← Back to Blog

Vectorize scanned maps in one tool — Extract Scanned Lines and Polygons in ArcGIS Pro 3.7

A scanned paper map being converted into clean vector line features in ArcGIS Pro

Every long-lived GIS shop has the same skeleton in the closet: a drawer (or a network share) full of scanned maps that never became data. Utility as-builts, old parcel books, soil surveys, hand-drawn contour sheets — georeferenced at best, traced by hand at worst. ArcGIS Pro 3.7 finally ships purpose-built geoprocessing for this: Extract Scanned Lines and Extract Scanned Polygons, two new tools that read the linework off a scanned image and hand back clean vector features.

Released in the May 2026 update, both tools live in the Conversion toolbox, in the From Raster toolset. If you remember ArcScan from the ArcMap era, this is the idea reborn as modern, scriptable geoprocessing — no extension, no raster-cleanup editing session, no separate toolbar. Here is what the tools actually do, the parameters that decide whether your output is usable, and the two catches (input format and licensing) to know before you plan a conversion project around them.

What the two tools do

Both tools take a scanned map image that has been classified into exactly two classes — foreground linework and background — and convert the foreground cells to features:

Extract Scanned Lines can even do both jobs at once: its optional Output Polygon Features parameter catches anything wider than your Maximum Line Width and writes it out as polygons, so a sheet that mixes thin linework with filled symbols doesn’t need two passes.

The workflow this replaces

Until now, Esri’s own recommended path for this problem was a relay race across toolboxes. The support knowledge base (Article 000034497) walks through it: georeference the scan, run Reclassify (Spatial Analyst license) to isolate the linework, run Raster to Polyline, then patch the result with Snap, Simplify Line, and Smooth Line, and finally Merge or Dissolve the pieces together.

That chain works, but it treats the symptoms one tool at a time — and none of those tools understand that the raster is scanned linework. Raster to Polyline has no concept of a line intersection, a scanning gap, or a corner; you fix those artifacts downstream, feature by feature. The new tools bake that understanding into the extraction itself, which is exactly where it belongs.

The parameters that decide your output quality

Intersection Solution — the one to get right first

Where three or more raster lines meet, the tool has to decide what geometry to build. Extract Scanned Lines gives you three strategies, and Esri’s documentation is unusually concrete about when to use each:

OptionWhat it doesUse it for
Geometrical (default)Preserves angles and straight lines at intersectionsSite plans, engineering drawings — anywhere angular accuracy is critical
MedianAverages directions where lines meet at irregular anglesNatural and organic features — environmental mapping, streams, soil lines
NoneNo special handling; extracts as-isData with no meaningful intersections — elevation contours

Cleanup built into the extraction

The rest of the parameter list reads like a checklist of everything that used to go wrong in the old multi-tool chain:

A practical first run

  1. Georeference the scan first. The extraction inherits the raster’s coordinates; vectorizing an unreferenced image gives you beautifully clean features in the wrong universe.
  2. Make it binary. Both tools require the raster classified into exactly two classes. A Reclassify pass (or any classification that yields foreground vs. background) gets you there; note the foreground value.
  3. Run Extract Scanned Lines with defaults on a representative subsection — set only the input, foreground value, output, and an Intersection Solution chosen from the table above.
  4. Iterate on the cleanup parameters where the output disappoints: gaps in faded lines → Close Gaps; speckle features → Noise Level; jagged strokes → smoothing; blobby wide symbols → Maximum Line Width + Output Polygon Features.
  5. QA with topology rules (dangles, self-intersections) before the data goes anywhere near production — automated extraction is a head start, not a substitute for validation.

Because both tools are standard geoprocessing in the Conversion toolbox, they are fully scriptable — arcpy.conversion.ExtractScannedLines and arcpy.conversion.ExtractScannedPolygons — so a drawer of two hundred sheets becomes a batch job over a folder of georeferenced, binarized rasters rather than two hundred editing sessions. That is the real payoff: the tools turn map modernization from an editing project into a pipeline.

The two catches

Licensing: both tools require an ArcGIS Pro Advanced license — they are not available at Basic or Standard. Check your entitlements before you scope a conversion project around them. (The old workflow wasn’t free either: Reclassify needs Spatial Analyst, and Snap/Simplify/Smooth need Standard or Advanced.)
Input prep still matters. The tools want a two-class binary image, and Esri’s usage notes are explicit that results depend on the state of the scan — eliminating noise and unwanted pixels beforehand is sometimes necessary. A clean, high-contrast, well-georeferenced scan in; clean features out. The tools forgive a lot, but they don’t forgive everything.

Where this pays off

The organizations that should care most are the ones sitting on paper-era records with modern-data obligations: utilities with scanned as-builts feeding a network migration, counties with parcel books that predate their geodatabase, environmental agencies with decades of hand-drawn survey sheets. For those archives, the difference between “a digitizing tech traces sheets for a year” and “a parameterized batch pipeline with human QA” is the difference between a project that gets funded and one that doesn’t.

References

Sitting on a drawer of scanned maps?

We build scan-to-vector pipelines — georeferencing, extraction parameter tuning, batch automation with ArcPy, and topology-driven QA — so your paper archive becomes production data, not a year of tracing.

Book a free intro call