matrix(matrix):
+    """Convert from transformation matrix to PDF shorthand."""
+    a, b = matrix[0][0], matrix[0][1]
+    c, d = matrix[1][0], matrix[1][1]
+    e, f = matrix[2][0], matrix[2][1]
+    return tuple(map(float, (a, b, c, d, e, f)))
+
+
+ContentsInfo = namedtuple('ContentsInfo', ['raster_settings', 'inline_images'])
+
+
+def _interpret_contents(contentstream):
+    operations = contentstream.operations
+    stack = []
+    ctm = _matrix_from_shorthand((1, 0, 0, 1, 0, 0))
+    image_raster_settings = []
+    inline_images = []
+
+    for op in operations:
+        operands, command = op
+        if command == b'q':
+            stack.append(ctm)
+            if len(stack) > 32:
+                raise RuntimeError("PDF graphics stack overflow")
+        elif command == b'Q':
+            ctm = stack.pop()
+        elif command == b'cm':
+            ctm = matrix_mult(
+                ctm, _matrix_from_shorthand(operands))
+        elif command == b'Do':
+            image_name = operands[0]
+            image_raster_settings.append(
+                (image_name, _shorthand_from_matrix(ctm)))
+        elif command == b'INLINE IMAGE':
+            settings = operands['settings']
+            inline_images.append(
+                (settings, _shorthand_from_matrix(ctm)))
+
+    return ContentsInfo(
+        raster_settings=image_raster_settings,
+        inline_images=inline_images)
+
+
+def _get_dpi(ctm_shorthand, image_size):
+    """Given the transformation matrix and image size, find the image DPI.
+
+    PDFs do not include image resolution information within image data.
+    Instead, the PDF page content stream describes the location where the
+    image will be rasterized, and the effective resolution is the ratio of the
+    pixel size to raster target size.
+
+    Normally a scanned PDF has the paper size set appropriately but this is
+    not guaranteed. The most common case is a cropped image will change the
+    page size (/CropBox) without altering the page content stream. That means
+    it is not sufficient to assume that the image fills the page, even though
+    that is the most common case.
+
+    A PDF image may be scaled (always), cropped, translated, rotated in place
+    to an arbitrary angle (rarely) and skewed. Only equal area mappings can
+    be expressed, that is, it is not necessary to consider distortions where
+    the effective DPI varies with position.
+
+    To determine the image scale, transform an offset axis vector v0 (0, 0),
+    width-axis vector v0 (1, 0), height-axis vector vh (0, 1) with the matrix,
+    which gives the dimensions of the image in PDF units. From there we can
+    compare to actual image dimensions. PDF uses
+    row vector * matrix_tranposed unlike the traditional
+    matrix * column vector.
+
+    The offset, width and height vectors can be combined in a matrix and
+    multiplied by the transform matrix. Then we want to calculated
+        magnitude(width_vector - offset_vector)
+    and
+        magnitude(height_vector - offset_vector)
+
+    When the above is worked out algebraically, the effect of translation
+    cancels out, and the vector magnitudes become functions of the nonzero
+    transformation matrix indices. The results of the derivation are used
+    in this code.
+
+    pdfimages -list does calculate the DPI in some way that is not completely
+    naive, but it does not get the DPI of rotated images right, so cannot be
+    used anymore to validate this. Photoshop works, or using Acrobat to
+    rotate the image back to normal.
+
+    It does not matter if the image is partially cropped, or even out of the
+    /MediaBox.
+
+    """
+
+    a, b, c, d, _, _ = ctm_shorthand
+
+    # Calculate the width and height of the image in PDF units
+    image_drawn_width = hypot(a, b)
+    image_drawn_height = hypot(c, d)
+
+    # The scale of the image is pixels per PDF unit (1/72")
+    scale_w = image_size[0] / image_drawn_width
+    scale_h = image_size[1] / image_drawn_height
+
+    # DPI = scale * 72
+    dpi_w = scale_w * 72.0
+    dpi_h = scale_h * 72.0
+
+    return (dpi_w, dpi_h)
+
+
+def _find_page_inline_images(page, pageinfo, contentsinfo):
+    "Find inline images on the page"
+
+    for n, im in enumerate(contentsinfo.inline_images):
+        settings, shorthand = im
+        image = {}
+        image['name'] = str('inline-%02d' % n)
+        image['width'] = settings['/W']
+        image['height'] = settings['/H']
+        image['bpc'] = settings['/BPC']
+        image['color'] = FRIENDLY_COLORSPACE.get(settings['/CS'], '-')
+        image['comp'] = FRIENDLY_COMP.get(image['color'], '?')
+        if '/F' in settings:
+            filter_ = settings['/F']
+            if isinstance(filter_, pypdf.generic.ArrayObject):
+                filter_ = filter_[0]
+            image['enc'] = FRIENDLY_ENCODING.get(filter_, 'image')
+        else:
+            image['enc'] = 'image'
+
+        dpi_w, dpi_h = _get_dpi(shorthand, (image['width'], image['height']))
+        image['dpi_w'], image['dpi_h'] = Decimal(dpi_w), Decimal(dpi_h)
+        yield image
+
+
+def _find_page_regular_images(page, pageinfo, contentsinfo):
+    "Find images stored in XObject resources"
+
     try:
         page['/Resources']['/XObject']
     except KeyError:
         return
-
-    # Look for XObject (out of line images)
     for xobj in page['/Resources']['/XObject']:
         # PyPDF2 returns the keys as an iterator
         pdfimage = page['/Resources']['/XObject'][xobj]
         if pdfimage['/Subtype'] != '/Image':
             continue
-        if '/ImageMask' in pdfimage:
-            if pdfimage['/ImageMask']:
-                continue
         image = {}
+        image['name'] = str(xobj)
         image['width'] = pdfimage['/Width']
         image['height'] = pdfimage['/Height']
         image['bpc'] = pdfimage['/BitsPerComponent']
@@ -98,12 +230,33 @@ def _find_page_images(page, pageinfo):
             image['color'] = 'jpx' if image['enc'] == 'jpx' else '?'
 
