v11.2.0 release notes

Change v11.1.3 to v11.2.0 since it contains functional changes.

azure-pipelines.yml

@@ -149,12 +149,6 @@ stages:
       - job: "macOS_Mojave"
         pool:
           vmImage: "macos-10.14"
-        strategy:
-          matrix:
-            Python37:
-              python.version: ""
-            Python38:
-              python.version: "python@3.8"
         steps:
           # https://github.com/actions/virtual-environments/issues/664
           # - task: UsePythonVersion@0
@@ -163,11 +157,8 @@ stages:
           - bash: |
               brew update
               brew unlink python@2
-              if [ "$(python.version)" != "" ]; then
-                brew upgrade $(python.version)
-              else
-                echo "Using Python `python3 --version`"
-              fi
+              brew upgrade python
+              echo "Using Python `python3 --version`"
             displayName: "Update brew and Python"
           - bash: |
               brew install \
@@ -177,8 +168,7 @@ stages:
                 leptonica \
                 openjpeg \
                 pngquant \
-                tesseract \
-                unpaper
+                tesseract
             displayName: "Install system packages"
           - bash: |
               pip3 install --upgrade pip

docs/installation.rst

@@ -190,7 +190,8 @@ of ocrmypdf, and install the following dependencies:
         python3-reportlab \
         qpdf \
         tesseract-ocr \
-        zlib1g
+        zlib1g \
+        unpaper
 
 We will need a newer version of ``pip`` then was available for Ubuntu 18.04:
 

docs/release_notes.rst

@@ -12,6 +12,45 @@ may be unreliable. Use the API to depend on precise behavior.
 The public API may be useful in scripts that launch OCRmyPDF processes or that
 wish to use some of its features for working with PDFs.
 
+v11.2.0
+=======
+
+-  Fixed an issue with optimizing PNG-type images that had soft masks or image masks.
+   This is a regression introduced in (or about) v11.1.0.
+-  Improved type checking of the ``plugins`` parameter for the ``ocrmypdf.ocr``
+   API call.
+
+v11.1.2
+=======
+
+-  Fixed hOCR renderer writing the text in roughly reverse order. This should not
+   affect reasonably smart PDF readers that properly locate the position of all
+   text, but may confuse those that rely on the order of objects in the content
+   stream. (#642)
+
+v11.1.1
+=======
+
+-  We now avoid using named temporary files when using pngquant allowing containerized
+   pngquant installs to be used.
+-  Clarified an error message.
+-  Highest number of 1's in a release ever!
+
+v11.1.0
+=======
+
+-  Fixed page rotation issues: #634, #589.
+-  Fixed some cases where optimization created an invalid image such as a
+   1-bit "RGB" image: #629, #620.
+-  Page numbers are now displayed in debug logs when pages are being grafted.
+-  ocrmypdf.optimize.rewrite_png and ocrmypdf.optimize.rewrite_png_as_g4 were
+   marked deprecated. Strictly speaking these should have been internal APIs,
+   but they were never hidden.
+-  As a precaution, pikepdf mmap-based file access has been disabled due to a
+   rare race condition that causes a crash when certain objects are deallocated.
+   The problem is likely in pikepdf's dependency pybind11.
+-  Extended the example plugin to demonstrate conversion to mono.
+
 v11.0.2
 =======
 

misc/example_plugin.py

@@ -18,6 +18,25 @@
 # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 # SOFTWARE.
 
+"""
+An example of an OCRmyPDF plugin.
+
+This plugin adds two new command line arguments
+    --grayscale-ocr: converts the image to grayscale before performing OCR on it
+        (This is occasionally useful for images whose color confounds OCR. It only
+        affects the image shown to OCR. The image is not saved.)
+    --mono-page: converts pages all pages in the output file to black and white
+
+To use this from the command line:
+    ocrmypdf --plugin path/to/example_plugin.py --mono-page input.pdf output.pdf
+
+To use this as an API:
+    import ocrmypdf
+    ocrmypdf.ocr('input.pdf', 'output.pdf',
+        plugins=['path/to/example_plugin.py'], mono_page=True
+    )
+"""
+
 import logging
 
 from PIL import Image
@@ -30,6 +49,7 @@ log = logging.getLogger(__name__)
 @hookimpl
 def add_options(parser):
     parser.add_argument('--grayscale-ocr', action='store_true')
+    parser.add_argument('--mono-page', action='store_true')
 
