fix pdf truncation after inline images

Author: Muhtasim-Munif-Fahim

packages/markitdown/README.md

@@ -16,6 +16,12 @@ From PyPI:
 pip install markitdown[all]
 ```
 
+If you need the optional PyMuPDF fallback for some PDFs:
+
+```bash
+pip install markitdown[pymupdf]
+```
+
 From source:
 
 ```bash

packages/markitdown/pyproject.toml

@@ -42,6 +42,7 @@ all = [
   "lxml",
   "pdfminer.six>=20251230",
   "pdfplumber>=0.11.9",
+  "pymupdf",
   "olefile",
   "pydub",
   "SpeechRecognition",
@@ -55,6 +56,7 @@ docx = ["mammoth~=1.11.0", "lxml"]
 xlsx = ["pandas", "openpyxl"]
 xls = ["pandas", "xlrd"]
 pdf = ["pdfminer.six>=20251230", "pdfplumber>=0.11.9"]
+pymupdf = ["pymupdf"]
 outlook = ["olefile"]
 audio-transcription = ["pydub", "SpeechRecognition"]
 youtube-transcription = ["youtube-transcript-api"]

packages/markitdown/src/markitdown/converters/_pdf_converter.py

@@ -57,6 +57,23 @@ def _merge_partial_numbering_lines(text: str) -> str:
     return "\n".join(result_lines)
 
 
+def _extract_with_pymupdf(pdf_bytes: io.BytesIO) -> str | None:
+    """Extract text with PyMuPDF when the primary PDF parsers look truncated."""
+    if fitz is None:
+        return None
+
+    pdf_bytes.seek(0)
+    chunks: list[str] = []
+    with fitz.open(stream=pdf_bytes.read(), filetype="pdf") as doc:
+        for page in doc:
+            text = page.get_text("text")
+            if text and text.strip():
+                chunks.append(text.strip())
+
+    markdown = "\n\n".join(chunks).strip()
+    return markdown or None
+
+
 # Load dependencies
 _dependency_exc_info = None
 try:
@@ -66,6 +83,11 @@ def _merge_partial_numbering_lines(text: str) -> str:
 except ImportError:
     _dependency_exc_info = sys.exc_info()
 
+try:
+    import fitz
+except ImportError:
+    fitz = None
+
 
 ACCEPTED_MIME_TYPE_PREFIXES = [
     "application/pdf",
@@ -536,6 +558,9 @@ def convert(
 
         assert isinstance(file_stream, io.IOBase)
 
+        markdown_chunks: list[str] = []
+        has_images = False
+
         # Read file stream into BytesIO for compatibility with pdfplumber
         pdf_bytes = io.BytesIO(file_stream.read())
 
@@ -545,12 +570,14 @@ def convert(
             # pages are collected separately. page.close() is called
             # after each page to free pdfplumber's cached objects and
             # keep memory usage constant regardless of page count.
-            markdown_chunks: list[str] = []
             form_page_count = 0
             plain_page_indices: list[int] = []
 
             with pdfplumber.open(pdf_bytes) as pdf:
                 for page_idx, page in enumerate(pdf.pages):
+                    has_images = has_images or bool(getattr(page, "images", None))
+
+                    # Try form-style word position extraction
                     page_content = _extract_form_content_from_words(page)
 
                     if page_content is not None:
@@ -583,6 +610,24 @@ def convert(
             pdf_bytes.seek(0)
             markdown = pdfminer.high_level.extract_text(pdf_bytes)
 
+        # Recover from inline-image truncation cases where the primary parsers
+        # return a much shorter body than an optional PyMuPDF pass.
+        if fitz is not None and has_images and markdown and len(markdown) < 2048:
+            try:
+                pymupdf_markdown = _extract_with_pymupdf(pdf_bytes)
+            except Exception:
+                pymupdf_markdown = None
+            else:
+                if pymupdf_markdown is not None:
+                    primary_length = len(markdown.strip())
+                    pymupdf_length = len(pymupdf_markdown.strip())
+                    if pymupdf_length > primary_length and (
+                        primary_length == 0
+                        or pymupdf_length >= primary_length * 1.5
+                        or pymupdf_length - primary_length >= 200
+                    ):
+                        markdown = pymupdf_markdown
+
         # Post-process to merge MasterFormat-style partial numbering with following text
         markdown = _merge_partial_numbering_lines(markdown)
 

packages/markitdown/tests/test_module_misc.py

@@ -6,6 +6,8 @@
 import pytest
 from unittest.mock import MagicMock
 
+import markitdown.converters._pdf_converter as pdf_converter_module
+
 from markitdown._uri_utils import parse_data_uri, file_uri_to_path
 
 from markitdown import (
@@ -432,6 +434,67 @@ def test_exceptions() -> None:
     assert type(exc_info.value.attempts[0].converter).__name__ == "PptxConverter"
 
 
+def test_pdf_converter_prefers_pymupdf_when_primary_extraction_is_truncated(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    class _FakePage:
+        images = [object()]
+
+        def extract_words(self, keep_blank_chars=True, x_tolerance=3, y_tolerance=3):
+            return []
+
+        def extract_text(self):
+            return "BEFORE_IMAGE: this text should be extracted"
+
+    class _FakePdf:
+        pages = [_FakePage()]
+
+        def __enter__(self):
+            return self
+
+        def __exit__(self, exc_type, exc, tb):
+            return False
+
+    class _FakePyMuPdfPage:
+        def get_text(self, kind):
+            assert kind == "text"
+            return (
+                "BEFORE_IMAGE: this text should be extracted\n"
+                "AFTER_IMAGE: this text should also be extracted"
+            )
+
+    class _FakePyMuPdfDoc:
+        def __enter__(self):
+            return self
+
+        def __exit__(self, exc_type, exc, tb):
+            return False
+
+        def __iter__(self):
+            return iter([_FakePyMuPdfPage()])
+
+    monkeypatch.setattr(pdf_converter_module.pdfplumber, "open", lambda _: _FakePdf())
+    monkeypatch.setattr(
+        pdf_converter_module.pdfminer.high_level,
+        "extract_text",
+        lambda _: "BEFORE_IMAGE: this text should be extracted",
+    )
+    monkeypatch.setattr(
+        pdf_converter_module.fitz,
+        "open",
+        lambda *args, **kwargs: _FakePyMuPdfDoc(),
+    )
+
+    converter = pdf_converter_module.PdfConverter()
+    result = converter.convert(
+        io.BytesIO(b"%PDF-1.4 fake"),
+        StreamInfo(mimetype="application/pdf", extension=".pdf"),
+    )
+
+    assert "BEFORE_IMAGE: this text should be extracted" in result.markdown
+    assert "AFTER_IMAGE: this text should also be extracted" in result.markdown
+
+
 @pytest.mark.skipif(
     skip_exiftool,
     reason="do not run if exiftool is not installed",