         image['comp'] = FRIENDLY_COMP.get(image['color'], '?')
-        image['dpi_w'] = image['width'] / pageinfo['width_inches']
-        image['dpi_h'] = image['height'] / pageinfo['height_inches']
+        image['dpi_w'] = image['dpi_h'] = 0
+
+        for raster in contentsinfo.raster_settings:
+            # Loop in case the same image is display multiple times on a page
+            if raster[0] != image['name']:
+                continue
+            shorthand = raster[1]
+
+            dpi_w, dpi_h = _get_dpi(
+                shorthand, (image['width'], image['height']))
+
+            # When image is used multiple times take the highest DPI it is
+            # rendered at
+            image['dpi_w'] = max(dpi_w, image.get('dpi_w', 0))
+            image['dpi_h'] = max(dpi_h, image.get('dpi_h', 0))
+
+        image['dpi_w'] = Decimal(image['dpi_w'])
+        image['dpi_h'] = Decimal(image['dpi_h'])
         image['dpi'] = (image['dpi_w'] * image['dpi_h']) ** Decimal(0.5)
         yield image
 
 
+def _find_page_images(page, pageinfo, contentsinfo):
+    yield from _find_page_inline_images(page, pageinfo, contentsinfo)
+    yield from _find_page_regular_images(page, pageinfo, contentsinfo)
+
+
 def _page_has_text(pdf, page):
     # Simple test
     text = page.extractText()
@@ -141,12 +294,14 @@ def _pdf_get_pageinfo(infile, pageno: int):
     pageinfo['width_inches'] = width_pt / Decimal(72.0)
     pageinfo['height_inches'] = height_pt / Decimal(72.0)
 
-    pageinfo['images'] = [im for im in _find_page_images(page, pageinfo)]
+    try:
+        contentstream = pypdf.pdf.ContentStream(page.getContents(), pdf)
+    except AttributeError as e:
+        return pageinfo
 
