Vectorization in Drawing Lab¶
Status: Drawing Lab uses vector evidence; deeper method detail stays in the Vectorization surface
Authority: Current contract plus existing Vectorization wiki
Last reviewed: 2026-08-14
Question¶
How does Drawing Lab use vectorization without turning this wiki into the vector method catalog?
Short answer¶
Drawing Lab consumes vector evidence for appearance review and geometry carriers. The Vectorization wiki owns deeper technique history and validation boundary detail. This page keeps only the Drawing Lab claim boundary.
Appearance evidence¶
Potrace recipe potrace-0.80 produces closed ink outlines.
Those outlines follow the boundary of black regions.
Inkscape four-scan produces a sharp-dark multi-scan appearance stack.
Multi-scan appearance remains a review benchmark.
Appearance evidence is not engineering geometry by itself.
Centreline recovery for Vector Lines¶
Vector Lines starts from Potrace potrace-0.80 closed outlines.
It reconstructs the fill of those closed regions.
It derives a one-pixel medial-axis centreline from that fill.
It then fits compatible Hough segments on the centreline.
Only after that conversion can the result act as an engineering carrier candidate.
Drawing Lab uses¶
- Protected whole-sheet vectorizer as a separately identified baseline
- Potrace recipe
potrace-0.80as fallback appearance evidence - Inkscape four-scan as an appearance benchmark
- Centreline recovery for Vector Lines dimension comparison
- Ridge vectors as centreline evidence where that path is available
Drawing Lab does not claim here¶
- That outline boundaries are engineering lines
- That whole-sheet topology candidates are semantic views
- That vector completeness equals validated engineering reconstruction
- That every vector experiment belongs in this surface
Naming note¶
The lab method ID potrace-0.80 names the brightness-cutoff recipe with threshold 0.80.
The Potrace engine version in the Detection worktree is Potrace 1.16.
Do not confuse recipe threshold with engine version.
Deep link¶
For pipeline stages, technique findings, and vector validation limits, open Vectorization.
Open Questions¶
- Which vector evidence layers become mandatory Detected Sheet appearance layers for every sheet?
Sources¶
- Current contract:
table-dimension-detection-architecture.md(2026-08-12) - Existing wiki surface: Vectorization
- Accepted decision: DL-010