extract_pymupdf.py 3.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101
  1. #!/usr/bin/env python3
  2. """Extract text from documents using pymupdf. Lightweight (~25MB), no models.
  3. Usage:
  4. python extract_pymupdf.py document.pdf
  5. python extract_pymupdf.py document.pdf --markdown
  6. python extract_pymupdf.py document.pdf --pages 0-4
  7. python extract_pymupdf.py document.pdf --images output_dir/
  8. python extract_pymupdf.py document.pdf --tables
  9. python extract_pymupdf.py document.pdf --metadata
  10. """
  11. import sys
  12. import json
  13. def extract_text(path, pages=None):
  14. import pymupdf
  15. doc = pymupdf.open(path)
  16. page_range = range(len(doc)) if pages is None else pages
  17. for i in page_range:
  18. if i < len(doc):
  19. print(f"\n--- Page {i+1}/{len(doc)} ---\n")
  20. print(doc[i].get_text())
  21. def extract_markdown(path, pages=None):
  22. import pymupdf4llm
  23. md = pymupdf4llm.to_markdown(path, pages=pages)
  24. print(md)
  25. def extract_tables(path):
  26. import pymupdf
  27. doc = pymupdf.open(path)
  28. for i, page in enumerate(doc):
  29. tables = page.find_tables()
  30. for j, table in enumerate(tables.tables):
  31. print(f"\n--- Page {i+1}, Table {j+1} ---\n")
  32. df = table.to_pandas()
  33. print(df.to_markdown(index=False))
  34. def extract_images(path, output_dir):
  35. import pymupdf
  36. from pathlib import Path
  37. Path(output_dir).mkdir(parents=True, exist_ok=True)
  38. doc = pymupdf.open(path)
  39. count = 0
  40. for i, page in enumerate(doc):
  41. for img_idx, img in enumerate(page.get_images(full=True)):
  42. xref = img[0]
  43. pix = pymupdf.Pixmap(doc, xref)
  44. if pix.n >= 5:
  45. pix = pymupdf.Pixmap(pymupdf.csRGB, pix)
  46. out_path = f"{output_dir}/page{i+1}_img{img_idx+1}.png"
  47. pix.save(out_path)
  48. count += 1
  49. print(f"Extracted {count} images to {output_dir}/")
  50. def show_metadata(path):
  51. import pymupdf
  52. doc = pymupdf.open(path)
  53. print(json.dumps({
  54. "pages": len(doc),
  55. "title": doc.metadata.get("title", ""),
  56. "author": doc.metadata.get("author", ""),
  57. "subject": doc.metadata.get("subject", ""),
  58. "creator": doc.metadata.get("creator", ""),
  59. "producer": doc.metadata.get("producer", ""),
  60. "format": doc.metadata.get("format", ""),
  61. }, indent=2))
  62. if __name__ == "__main__":
  63. import argparse
  64. parser = argparse.ArgumentParser(
  65. description="Extract text/tables/images/metadata from documents using pymupdf (lightweight, no models)."
  66. )
  67. parser.add_argument("path", help="Document to read")
  68. parser.add_argument("--pages", help="Page selection: N or START-END (0-indexed)")
  69. parser.add_argument("--markdown", action="store_true", help="Markdown output via pymupdf4llm")
  70. parser.add_argument("--tables", action="store_true", help="Extract tables as markdown")
  71. parser.add_argument("--images", nargs="?", const="./images", metavar="OUTPUT_DIR",
  72. help="Extract embedded images to OUTPUT_DIR (default ./images)")
  73. parser.add_argument("--metadata", action="store_true", help="Show document metadata as JSON")
  74. args = parser.parse_args()
  75. pages = None
  76. if args.pages:
  77. if "-" in args.pages:
  78. start, end = args.pages.split("-")
  79. pages = list(range(int(start), int(end) + 1))
  80. else:
  81. pages = [int(args.pages)]
  82. if args.metadata:
  83. show_metadata(args.path)
  84. elif args.tables:
  85. extract_tables(args.path)
  86. elif args.images is not None:
  87. extract_images(args.path, args.images)
  88. elif args.markdown:
  89. extract_markdown(args.path, pages=pages)
  90. else:
  91. extract_text(args.path, pages=pages)