-    # Look for inline images
-    if _page_has_inline_images(page):
-        raise NotImplementedError(
-            "Warning: input PDF contains inline images - not supported")
+    contentsinfo = _interpret_contents(contentstream)
+    pageinfo['images'] = [im for im in _find_page_images(
+                                page, pageinfo, contentsinfo)]
 
     if pageinfo['images']:
         xres = max(image['dpi_w'] for image in pageinfo['images'])
@@ -164,3 +319,17 @@ def pdf_get_all_pageinfo(infile):
     pdf = pypdf.PdfFileReader(infile)
     getcontext().prec = 6
     return [_pdf_get_pageinfo(infile, n) for n in range(pdf.numPages)]
+
+
+def main():
+    import argparse
+    parser = argparse.ArgumentParser()
+    parser.add_argument('infile')
+    args = parser.parse_args()
+    info = pdf_get_all_pageinfo(args.infile)
+    from pprint import pprint
+    pprint(info)
+
+
+if __name__ == '__main__':
+    main()

ocrmypdf/pdfa.py

@@ -5,10 +5,13 @@
 
 from __future__ import print_function, absolute_import, division
 from string import Template
-from subprocess import Popen, PIPE
-import os
 import codecs
-from . import get_program
+import pkg_resources
+
+ICC_PROFILE_RELPATH = 'data/sRGB_IEC61966-2-1_black_scaled.icc'
+
+SRGB_ICC_PROFILE = pkg_resources.resource_filename(
+    'ocrmypdf', ICC_PROFILE_RELPATH)
 
 
 # This is a template written in PostScript which is needed to create PDF/A
@@ -93,38 +96,9 @@ def _get_pdfa_def(icc_profile, icc_identifier, pdfmark):
     return result
 
 
-def _get_postscript_icc_path():
-    "Parse Ghostscript's help message to find where iccprofiles are stored"
-
-    p_gs = Popen([get_program('gs'), '--help'], close_fds=True,
-                 universal_newlines=True,
-                 stdout=PIPE, stderr=PIPE)
-    out, _ = p_gs.communicate()
-    lines = out.splitlines()
-
-    def search_paths(lines):
-        seeking = True
-        for line in lines:
-            if seeking:
-                if line.startswith('Search path'):
-                    seeking = False
-                    continue
-            else:
-                if line.strip().startswith('/'):
-                    yield from (
-                        path.strip() for path in line.split(':')
-                        if path.strip() != '')
-    for root in search_paths(lines):
-        path = os.path.realpath(os.path.join(root, '../iccprofiles'))
-        if os.path.exists(path):
-            return path
-
-    raise FileNotFoundError("Could not find Ghostscript's iccprofiles")
-
-
 def generate_pdfa_def(target_filename, pdfmark, icc='sRGB'):
     if icc == 'sRGB':
-        icc_profile = os.path.join(_get_postscript_icc_path(), 'srgb.icc')
+        icc_profile = SRGB_ICC_PROFILE
     else:
         raise NotImplementedError("Only supporting sRGB")
 

ocrmypdf/tesseract.py

@@ -98,7 +98,8 @@ def get_orientation(input_file, language: list, timeout: float, log):
         return OrientationConfidence(angle=0, confidence=0.0)
     except CalledProcessError as e:
         tesseract_log_output(log, e.output, input_file)
-        if 'Image too large' in e.output:
+        if ('Too few characters. Skipping this page' in e.output or
+                'Image too large' in e.output):
             return OrientationConfidence(0, 0)
         raise e from e
     else:

ocrmypdf/unpaper.py

@@ -3,7 +3,7 @@
 # unpaper documentation:
 # https://github.com/Flameeyes/unpaper/blob/master/doc/basic-concepts.md
 
