Math Formula PDF to Editable Word

The Problem

Scanned math exams or supplementary PDF files are often image-only, with no selectable text. Formulas and tables cannot be edited directly in Word, and raw OCR output can still produce broken formulas, HTML tables, and missing layout rules. This skill targets math PDFs that need to become Word documents for further editing: formulas become OMML, tables become native Word tables, and the layout follows A4 exam-style conventions.

How It Works

The skill continues from existing MinerU output, usually checking a directory that contains full.md and images/, and asks the user to remove unrelated images such as QR-code ads or publisher logos. It then runs a conversion pipeline: normalize HTML tables into Pandoc-compatible Markdown pipe tables; generate DOCX with Pandoc, converting formulas to editable OMML and tables into Word tables; apply fixed layout parameters such as A4 paper, 2.5cm top and bottom margins, 3.2cm left and right margins, body text settings, and 1.5 line spacing; and clean MinerU artifacts like text_image or geometry-caption noise.

Boundaries and Caveats

It is best suited to image-only math PDFs, especially exams, workbooks, and study materials. The local environment needs MinerU, Pandoc, and Python. Pandoc does not directly convert HTML tables, so table normalization is required; image paths also require running from the Markdown directory to avoid broken references. Some tables may remain as images if MinerU does not recognize them, and merged cells cannot be fully represented in Markdown pipe tables, so manual adjustment in Word may still be needed.