From masterpieces to facial recognition: ending the year with GIVE

12 Dec 2022

December is… gradually counting down to the New Year. But partying too hard is not an option for us because we need to round off our GIVE - Gecoördineerd Initiatief voor Vlaamse Erfgoeddigitalisering (Coordinated Initiative for Flemish Heritage Digitisation) project on digitisation and automatic metadata creation before the end 2023.

Not much time to read? This update in brief: we’ve decided what approach to use for facial recognition, the actual digitisation process for newspapers and glass plates is almost underway, and we’ve been on a site visit to meet with the masterpiece managers.

Pictured: Reading a newspaper at the train station, Fortepan collection, photo by Lipovits Károly, licence: CC BY-SA 3.0

We will be processing some 250 masterpieces in this project, to create high-quality digital reproductions of each one. What makes a piece of work a masterpiece? (link in Dutch).

Preliminary site visits

After selecting 40 works on paper or parchment, including municipal registers, musical scores, choir books, clippings, a diary... we get everything ready to start the actual digitisation. Project leader Lobke has travelled all across Flanders over recent months putting the different pieces of the puzzle in place and talking to heritage managers about:

  1. possible preparatory treatments and points of attention for digitisation;

  2. a suitable location for the digitisation to take place;

  3. what metadata to include;

  4. the steps towards providing access.

Then we’re planning our final site visits and preparing everything for digitisation at eleven different locations. 

Project leader Lobke on a site visit at the Cathedral of Our Lady in Antwerp, pictured with the Gulden Boek van de Kapel van het Venerabel (‘Golden Book of the Chapel of the Venerable’). © meemoo

Bonus!

This project grabbed the attention of OKRA, a magazine crammed full of art and heritage news. We caught up with them in an interview. Want to read about it?

Good progress with (gigapixel) photography and 3D scanning

  • We’ve already photographed almost half the masterpieces made from 2D materials (such as paintings, drawings and prints). The majority of these works are kept in museums – and we’re waiting until the winter months to take pictures in churches, when they are quieter and more favourable in terms of light.

  • At least one-third of the masterpieces that we’re taking gigapixel pictures of have already been captured in great detail. You can see the process in the video below, for the Panorama of Zeeland at the Plantin-Moretus Museum:

As we continue to tick things off our list, we’re also preparing to archive the digital versions and their metadata, and using the Rijksmuseum’s Image Processing Tool to help carry out quality controls.

We’ve reached a new milestone with our newspaper registration since our previous update: together with our project partner, Flanders Heritage Library, we’ve now registered over 80,000 newspapers and carried out extensive checks. It’s not long now until we can start the actual digitisation, with the project shifting up a gear last week. We’ve finalised the form (the XML file) that we’ll be using to enter the metadata for the newspapers, and taken the final preparatory steps. So we’re on course to start digitising in 2023!

The set-up is ready to go at Picturae, with a supply of newspapers patiently waiting in the cupboards for the end of the festive season...

Digitisation set-up at Picturae. © meemoo

Cupboard with newspapers to be digitised at Picturae. © meemoo

What role can machine learning play in the cultural heritage sector? We’re investigating how we can enrich metadata in a semi-automatic way and make archival content easier and quicker to find in three sub-projects (facial, speech and entity recognition). Meemoo content partners with an interest in this topic can find reports from all our working group meeting on the partner portal (link in Dutch).

Where are we at?

Following months of analysis, we took the decision in October to build the system for facial detection and recognition ourselves, building on the outcomes from our recently completed FAME project and subsequent research. This means we’ll definitely be scaling up our video content processing system, which is already under full development.

We’re choosing a supplier for the speech recognition part this month, and will be integrating the system we need in our architecture at the start of next year before transcribing the audio. We will then run named-entity recognition (NER) on these texts. More to follow.

Pictured: Bart De Wever, photo by Michiel Hendryckx, CC BY-SA 3.0, via Wikimedia Commons.

Ethical and legal aspects tightened up

You need to proceed with caution when using artificial intelligence, which is why we’re building a robust ethical and legal framework together with our partners. We’ve already set sail with Knowledge Centre Data & Society, who entered into dialogue with project stakeholders in three workshops on ethics. And we’re also working on a legal framework together with IFORI.

We’ve known that there’s a need to digitise photographic materials as well as audiovisual content for quite some time. And now we’re taking a first step with this glass plates project. Since February, our content partners have been working hard to register and package some 170,000 plates, and we recently reached the milestone of over 100,000 of them – 70% of the total!

Lantern plates, glass negatives and positives, plates with different dimensions, in colour or black-and-white… it’s a hugely diverse collection. Indeed, we have custom technical agreements in place for each glass plate variant, even though we’re aiming to keep the digitisation process as uniform and controllable as possible. And this process is getting off to a rapid start – we completed the testing phase last week!

PS. As a nice follow-up to the media briefing, we wrote an Inspiration article for the latest edition of META, which subscribers can find here (link in Dutch).

Pictured: preparations fully underway for the on-site digitisation in Ghent University’s Boekentoren (‘Book Tower’). © meemoo

Pictured: project leader Axelle in META. © meemoo

The GIVE projects have been made possible thanks to support from the European Regional Development Fund (ERDF) and are part of the Flemish Government’s ‘Resilience Recovery Plan’.

This page is loading...