Orientation and line progression are separate layout decisions
Set the expected orientation and line progression before recognition, then verify vertical columns and horizontal inserts as separate layout decisions.
The intended output is a documented region order for vertical columns and horizontal inserts, with orientation, column progression, punctuation, and rotated characters checked independently.
Test column progression and horizontal inserts with Bilingual OCR. Then compare vertical region sequence with A Reviewable Workflow for Mixed-Language OCR and Multi-Column OCR and Reading Order Review.
Vertical-layout review
- Identify vertical line progression
- Map horizontal inserts
- Check punctuation orientation
- Separate small annotations
- Verify first and last column explicitly
Map vertical columns and horizontal inserts before export
Identify the writing mode
Determine whether lines run vertically and which direction successive columns progress. Mixed pages may contain horizontal headings, page numbers, captions, or Latin identifiers that need separate region treatment.
Preserve character orientation
Some characters remain upright while others rotate in vertical layout. Do not judge OCR only from a rotated preview; compare exported characters and punctuation with the original orientation.
Handle annotations deliberately
Small pronunciation guides, interlinear notes, and marginal text can be merged into the main line. Decide whether they belong in the main transcription, annotations, or a separate layer.
Verify column sequence
Mark the first and last column and trace progression across the page. Multi-region historical material may mix vertical body text with horizontal tables or labels.
Map column progression and horizontal inserts separately
On a vertical page, identify the orientation of glyphs, the direction within each column, and the order in which columns progress. Mark any horizontal dates, page numbers, captions, or Latin fragments as separate regions before recognition.
Read the exported text without relying on the page image. Check whether columns were reversed, split, or joined across annotations. If the engine performs better on rotated regional crops, retain coordinates and orientation metadata so the pieces can be returned to the correct page sequence.
- Separate orientation from reading order
- Isolate horizontal inserts
- Retain coordinates when rotating regions
Vertical-layout components
| Component | Likely failure | Treatment |
|---|---|---|
| Main vertical columns | Wrong column sequence | Map progression before export |
| Horizontal heading | Rotated or merged | Recognize as a separate region |
| Small reading annotation | Inserted into body | Store as an annotation layer |
| Latin identifier | Character orientation changes | Verify exact token separately |
Rotation alone does not solve vertical layout
Glyph orientation, line direction, column progression, punctuation, and horizontal inserts are separate decisions. Rotating the whole page can create a readable but incorrectly ordered result.