 
 @hookimpl
@@ -52,7 +72,13 @@ def filter_ocr_image(page, image):
 
 @hookimpl
 def filter_page_image(page, image_filename):
-    output = image_filename.with_suffix('.jpg')
-    with Image.open(image_filename) as im:
-        im.save(output)
-    return output
+    if page.options.mono_page:
+        with Image.open(image_filename) as im:
+            im = im.convert('1')
+            im.save(image_filename)
+        return image_filename
+    else:
+        output = image_filename.with_suffix('.jpg')
+        with Image.open(image_filename) as im:
+            im.save(output)
+        return output

src/ocrmypdf/_exec/pngquant.py

@@ -7,7 +7,11 @@
 
 """Interface to pngquant executable"""
 
+from contextlib import contextmanager
+from io import BytesIO
 from os import fspath
+from pathlib import Path
+from subprocess import PIPE
 from tempfile import NamedTemporaryFile
 
 from PIL import Image
@@ -28,34 +32,32 @@ def available():
     return True
 
 
-def quantize(input_file, output_file, quality_min, quality_max):
-    input_file = fspath(input_file)
-    output_file = fspath(output_file)
-    if input_file.endswith('.jpg'):
-        with Image.open(input_file) as im, NamedTemporaryFile(suffix='.png') as tmp:
-            im.save(tmp)
-            args = [
-                'pngquant',
-                '--force',
-                '--skip-if-larger',
-                '--output',
-                output_file,
-                '--quality',
-                f'{quality_min}-{quality_max}',
-                '--',
-                tmp.name,
-            ]
-            run(args)
+@contextmanager
+def input_as_png(input_file: Path):
+    if not input_file.name.endswith('.png'):
+        with Image.open(input_file) as im:
+            bio = BytesIO()
+            im.save(bio, format='png')
+            bio.seek(0)
+            yield bio
     else:
+        with open(input_file, 'rb') as f:
+            yield f
+
+
+def quantize(input_file: Path, output_file: Path, quality_min: int, quality_max: int):
+    with input_as_png(input_file) as input_stream:
         args = [
             'pngquant',
             '--force',
             '--skip-if-larger',
-            '--output',
-            output_file,
             '--quality',
             f'{quality_min}-{quality_max}',
-            '--',
-            input_file,
+            '--',  # pngquant: stop processing arguments
+            '-',  # pngquant: stream input and output
         ]
-        run(args)
+        result = run(args, stdin=input_stream, stdout=PIPE, stderr=PIPE, check=False)
+
+    if result.returncode == 0:
+        # input_file could be the same as output_file, so we defer the write
+        output_file.write_bytes(result.stdout)

src/ocrmypdf/_exec/unpaper.py

@@ -107,7 +107,7 @@ def run(input_file, output_file, dpi, mode_args):
 
 def validate_custom_args(args: str):
     unpaper_args = shlex.split(args)
-    if any('/' in arg for arg in unpaper_args):
+    if any(('/' in arg or arg == '.' or arg == '..') for arg in unpaper_args):
         raise ValueError('No filenames allowed in --unpaper-args')
     return unpaper_args
 

src/ocrmypdf/_graft.py

@@ -109,7 +109,6 @@ class OcrGrafter:
         if textpdf and not self.font:
             self.font, self.font_key = self._find_font(textpdf)
 
-        emplaced_page = False
         content_rotation = self.pdfinfo[pageno].rotation
         path_image = Path(image).resolve() if image else None
         if path_image is not None and path_image != self.path_base:
@@ -123,17 +122,21 @@ class OcrGrafter:
                 local_image_page = self.pdf_base.pages[-1]
                 self.pdf_base.pages[pageno].emplace(local_image_page)
                 del self.pdf_base.pages[-1]
-            emplaced_page = True
+            # The pdf_image_page will always be created with any /Rotate applied
+            # applied already
+            content_rotation = 0
 