-from subprocess import Popen, PIPE
+from subprocess import CalledProcessError, STDOUT, check_output, check_call
 from tempfile import NamedTemporaryFile
 import sys
 import os
@@ -17,10 +17,9 @@ def version():
         get_program('unpaper'),
         '--version'
     ]
-    p_unpaper = Popen(args_unpaper, close_fds=True, universal_newlines=True,
-                      stdout=PIPE, stderr=PIPE)
-    version, _ = p_unpaper.communicate(timeout=5)
-
+    version = check_output(
+        args_unpaper, close_fds=True, universal_newlines=True,
+        stderr=STDOUT, timeout=5)
     return version.strip()
 
 
@@ -68,15 +67,17 @@ def run(input_file, output_file, dpi, log, mode_args):
         os.unlink(output_pnm.name)
 
         args_unpaper.extend([input_pnm.name, output_pnm.name])
-        p_unpaper = Popen(
-            args_unpaper, close_fds=True,
-            universal_newlines=True, stdout=PIPE, stderr=PIPE
-            )
-        out, err = p_unpaper.communicate()
-        log.debug(out)
-        log.debug(err)
-
-        Image.open(output_pnm.name).save(output_file)
+        try:
+            stdout = check_output(
+                args_unpaper, close_fds=True,
+                universal_newlines=True, stderr=STDOUT,
+                )
+        except CalledProcessError as e:
+            log.debug(e.output)
+            raise e from e
+        else:
+            log.debug(stdout)
+            Image.open(output_pnm.name).save(output_file)
 
 
 def deskew(input_file, output_file, dpi, log):

setup.py

@@ -209,7 +209,6 @@ setup(
         ],
     setup_requires=[
         'setuptools_scm',
-        'setuptools_scm_git_archive',
         'cffi>=1.5.0',
         'pytest-runner'
     ],
@@ -218,7 +217,7 @@ setup(
         'ocrmypdf/lib/compile_leptonica.py:ffi'
     ],
     install_requires=[
-        'ruffus>=2.6.3',
+        'ruffus==2.6.3',
         'Pillow>=3.0.0',
         'reportlab>=3.1.44',
         'PyPDF2>=1.25.1',
@@ -231,5 +230,6 @@ setup(
             'ocrmypdf = ocrmypdf.main:run_pipeline'
         ],
     },
+    package_data={'ocrmypdf': ['data/sRGB_IEC61966-2-1_black_scaled.icc']},
     include_package_data=True,
     zip_safe=False)

share/etc-apt-sources.list

@@ -0,0 +1,4 @@
+deb http://httpredir.debian.org/debian stretch main
+deb http://httpredir.debian.org/debian stretch-updates main
+deb http://security.debian.org stretch/updates main
+deb http://ftp.de.debian.org/debian sid main contrib non-free

tests/spoof/unpaper_oldversion.py

@@ -0,0 +1,14 @@
+#!/usr/bin/env python3
+import sys
+
+def main():
+    if sys.argv[1] == '--version':
+        print('0.5')
+        sys.exit(0)
+
+    print("Only supports --version")
+    sys.exit(1)
+
+
+if __name__ == '__main__':
+    main()

tests/test_main.py

@@ -325,11 +325,6 @@ def test_monochrome_correlation():
     'tesseract',
     ])
 def test_autorotate(spoof_tesseract_cache, renderer):
-    import ocrmypdf.ghostscript as ghostscript
-    import logging
-
-    gslog = logging.getLogger()
-
     # cardinal.pdf contains four copies of an image rotated in each cardinal
     # direction - these ones are "burned in" not tagged with /Rotate
     out = check_ocrmypdf('cardinal.pdf', 'test_autorotate_%s.pdf' % renderer,
@@ -343,6 +338,36 @@ def test_autorotate(spoof_tesseract_cache, renderer):
         assert correlation > 0.80
 
