Linking Page Images to Transcriptions with SVG

Hugh A. Cayless

Head, R&D Group, Carolina Digital Library and Archives

So, what's the point of this?

• Lots of chat this summer about text-image linking
• Moderate to gigantic manuscript digitization projects on my horizon
• What's the state of the tools for:
  • analysing images?
  • linking image to text?
  • displaying and manipulating the results?

Desiderata

• the process should be automated (as much as possible)
• the result should scale down as much as possible: from line to word/letter (or at least contiguous text segment)
• should support linking by URI
• should be able to be manipulated in arbitrary ways

Goals

• Create an SVG overlay of the manuscript page image
• Analyse the structure of the SVG document to detect lines, etc.
• Link the groups so produced to structures in a TEI transcription
• Display the results in a usable GUI

Method

• Turn the image into an SVG
• Organise the paths into groups
• Link the ids of the paths to ids of structures in the transcript
• Serialise the results of the analysis into a form that can be displayed and worked with

Example: papyrus

Toolchain: prep

• potrace (http://potrace.sf.net)
  • converts a bitmap into vector graphics (including SVG)
• ImageMagick (http://www.imagemagick.org/)
  • used to preprocess the image (converting JPEG to bitmap, e.g.)
• Inkscape (http://www.inkscape.org/)
  • SVG graphical editor, used from the command line to process the SVG produced by potrace
• XSLT (used for cleanup)

papyrus image as SVG

Toolchain: analysis

• Python
  • lxml - ElementTree
  • numpy
• the script reads in the SVG produced by potrace and filtered through Inkscape, does some filtering, detects the lines, then serializes the results back to SVG and Javascript

Example: analysed SVG

Toolchain: display

• Javascript:
  • OpenLayers
    • hacked (I use the term advisedly) to accomodate paths and groups of paths
  • jQuery
• XSLT (to turn the TEI transcription into HTML)

Example

What's next?

• Lots of questions:
  • How much can be automatic?
  • How deeply can we analyse?
  • How best to link?
• Need to refine and test on lots more content

Automation

• Can we compute the right black/white cutoff point?
• How much, and what kind of image preprocessing should be done?
• What's the best way of throwing out extraneous paths?
• Can the linking be automated?

Analysis

• Gosh, I've made rectangles around lines.
• Structure is a problem:
  

Linking

• The demo version uses JSON, basically a pair of hash maps, id -> list of ids
• TEI has ways of doing this, but is that the right place for it?
• linkbases?

Conclusions

• Method seems promising
• Tools are in an advanced enough state to do really useful work
• References:
  • P. Mich. 1.78 at Perseus
  • P. Mich. 1.78 in APIS
  • Letter from John and Ebenezer Pettigrew to Charles Pettigrew at Documenting the American South