-        if emplaced_page:
-            content_rotation = autorotate_correction
-        text_rotation = autorotate_correction
-        text_misaligned = (text_rotation - content_rotation) % 360
-        log.debug(
-            f"Rotations for page {pageno}: [text, auto, misalign, content] = "
-            f"{text_rotation}, {autorotate_correction}, "
-            f"{text_misaligned}, {content_rotation}"
-        )
+        if content_rotation != 0:
+            # Text can be misaligned on a /Rotate'd page.
+            # That is because we rasterize pages with /Rotate applied,
+            # so that the OCR image text is upright and comes back upright.
+            text_misaligned = (autorotate_correction - content_rotation) % 360
+            log.debug(
+                f"Text rotation: (autorotate, content) -> text misalignment = "
+                f"({autorotate_correction}, {content_rotation}) -> {text_misaligned}"
+            )
+        else:
+            text_misaligned = 0
 
         if textpdf and self.font:
             # Graft the text layer onto this page, whether new or old
@@ -143,15 +146,18 @@ class OcrGrafter:
                 textpdf=textpdf,
                 font=self.font,
                 font_key=self.font_key,
-                rotation=text_misaligned,
+                text_rotation=text_misaligned,
                 procset=self.procset,
                 strip_old_text=strip_old,
             )
 
-        # Correct the rotation if applicable
-        self.pdf_base.pages[pageno].Rotate = (
-            content_rotation - autorotate_correction
-        ) % 360
+        # Correct the page rotation
+        page_rotation = (content_rotation - autorotate_correction) % 360
+        self.pdf_base.pages[pageno].Rotate = page_rotation
+        log.debug(
+            f"Page rotation: (content, auto) -> page = "
+            f"({content_rotation}, {autorotate_correction}) -> {page_rotation}"
+        )
 
         if self.emplacements % MAX_REPLACE_PAGES == 0:
             self.save_and_reload()
@@ -226,7 +232,7 @@ class OcrGrafter:
         font: pikepdf.Object,
         font_key: pikepdf.Object,
         procset: pikepdf.Object,
-        rotation: int,
+        text_rotation: int,
         strip_old_text: bool,
     ):
         """Insert the text layer from text page 0 on to pdf_base at page_num"""
@@ -256,13 +262,13 @@ class OcrGrafter:
             corner = pikepdf.PdfMatrix().translated(mediabox[0], mediabox[1])
             # -rotation because the input is a clockwise angle and this formula
             # uses CCW
-            rotation = -rotation % 360
-            rotate = pikepdf.PdfMatrix().rotated(rotation)
+            text_rotation = -text_rotation % 360
+            rotate = pikepdf.PdfMatrix().rotated(text_rotation)
 
             # Because of rounding of DPI, we might get a text layer that is not
             # identically sized to the target page. Scale to adjust. Normally this
             # is within 0.998.
-            if rotation in (90, 270):
+            if text_rotation in (90, 270):
                 wt, ht = ht, wt
             scale_x = wp / wt
             scale_y = hp / ht

src/ocrmypdf/_sync.py

@@ -252,15 +252,19 @@ def exec_concurrent(context: PdfContext):
     ocrgraft = OcrGrafter(context)
 
     def update_page(result: PageResult, pbar):
-        sidecars[result.pageno] = result.text
-        pbar.update()
-        ocrgraft.graft_page(
-            pageno=result.pageno,
-            image=result.pdf_page_from_image,
-            textpdf=result.ocr,
-            autorotate_correction=result.orientation_correction,
-        )
-        pbar.update()
+        try:
+            tls.pageno = result.pageno + 1
+            sidecars[result.pageno] = result.text
+            pbar.update()
+            ocrgraft.graft_page(
+                pageno=result.pageno,
+                image=result.pdf_page_from_image,
+                textpdf=result.ocr,
+                autorotate_correction=result.orientation_correction,
+            )
+            pbar.update()
+        finally:
+            tls.pageno = None
 
     exec_progress_pool(
         use_threads=context.options.use_threads,

src/ocrmypdf/_validation.py

@@ -308,17 +308,6 @@ def check_closed_streams(options):  # pragma: no cover
     return True
 