 
+def test_autorotate_threshold_low(spoof_tesseract_cache):
+    out = check_ocrmypdf('cardinal.pdf', 'test_autorotate_threshold_low.pdf',
+                         '--rotate-pages-threshold', '1',
+                         '-r', '-v', '1', env=spoof_tesseract_cache)
+
+    # Low threshold -> always rotate -> expect high correlation between
+    # reference page and test page
+    correlation = check_monochrome_correlation(
+        reference_pdf=_infile('cardinal.pdf'),
+        reference_pageno=1,
+        test_pdf=out,
+        test_pageno=3)
+    assert correlation > 0.80
+
+
+def test_autorotate_threshold_high(spoof_tesseract_cache):
+    out = check_ocrmypdf('cardinal.pdf', 'test_autorotate_threshold_high.pdf',
+                         '--rotate-pages-threshold', '99',
+                         '-r', '-v', '1', env=spoof_tesseract_cache)
+
+    # High threshold -> never rotate -> expect low correlation since
+    # test page will not be rotated
+    correlation = check_monochrome_correlation(
+        reference_pdf=_infile('cardinal.pdf'),
+        reference_pageno=1,
+        test_pdf=out,
+        test_pageno=3)
+    assert correlation < 0.10
+
+
 @pytest.mark.parametrize('renderer', [
     'hocr',
     'tesseract',
@@ -493,6 +518,8 @@ def test_tesseract_crash_autorotate(spoof_tesseract_crash):
     assert sh.returncode == ExitCode.child_process_error
     assert not os.path.exists(_outfile('wontwork.pdf'))
     assert "ERROR" in err
+    print(out)
+    print(err)
 
 
 @pytest.mark.parametrize('renderer', [
@@ -503,3 +530,20 @@ def test_tesseract_image_too_big(renderer, spoof_tesseract_big_image_error):
     check_ocrmypdf(
         'hugemono.pdf', 'hugemono_%s.pdf' % renderer, '-r',
         '--pdf-renderer', renderer, env=spoof_tesseract_big_image_error)
+
+
+def test_no_unpaper():
+    env = os.environ.copy()
+    env['OCRMYPDF_UNPAPER'] = os.path.abspath('./spoof/no_unpaper_here.py')
+    sh, out, err = run_ocrmypdf_env(
+        'c02-22.pdf', 'wont_be_created.pdf', '--clean', env=env)
+    assert sh.returncode == ExitCode.missing_dependency
+
+
+def test_old_unpaper():
+    env = os.environ.copy()
+    env['OCRMYPDF_UNPAPER'] = os.path.abspath('./spoof/unpaper_oldversion.py')
+    sh, out, err = run_ocrmypdf_env(
+        'c02-22.pdf', 'wont_be_created.pdf', '--clean', env=env)
+    assert sh.returncode == ExitCode.missing_dependency
+

tests/test_pageinfo.py

@@ -103,8 +103,8 @@ def test_single_page_image():
     assert pdfimage['bpc'] == 8
 
     # DPI in a 1"x1" is the image width
-    assert pdfimage['dpi_w'] == 8
-    assert pdfimage['dpi_h'] == 8
+    assert abs(pdfimage['dpi_w'] - 8) < 1e-5
+    assert abs(pdfimage['dpi_h'] - 8) < 1e-5
 
 
 def test_single_page_inline_image():
@@ -120,8 +120,12 @@ def test_single_page_inline_image():
         pdf.showPage()
         pdf.save()
 
-    with pytest.raises(NotImplementedError):
-        pageinfo.pdf_get_all_pageinfo(filename)
+    pdfinfo = pageinfo.pdf_get_all_pageinfo(filename)
+    print(pdfinfo)
+    pdfimage = pdfinfo[0]['images'][0]
+    assert (pdfimage['dpi_w'] - 8) < 1e-5
+    assert pdfimage['color'] != '-'
+    assert pdfimage['width'] == 8
 
 
 def test_jpeg():
@@ -131,4 +135,5 @@ def test_jpeg():
 
     pdfimage = pdfinfo[0]['images'][0]
     assert pdfimage['enc'] == 'jpeg'
+    assert (pdfimage['dpi_w'] - 150) < 1e-5