feat: anti-alias Ghostscript rasterization to improve OCR quality

Ghostscript 10.x renders aliased glyphs that OCR frequently misreads as
extra word breaks or substituted characters. Enable text and graphics
anti-aliasing (-dTextAlphaBits=4 -dGraphicsAlphaBits=4) for the contone
raster devices, which empirically improves OCR accuracy on the
Ghostscript path, especially for small fonts at moderate DPI. The 1-bit
mono devices are excluded, since older Ghostscript rejects alpha bits on
them and pngmonod performs its own anti-aliased downscaling.

Also log which rasterizer rendered each page at debug verbosity and
clarify the --rasterizer help text, so quality reports are easier to
diagnose. The default rasterizer (auto) already prefers pypdfium2, which
anti-aliases; this primarily benefits --rasterizer ghostscript and
installs without pypdfium2.

Closes #1439.

docs/releasenotes/version17.md

@@ -62,6 +62,20 @@
   better input for OCR on faint or anti-aliased scans at negligible cost and
   no change to output file size, since the rasterized image is an
   intermediate that is discarded after OCR.
+- When rasterizing pages with Ghostscript, OCRmyPDF now enables text and
+  graphics anti-aliasing (``-dTextAlphaBits=4 -dGraphicsAlphaBits=4``) for the
+  grayscale and color raster devices. Ghostscript 10.x renders aliased glyphs
+  that OCR frequently misreads as extra word breaks or substituted characters;
+  anti-aliasing materially improves OCR accuracy on the Ghostscript
+  rasterization path, especially for small fonts at moderate resolution. The
+  1-bit monochrome devices are unaffected, since they perform their own
+  anti-aliased downscaling and older Ghostscript versions reject alpha-bit
+  options on them. Note that the default rasterizer (``--rasterizer auto``)
+  prefers pypdfium2, which already anti-aliases; this change benefits users who
+  select ``--rasterizer ghostscript`` or do not have pypdfium2 installed.
+  OCRmyPDF now also logs which rasterizer rendered each page at debug verbosity
+  (``-v 1``), and the ``--rasterizer`` help text explains the OCR-quality
+  trade-off, to make such reports easier to diagnose. {issue}`1439`
 
 ## v17.5.0
 

src/ocrmypdf/_exec/ghostscript.py

@@ -150,6 +150,19 @@ def rasterize_pdf(
     else:
         effective_dpi = raster_dpi
 
+    # Anti-alias text and vector graphics when rendering to a contone device.
+    # Ghostscript 10.x renders aliased glyphs that OCR frequently misreads as
+    # extra word breaks; anti-aliasing empirically improves OCR accuracy on the
+    # Ghostscript path, especially for small fonts at moderate DPI (#1439).
+    # The 1-bit mono devices do not accept alpha bits (older Ghostscript
+    # rejects them) and pngmonod performs its own anti-aliased downscaling.
+    mono_devices = (GhostscriptRasterDevice.PNGMONO, GhostscriptRasterDevice.PNGMONOD)
+    antialias_args = (
+        []
+        if raster_device in mono_devices
+        else ['-dTextAlphaBits=4', '-dGraphicsAlphaBits=4']
+    )
+
     args_gs = (
         [
             GS,
@@ -162,6 +175,7 @@ def rasterize_pdf(
             f'-dLastPage={pageno}',
             f'-r{effective_dpi.x:f}x{effective_dpi.y:f}',
         ]
+        + antialias_args
         + (['-dUseCropBox'] if use_cropbox else [])
         + (['-dFILTERVECTOR'] if filter_vector else [])
         + (['-dPDFSTOPONERROR'] if stop_on_error else [])

src/ocrmypdf/builtin_plugins/ghostscript.py

@@ -256,6 +256,8 @@ def rasterize_pdf_page(
         # Let pypdfium handle it (it will error in check_options if unavailable)
         return None
 
+    log.debug("Rasterizing page %d with the Ghostscript rasterizer", pageno)
+
     ghostscript.rasterize_pdf(
         input_file,
         output_file,

src/ocrmypdf/builtin_plugins/pypdfium.py

@@ -255,6 +255,8 @@ def rasterize_pdf_page(
     if pdfium is None:
         return None  # Fall back to Ghostscript
 