 
-def log_page_orientations(pdfinfo):
-    direction = {0: 'n', 90: 'e', 180: 's', 270: 'w'}
-    orientations = []
-    for n, page in enumerate(pdfinfo):
-        angle = page.rotation or 0
-        if angle != 0:
-            orientations.append('{0}{1}'.format(n + 1, direction.get(angle, '')))
-    if orientations:
-        log.info('Page orientations detected: %s', ' '.join(orientations))
-
-
 def create_input_file(options, work_folder: Path) -> Tuple[Path, str]:
     if options.input_file == '-':
         # stdin

src/ocrmypdf/api.py

@@ -226,7 +226,7 @@ def ocr(  # pylint: disable=unused-argument
     user_words: os.PathLike = None,
     user_patterns: os.PathLike = None,
     fast_web_view: float = None,
-    plugins: Iterable[str] = None,
+    plugins: Iterable[Union[str, Path]] = None,
     keep_temporary_files: bool = None,
     progress_bar: bool = None,
     **kwargs,
@@ -280,6 +280,8 @@ def ocr(  # pylint: disable=unused-argument
     """
     if not plugins:
         plugins = []
+    elif isinstance(plugins, (str, Path)):
+        plugins = [plugins]
     else:
         plugins = list(plugins)
 

src/ocrmypdf/builtin_plugins/ghostscript.py

@@ -39,13 +39,11 @@ def check_options(options):
     if gs_version < '9.20' and options.output_type != 'pdf' and not is_latin:
         # https://bugs.ghostscript.com/show_bug.cgi?id=696874
         # Ghostscript < 9.20 fails to encode multibyte characters properly
-        msg = (
-            "The installed version of Ghostscript does not work correctly "
-            "with the OCR languages you specified. Use --output-type pdf or "
+        log.warning(
+            f"The installed version of Ghostscript ({gs_version}) does not work "
+            "correctly with the OCR languages you specified. Use --output-type pdf or "
             "upgrade to Ghostscript 9.20 or later to avoid this issue."
         )
-        msg += f"Found Ghostscript {gs_version}"
-        log.warning(msg)
 
     if options.output_type == 'pdfa':
         options.output_type = 'pdfa-2'

src/ocrmypdf/helpers.py

@@ -213,11 +213,15 @@ def clamp(n: T, smallest: T, largest: T) -> T:
 
 
 def pikepdf_enable_mmap():
-    try:
-        if pikepdf._qpdf.set_access_default_mmap(True):
-            log.debug("pikepdf mmap enabled")
-    except AttributeError:
-        log.debug("pikepdf mmap not available")
+    # try:
+    #     if pikepdf._qpdf.set_access_default_mmap(True):
+    #         log.debug("pikepdf mmap enabled")
+    # except AttributeError:
+    #     log.debug("pikepdf mmap not available")
+    # We found a race condition probably related to pybind issue #2252 that can
+    # cause a crash. For now, disable pikepdf mmap to be on the safe side.
+    log.debug("pikepdf mmap disabled")
+    return
 
 
 def deprecated(func):

src/ocrmypdf/hocrtransform.py

@@ -167,7 +167,10 @@ class HocrTransform:
     def topdown_position(self, element):
         pxl_line_coords = self.element_coordinates(element)
         line_box = self.pt_from_pixel(pxl_line_coords)
-        return -line_box.y2
+        # Coordinates here are still in the hocr coordinate system, so 0 on the y axis
+        # is the top of the page and increasing values of y will move towards the
+        # bottom of the page.
+        return line_box.y2
 
     def to_pdf(
         self,

src/ocrmypdf/leptonica.py

@@ -58,6 +58,24 @@ if not _libpath:
         ---------------------------------------------------------------------
         """
     )
+if os.name == 'nt':  
+    # On Windows, recent versions of libpng require zlib. We have to make sure 
+    # the zlib version being loaded is the same one that libpng was built with. 
+    # This tries to import zlib from Tesseract's installation folder, falling back
+    # to find_library() if liblept is being loaded from somewhere else.
+    # Loading zlib from other places could cause a version mismatch
+    _zlib_path = os.path.join(os.path.dirname(_libpath), 'zlib1.dll')
+    if not os.path.exists(_zlib_path):
+        _zlib_path = find_library('zlib')
+    try:
+        zlib = ffi.dlopen(_zlib_path)
+    except ffi.error as e:
+        raise MissingDependencyError(
+            """
+            Could not load the zlib library. It could be that Tesseract is not installed properly,
+            we can't find the installation on your system PATH environment variable.
+            """
+        ) from e
 try:
     lept = ffi.dlopen(_libpath)
     lept.setMsgSeverity(lept.L_SEVERITY_WARNING)

src/ocrmypdf/optimize.py

@@ -10,6 +10,7 @@ import sys
 import tempfile
 from collections import defaultdict
 from functools import partial
+from io import BytesIO
 from os import fspath
 from pathlib import Path
 from typing import (
@@ -27,6 +28,7 @@ from typing import (
     Union,
 )
 
