Heading Detection

Approach

EdgeParse determines heading levels by analyzing:

1. Font size — larger text is more likely to be a heading
2. Font weight — bold text signals heading intent
3. Spacing — headings typically have more vertical space above
4. Position — headings often start at the left margin
5. Tagged PDF structure — H1–H6 tags when available

Heading Levels

Level	Typical Characteristics
title (H1)	Largest font size, first element
section (H2)	Second-largest font, bold
subsection (H3)	Third-largest font, bold
sub-subsection (H4)	Bold, slightly larger than body

MHS Score

EdgeParse achieves a MHS (Markdown Heading Similarity) score of 0.818:

Tool	MHS Score
Docling	0.824
EdgeParse	0.818
Marker	0.794
PyMuPDF4LLM	0.774

Output

{
  "type": "heading",
  "id": 1,
  "level": "section",
  "heading level": 2,
  "page number": 1,
  "content": "Financial Overview",
  "font": "Helvetica-Bold",
  "font size": 18.0
}