Transcrire d'anciens recensements avec l'IA

500 000+

Utilisateurs dans le monde

200 M+

Pages traitées

300+

Modèles IA publics

100+

Langues et écritures

Le problème

Why old census handwriting defeats standard OCR

Census records are structured as tables, but enumerators filled them by hand — often in a rush, going door to door. Every enumerator had a different hand, and abbreviations varied by district, decade, and country. US Federal Census sheets from 1790 to 1950 use cramped columns with faded pencil or iron gall ink. UK Census returns from 1841 to 1921 pack entire households into tight rows. Standard OCR sees a grid of noise, not the names, ages, and occupations your research depends on.

Enumerator handwriting varies wildly — each census district has a different hand

Tabular layouts with narrow columns confuse standard OCR text-flow detection

Faded pencil entries, ink bleed-through, and microfilm artefacts on nearly every page

Heavy use of abbreviations: 'do.' for ditto, occupation shorthand, birthplace codes

Name spellings are phonetic — the same surname appears three different ways on one page

Example of difficult census record handwriting

La solution

AI that reads census record handwriting across countries and centuries

Transkribus uses handwritten text recognition (HTR) trained on millions of historical document samples — including census records. Unlike standard OCR, it understands connected cursive, tabular layouts, and the abbreviation patterns census enumerators actually used. Upload a scan or phone photo of your census page, select a model, and get structured text back in minutes. The AI detects columns, reads across rows, and preserves the table structure that makes census data usable.

300+ public models — including models trained on census-era handwriting styles

Table detection recognises column headers, row boundaries, and cell content automatically

Handles faded pencil, microfilm scans, and low-contrast images that defeat standard OCR

Confidence scores on every line — see exactly which entries need your review

Export as plain text, CSV, Excel, searchable PDF, or structured XML

Comment fonctionne l'OCR manuscrite

Transkribus transcription editor showing a census record

How to transcribe census records in 4 steps

Upload your census page

Take a photo with your phone or upload a scan. Transkribus accepts JPG, PNG, PDF, and TIFF — including multi-page census schedules.

Sélectionner un modèle IA

Choose from 300+ public models. Filter by language, century, and script type to find the best match for your census record's era and region.

Lancer la reconnaissance de texte

Click 'Recognise' and let the AI work. It detects the table structure, reads each cell, and processes the full page in about 30 seconds.

Vérifier et exporter

Check the transcription against the original image. Fix any misread entries in the built-in editor, then export as text, CSV, Excel, PDF, or XML.

Models for census records

AI models for historical census digitization and enumeration transcription

The Transkribus public model catalog includes models suited to the handwriting styles and tabular formats found in census records from multiple countries and centuries. Many were trained by archivists and genealogists on real census collections, so they handle the specific abbreviations, column layouts, and script variations that generic tools miss.

US Federal Census: models covering 19th and early 20th-century American enumerator handwriting

UK Census: models for Victorian and Edwardian-era enumeration book entries (1841-1921)

Canadian Census: models for bilingual English/French population schedules

European census records: models for German, Austrian, Scandinavian, and Dutch population registers

New models added by the community regularly — covering more regions and time periods

Browse public models for census records

Transkribus model catalog filtered for census record models

Questions fréquemment posées

Vos documents, vos données, votre contrôle.

Transkribus est développé et hébergé en Europe par une coopérative de plus de 250 institutions. Vos données restent les vôtres.

Propriété totale des données

Vos documents et transcriptions vous appartiennent. Suppression à tout moment.

Hébergé en Europe

Traitement sur nos propres serveurs en Autriche. Conforme au RGPD. Aucune dépendance Big Tech.

Construit pour durer

Une coopérative, pas une startup. Plus de 250 copropriétaires garantissent la pérennité de Transkribus.

Ready to unlock your census records?

Create a free account and start transcribing census pages today. 50 free credits every month — no credit card required.

Commencer gratuitement Voir les offres et tarifs

50 crédits gratuits par mois — Aucune carte bancaire requise

200 M+Pages traitées

500 000+Utilisateurs dans le monde

300+Modèles IA publics

Transcrire d'anciens recensements avec la reconnaissance d'écriture par IA