+import img2pdf
 import pikepdf
 from pikepdf import Dictionary, Name, Object, Pdf, PdfImage
 from PIL import Image
@@ -37,7 +39,7 @@ from ocrmypdf._concurrent import exec_progress_pool
 from ocrmypdf._exec import jbig2enc, pngquant
 from ocrmypdf._jobcontext import PdfContext
 from ocrmypdf.exceptions import OutputFileAccessError
-from ocrmypdf.helpers import safe_symlink
+from ocrmypdf.helpers import deprecated, safe_symlink
 
 log = logging.getLogger(__name__)
 
@@ -75,23 +77,26 @@ def extract_image_filter(
     if image.Subtype != Name.Image:
         return None
     if image.Length < 100:
-        log.debug("Skipping small image, xref %s", xref)
+        log.debug(f"Skipping small image, xref {xref}")
         return None
 
     pim = PdfImage(image)
 
     if len(pim.filter_decodeparms) > 1:
-        log.debug("Skipping multiply filtered, xref %s", xref)
+        log.debug(f"Skipping multiply filtered image, xref {xref}")
         return None
     filtdp = pim.filter_decodeparms[0]
 
     if pim.bits_per_component > 8:
+        log.debug(f"Skipping wide gamut image, xref {xref}")
         return None  # Don't mess with wide gamut images
 
     if filtdp[0] == Name.JPXDecode:
+        log.debug(f"Skipping JPEG2000 iamge, xref {xref}")
         return None  # Don't do JPEG2000
 
     if Name.Decode in image:
+        log.debug(f"Skipping image with Decode table, xref {xref}")
         return None  # Don't mess with custom Decode tables
 
     return pim, filtdp
@@ -227,7 +232,9 @@ def extract_images(
                 # Ignore soft masks
                 smask_xref = Xref(image.SMask.objgen[0])
                 exclude_xrefs.add(smask_xref)
+                log.debug(f"Skipping image {smask_xref} because it is an SMask")
             include_xrefs.add(xref)
+            log.debug(f"Treating {xref} as an optimization candidate")
             if xref not in pageno_for_xref:
                 pageno_for_xref[xref] = pageno
 
@@ -393,6 +400,47 @@ def transcode_jpegs(pike: Pdf, jpegs: Sequence[Xref], root: Path, options) -> No
         im_obj.write(compdata.read(), filter=Name.DCTDecode)
 
 
+def _transcode_png(pike: Pdf, filename: Path, xref: Xref) -> bool:
+    output = filename.with_suffix('.png.pdf')
+    with output.open('wb') as f:
+        img2pdf.convert(fspath(filename), outputstream=f)
+
+    with pikepdf.open(output) as pdf_image:
+        foreign_image = next(pdf_image.pages[0].images.values())
+        local_image = pike.copy_foreign(foreign_image)
+
+        im_obj = pike.get_object(xref, 0)
+        im_obj.write(
+            local_image.read_raw_bytes(),
+            filter=local_image.Filter,
+            decode_parms=local_image.DecodeParms,
+        )
+
+        # Don't copy keys from the new image...
+        del_keys = set(im_obj.keys()) - set(local_image.keys())
+        # ...except for the keep_fields, which are essential to displaying
+        # the image correctly and preserving its metadata. (/Decode arrays
+        # and /SMaskInData are implicitly discarded prior to this point.)
+        keep_fields = {
+            '/ID',
+            '/Intent',
+            '/Interpolate',
+            '/Mask',
+            '/Metadata',
+            '/OC',
+            '/OPI',
+            '/SMask',
+            '/StructParent',
+        }
+        del_keys -= keep_fields
+        for key in local_image.keys():
+            if key != Name.Length and str(key) not in keep_fields:
+                im_obj[key] = local_image[key]
+        for key in del_keys:
+            del im_obj[key]
+    return True
+
+
 def transcode_pngs(
     pike: Pdf,
     images: Sequence[Xref],
@@ -435,34 +483,11 @@ def transcode_pngs(
         )
 
