jedarden
|
a79260b139
|
feat(pdftract-h2s0z): implement adaptive word boundary detector
Implement Phase 3.2 word boundary detection algorithm:
- Bootstrap threshold = 0.25 × font_size for first 20 glyphs
- Recalibrate to 1.5× median of last 20 gaps every 5 samples
- Exclude outliers > 4× current threshold
- Reset on Tf (font switch) and BT operators
- Negative gaps never trigger word boundaries
Closes: pdftract-h2s0z
Files:
- crates/pdftract-core/src/word_boundary.rs (NEW): WordBoundaryDetector, WordBoundaryManager, TextState
- crates/pdftract-core/src/lib.rs: Export word_boundary module
- crates/pdftract-core/src/font/resolver.rs: Add from_usize test constructor
- notes/pdftract-h2s0z.md: Verification note
Tests: 27 word_boundary tests all passing
|
2026-05-24 06:06:56 -04:00 |
|