+    log.debug("Rasterizing page %d with the pypdfium2 rasterizer", pageno)
+
     # Acquire lock to ensure thread-safe access to pypdfium2
     with (
         _pdfium_lock,

src/ocrmypdf/cli.py

@@ -423,9 +423,13 @@ Online documentation is located at:
         '--rasterizer',
         choices=['auto', 'ghostscript', 'pypdfium'],
         default='auto',
-        help="Choose PDF page rasterizer. 'auto' prefers pypdfium when available, "
-        "falling back to Ghostscript. 'pypdfium' is faster but requires the "
-        "pypdfium2 package. 'ghostscript' uses the traditional Ghostscript rasterizer.",
+        help="Choose PDF page rasterizer. 'auto' (the default) prefers pypdfium2 "
+        "when the pypdfium2 package is installed, falling back to Ghostscript "
+        "otherwise. pypdfium2 anti-aliases page content and generally produces "
+        "better input for OCR than Ghostscript 10.x, which can render aliased "
+        "glyphs that OCR misreads as extra word breaks. 'pypdfium' forces the "
+        "pypdfium2 rasterizer (requires the pypdfium2 package); 'ghostscript' "
+        "forces the traditional Ghostscript rasterizer.",
     )
     advanced.add_argument(
         '--rotate-pages-threshold',

tests/test_ghostscript.py

@@ -141,6 +141,62 @@ def test_rasterize_low_dpi_one_axis(francais, outdir):
         assert im.info['dpi'] == forced_dpi
 
 
+def _capture_rasterize_args(resources, outdir, raster_device):
+    """Run rasterize_pdf with the gs subprocess mocked; return the gs argv."""
+    out = outdir / 'out.png'
+    captured = {}
+
+    def fake_run(args, **kwargs):
+        captured['args'] = list(args)
+        # Produce a valid PNG so rasterize_pdf's post-processing succeeds.
+        Image.new('RGB', (2, 2)).save(out)
+        return subprocess.CompletedProcess(args, returncode=0, stdout=b'', stderr=b'')
+
+    with patch('ocrmypdf._exec.ghostscript.run', side_effect=fake_run):
+        rasterize_pdf(
+            resources / 'francais.pdf',
+            out,
+            raster_device=raster_device,
+            raster_dpi=Resolution(150.0, 150.0),
+        )
+    return captured['args']
+
+
+@pytest.mark.parametrize(
+    'raster_device',
+    [
+        GhostscriptRasterDevice.PNGGRAY,
+        GhostscriptRasterDevice.PNG256,
+        GhostscriptRasterDevice.PNG16M,
+    ],
+)
+def test_rasterize_antialiases_contone_devices(resources, outdir, raster_device):
+    """Contone raster devices receive anti-aliasing flags to aid OCR.
+
+    Ghostscript 10.x renders aliased glyphs that OCR misreads as extra word
+    breaks; -dTextAlphaBits/-dGraphicsAlphaBits markedly improve accuracy,
+    especially for small fonts at moderate DPI (see issue #1439).
+    """
+    args = _capture_rasterize_args(resources, outdir, raster_device)
+    assert '-dTextAlphaBits=4' in args
+    assert '-dGraphicsAlphaBits=4' in args
+
+
+@pytest.mark.parametrize(
+    'raster_device',
+    [GhostscriptRasterDevice.PNGMONO, GhostscriptRasterDevice.PNGMONOD],
+)
+def test_rasterize_no_antialias_on_mono_devices(resources, outdir, raster_device):
+    """1-bit mono devices must not receive alpha-bit flags.
+
+    Older Ghostscript versions reject -dTextAlphaBits on 1-bit devices, and
+    pngmonod performs its own anti-aliased downscaling.
+    """
+    args = _capture_rasterize_args(resources, outdir, raster_device)
+    assert not any(a.startswith('-dTextAlphaBits') for a in args)
+    assert not any(a.startswith('-dGraphicsAlphaBits') for a in args)
+
+
 def test_generate_pdfa_default_jpeg_quality(outdir):
     """When jpeg_quality is None, Ghostscript receives -dJPEGQ=95 (default)."""
     with (