Slide 1

Slide 1 text

Bringing Old Maps & Illustrated Maps to the Web: 10 Years of M and Turning Misalignment into Innovation KOHEI OTSUKA — CTO, Nayuta, Inc. FOSS4G HIROSHIMA 2026 | 2 Sep 2026, 11:00 | Dahlia2

Slide 2

Slide 2 text

P ar t 1 M — ten years Where we came from, and where we stand 2

Slide 3

Slide 3 text

Why conventional GIS tools struggle A survey The same town, drawn • Distortion, scaling and rotation are intrinsic — not errors • Georeferencing warps the image; the drawing is destroyed • No runtime bidirectional transform Not error. Intent. 3

Slide 4

Slide 4 text

Yamagata Archive — a public archive of old maps • Yamagata University Museum, published since 2023 • Public-domain maps of the castle town, on the web • Built with the same stack a small team can run "Map of Yamagata Castle Town, Mogami Era" — Yamagata University Museum | © OpenStreetMap contributors 4

Slide 5

Slide 5 text

What appears when you switch between them "Map of Yamagata Castle Town, Mogami Era" — Yamagata University Museum | © OpenStreetMap contributors The Mamigasaki river moved. Downstream it still matches; upstream it does not. 5 © OpenStreetMap contributors

Slide 6

Slide 6 text

Recorded demo — Hiroshima, including this venue Your position, on the old map The drawing is untouched A published site — it covers this room s.maplat.jp/r/hiroshimamap/ Base map: 地理院タイル © 国土地理院 | Old map: 安芸国広島城所絵図 — National Archives of Japan 6

Slide 7

Slide 7 text

2026 — three anniversaries at once 20 10 1st 20 years since Nayuta, Inc. was founded 10 years of Maplat First year of the relaunched Nayuta 7

Slide 8

Slide 8 text

M — seventeen years, three principles 2008 2015 2017 2020 Started on the problem Prototype — yet another principle Patent filed (2016 principle) Patent granted 2009 2016 2018 2026 First solution — a different principle New principle; renamed Maplat Triple award, MLIT contest Nayuta succeeded; business begins 8

Slide 9

Slide 9 text

Patent and open source, together • Japanese patent 6684776 — held by the inventor, personally • Nayuta, Inc. develops and delivers it as a business • Shipped under the Apache License 2.0 • Five repositories — Version 1.0 (release candidate) Earlier patent 5810411 — filed 2011, granted 2015. Sole inventor; rights not held (work for hire) 9

Slide 10

Slide 10 text

P ar t 2 Turning misalignment into innovation The philosophy at the heart of Maplat: every map has its own rationality 10

Slide 11

Slide 11 text

This is a map. It shows almost no positions. • Marshall Islands, Micronesia • Sticks trace swells refracting around islands • Shells mark the islands themselves • Never carried to sea — it was for learning Photo: Wikimedia user Mattes, CC BY-SA 3.0 — Staatliches Museum für Völkerkunde München 11

Slide 12

Slide 12 text

One sheet, two scales A A B B C C Kyoto, 1696 The same ground today "Map of Kyoto", Genroku 9 (1696) — public domain, via Wikimedia Commons | © OpenStreetMap contributors 12 • A: inside the grid • B: the southern edge of town • C: where the rivers meet • A–B is 4 km. B–C is 11 km.

Slide 13

Slide 13 text

Nineteen forty-one. Accurate maps already existed. "Nara Famous Sightseeing Place Map", 1941 — public domain 13 • The Buddha and the shrine: drawn, circled in red • The street grid: thin, uniform, anonymous • The eastern hills: bird's-eye, with autumn leaves

Slide 14

Slide 14 text

Then you try to connect them • To combine maps, you have to make them agree As drawn • Warping makes them agree — and ruins them • Labels stretch, strokes smear, emphasis dies • The reason the map existed is what you destroy 14 Warped to fit

Slide 15

Slide 15 text

Keep the drawing. Connect it anyway. Then inaccuracy stops being a weakness. The gap records the era, the purpose and the culture the map was drawn for. 15

Slide 16

Slide 16 text

You can have two of these three Keep the drawing Both directions Standard tech Warp and bake the image — refit ✓ Warp onto a basemap — refit ✓ Similarity, to show the old map ✓ poor fit ✓ Nearest-points local transform ✓ — — Maplat ✓ ✓ — "refit" = the way back is a different function, fitted separately. 16

Slide 17

Slide 17 text

A question about your own tools Is your backward transform the inverse of your forward transform? "the 'backward' … methods are in general not the exact inverse of the 'forward' … methods" A widely used georeferencing library, in its own docs "SWITCH THE 1 AND 2 EASTING AND NORTHING ARRAYS" A standard geospatial toolkit — the comment sits immediately before it computes the backward coefficients "the adjust and spline transformations give an RMS of nearly zero … does not mean that the image will be perfectly georeferenced" A commercial GIS, in its own documentation 17

Slide 18

Slide 18 text

Ours is the same function, read backwards strict • Control points build one triangulation; we reverse it for the inverse • Then we check both directions for selfintersection • If it fails, we say so — and refuse the inverse 18 error loose exact inverse, verified inverse refused degraded; round trip not exact

Slide 19

Slide 19 text

What a real inverse buys you Lines stay lines No edge of the world Two views, one truth Pin edges to a river or a Defined beyond the control Switch without dropping to road. They come out points. Your GPS can walk off an approximation. unbroken. the sheet. 19

Slide 20

Slide 20 text

Recorded demo — building one, from the same map Map image: National Archives of Japan | digital.archives.go.jp/item/697877 | Base map: © OpenStreetMap contributors (ODbL) 20 1 Already paired: old map left, modern right 2 Several hundred control points, already placed 3 The image is untouched — only a transform is saved 4 Red lines are fixed triangulation edges 5 Click a point: its match centres opposite

Slide 21

Slide 21 text

Not a demo — a delivered solution Real time Offline Transform runs client-side, at interaction speed PWA — works with no network, once the tiles are cached Eleven languages Open Built for visitors, not only for locals Apache License 2.0, five repositories on GitHub Historical GIS, cultural heritage, tourism — humanities-oriented work. 21

Slide 22

Slide 22 text

P ar t 3 Where this goes next Ongoing work, and one problem I have not solved 22

Slide 23

Slide 23 text

Ongoing — vector data, and where the product goes Today: an image • Mapbox / MapLibre basemaps: already supported • Wider vector data support: in design • A new desktop platform — Tauri, Electrobun: undecided • A SaaS edition • An MCP interface, for generative AI 23 Next: queryable features

Slide 24

Slide 24 text

The open problem — 3D • In 2D, topology preservation is guaranteed 2D — a triangle 3D — a tetrahedron sign of area tells you this one turned inside out • In 3D, the same construction stops working • Tetrahedralization can invert where triangulation cannot • I do not have an answer to this 24

Slide 25

Slide 25 text

— We would love to hear ideas from this community — github.com/code4history/Maplat | www.maplat.jp/en/ s.maplat.jp/r/hiroshimamap/