     for xref in modified:
-        im_obj = pike.get_object(xref, 0)
-        try:
-            pix = leptonica.Pix.open(png_name(root, xref))
-            if pix.mode == '1':
-                compdata = pix.generate_pdf_ci_data(leptonica.lept.L_G4_ENCODE, 0)
-            else:
-                compdata = leptonica.CompressedData.open(png_name(root, xref))
-        except leptonica.LeptonicaError as e:
-            # Most likely this means file not found, i.e. quantize did not
-            # produce an improved version
-            log.error(e)
-            continue
-
-        # If re-coded image is larger don't use it - we test here because
-        # pngquant knows the size of the temporary output file but not the actual
-        # object in the PDF
-        if len(compdata) > int(im_obj.stream_dict.Length):
-            log.debug(
-                f"pngquant: pngquant did not improve over original image "
-                f"{len(compdata)} > {int(im_obj.stream_dict.Length)}"
-            )
-            continue
-        if compdata.type == leptonica.lept.L_FLATE_ENCODE:
-            rewrite_png(pike, im_obj, compdata)
-        elif compdata.type == leptonica.lept.L_G4_ENCODE:
-            rewrite_png_as_g4(pike, im_obj, compdata)
+        filename = png_name(root, xref)
+        _transcode_png(pike, filename, xref)
 
 
+@deprecated
 def rewrite_png_as_g4(pike: Pdf, im_obj: Object, compdata) -> None:
     im_obj.BitsPerComponent = 1
     im_obj.Width = compdata.w
@@ -483,6 +508,7 @@ def rewrite_png_as_g4(pike: Pdf, im_obj: Object, compdata) -> None:
     return
 
 
+@deprecated
 def rewrite_png(pike: Pdf, im_obj: Object, compdata) -> None:
     # When a PNG is inserted into a PDF, we more or less copy the IDAT section from
     # the PDF and transfer the rest of the PNG headers to PDF image metadata.

src/ocrmypdf/pluginspec.py

@@ -154,11 +154,29 @@ def filter_page_image(page: 'PageContext', image_filename: Path) -> Path:
     produced for a given page, this function will not be called. This is not
     the image that will be shown to OCR.
 
-    ocrmypdf will create the PDF page based on the image format used. If you
+    If the function does not want to modify the image, it should return
+    ``image_filename``. The hook may overwrite ``image_filename`` with a new file.
+
+    The output image should preserve the same physical unit dimensions, that is
+    (width * dpi_x, height * dpi_y). That is, if the image is resized, the DPI
+    must be adjusted by the reciprocal. If this is not preserved, the PDF page
+    will be resized and the OCR layer misaligned. OCRmyPDF does not nothing
+    to enforce these constraints; it is up to the plugin to do sensible things.
+
+    OCRmyPDF will create the PDF page based on the image format used. If you
     convert the image to a JPEG, the output page will be created as a JPEG, etc.
-    Note that the ocrmypdf image optimization stage may ultimately chose a
+    If you change the colorspace, that change will be kept. Note that the
+    OCRmyPDF image optimization stage, if enabled, may ultimately chose a
     different format.
 
+    If the return value is a file that does not exist, ``FileNotFoundError``
+    will occur. The return value should be a path to a file in the same folder
+    as ``image_filename``.
+
+    Implementation detail: If the value returned is falsy, OCRmyPDF will ignore
+    the return value and assume the input file was unmodified. This is deprecated.
+    To leave the image unmodified, ``image_filename`` should be returned.
+
     Note:
         This hook will be called from child processes. Modifying global state
         will not affect the main process or other child processes.

tests/test_validation.py

@@ -55,7 +55,7 @@ def test_old_ghostscript(caplog):
         vd._check_options(
             *make_opts_pm(language='chi_sim', output_type='pdfa'), {'chi_sim'}
         )
-        assert 'Ghostscript does not work correctly' in caplog.text
+        assert 'does not work correctly' in caplog.text
 
     with patch('ocrmypdf._exec.ghostscript.version', return_value='9.18'), patch(
         'ocrmypdf._exec.tesseract.has_textonly_pdf', return_value=True