HexRead
DeutschSample conversion

Full-text digitisation of historical prints

Full text from your scans

Searchable, reusable full text from what you have already digitised. Every word with a coordinate and a confidence value, delivered in the formats your systems expect, with contractually agreed, sample-tested accuracy.

Free sample conversionSee delivery formats

  • ALTO v4, PAGE-XML, TEI P5, JATS and BITS, every XML delivery schema-valid
  • Coordinates and a confidence value per word, reading order per page
  • Sampling-based acceptance per the DFG Practical Guidelines
  • Nothing is stored after delivery

Exploded view of a schematic page: the recognised regions, the lines, the word rectangles and the ALTO file float above the sheet as separate layers.

  1. Source
  2. Regions
  3. Lines
  4. Words
  5. ALTO

The delivery, layer by layer. Schematic page, generated in the browser.

Segmentation
Region, line, word
Per word
Coordinate and confidence
Delivery
ALTO, PAGE, TEI, JATS
Validation
every file against its schema
Acceptance
DFG guidelines
Retention
none after delivery

The run

Nine stages, one page

What happens between the scan and the file, shown on a schematic page. The order is the order of processing. The figures beside it are measured on this page, not typed.

  1. T+00

    Capture

    The scan as it arrives: TIFF, JPEG, JP2 or PDF, 300 ppi and up. Microfilm and older legacy scans too.

    Input
    TIFF, JPEG, JP2, PDF
    Handover
    link or media

  2. T+01

    Deskew

    Skew and distortion are measured and corrected. Without this step, line detection runs straight through the line.

    Skew
    measured per page
    Correction
    per page

  3. T+02

    Binarisation

    Separating type from carrier. Grain and show-through stay out; the hairlines of the type stay in.

    Carrier
    removed
    Hairlines
    kept

  4. T+03

    Column analysis

    The analysis finds the measures and the gutters between them. Miss the column and you read straight across the page.

    Measures
    3
    Head
    across all three

  5. T+04

    Reading order

    Head first, then column by column, top to bottom. It decides whether an article holds together in the catalogue or arrives in fragments.

    Regions
    4
    Articles
    split on request

  6. T+05

    Line segmentation

    Ascenders and descenders interlock between lines. The lines are separated cleanly all the same.

    Lines
    241
    Words per line
    8,1

  7. T+06

    Word coordinates

    Every word a rectangle, written in reading order. HPOS, VPOS, WIDTH and HEIGHT sit in the ALTO file exactly like this.

    Words
    1.951
    Per word
    one rectangle

  8. T+07

    Confidence

    One value per word. Whatever falls below the agreed threshold goes to post-correction; the rest does not.

    Below 0.90
    125 words
    Threshold
    agreed per project

  9. T+08

    ALTO

    String by String, schema-valid. The rectangle on the page and the element in the file are the same object.

    Schema
    ALTO v4
    Validation
    valid

The way back

In the delivery they are the same object. Hover a word on the page and its String element lights up. Hover an element and the word lights up on the page. That is what turns a scan into a searchable collection: the hit in the catalogue knows where on the page it sits.

Page, column 1·

ALTO v4, TextLine·

<TextBlock ID="block_1" HPOS="81" VPOS="700">
  <TextLine>
    <String CONTENT="..." WC="..." HPOS="..." VPOS="..." WIDTH="..." HEIGHT="..."/>
  </TextLine>
</TextBlock>

Schematic page with generated values. The structure is that of the delivery.

Buyers

Three buyers, one recognition stage

Libraries and archives

ALTO in the METS application profile, Kitodo.Production integration, delivery packages fit for long-term preservation. Full text without rights reservation.

Read more

Digitisation bureaus

The recognition stage as a white-label component in your workflow. Collection-wise tranches, confidence-driven post-correction, fixed cost per page.

Read more

Publishers with backfiles

Whole volumes from scan PDFs or existing datasets: one JATS file per article, one Crossref deposit per volume, a proof view page by page.

Read more

Delivery formats

In the formats your systems expect

One recognition yields the files your viewer, your repository and your long-term archive each expect. Every XML delivery schema-valid.

  1. ALTO v4LibrariesWord and line coordinates with confidence values, embedded as fileGrp FULLTEXT per the METS application profile.
  2. PAGE-XMLResearchRegions, reading order and line polygons per OCR-D, for Transkribus and eScriptorium workflows.
  3. TEI P5EditionsStructured text with typographic markup and a facsimile reference per line, DTABf-aligned on request.
  4. JATS and BITSPublishersArticle and book XML with per-article metadata, ready for Crossref registration.
  5. METS and MODSArchivingDelivery packages per binding delivery specifications, ready for the long-term archive.
  6. PDF/UA and EPUB 3AccessibilityAccessible outputs with correct reading order, from the same segmentation.

Choose a format to bring it forward.

All delivery formats and source material

Six delivery formats fanning out from one schematic source.

Delivery

A package you check yourself

What comes back is not a text export but a package your production team, your repository or your workflow can check without us: every file valid against its schema, every page beside its facsimile, every deviation from the source shown individually.

  • Proof view. Every page beside its facsimile, opened in a browser, no installation, no network.
  • Delivery documentation. Subject, scope, capture conventions and the validation status of every file, reproducible with the enclosed schemas.
  • Findings on the source. The typesetter's errors shown at the scan and left unchanged in the delivery. Whether to correct them is your decision.

The volume delivery in detail

delivery/

  • pruefansicht/every page beside its facsimile
  • alto/ page/ tei/ jats/by target format, every file schema-valid
  • mets.xmllibraries: the full text as fileGrp FULLTEXT, linked to the image per page
  • crossref-deposit.xmlpublishers: one deposit per volume, XSD-valid
  • lesefassung/the running text with the printed page numbers
  • faksimile/named by printed page, referenced from the full text
  • Lieferdokumentation.pdfwhat is delivered and how it was checked
  • Befunde-zur-Vorlage.pdfthe errors of the source, each at the scan

Confidence

Post-correction by threshold

One value per word means you do not correct the page, you correct the places where recognition was uncertain. Whatever lies above the agreed threshold stays untouched. The accuracy target is fixed per project in the contract and sample-tested.

6.4 % of this schematic page's words would go to post-correction

How we measure

Relief of a schematic page: the words with low confidence stand up from the sheet as columns.

Schematic page, generated in the browser. Each column is one word's uncertainty; whatever breaks through the threshold plane goes to post-correction.

Process

Four steps, an exit at each

  1. Sample conversion

    Up to 50 sample pages, free of charge, result with an error-rate measurement in 5 working days.

  2. Pilot

    A delimited collection at a fixed price, with a contractually agreed accuracy target.

  3. Production

    Agreed tranches, a fixed throughput, continuous sample checks.

  4. Acceptance

    Sampling per the DFG Practical Guidelines, error-rate report per material class.

Process, pricing and contact

A sample from your holdings

Up to 50 sample pages from your holdings, ideally your hardest volume year. No obligation, result with an error-rate measurement within 5 working days. Only then do we talk about scope and price.

Free sample conversionSee the process