Collaborators

This activity is part of the FUSION project .

It is a collaboration with:

• Elodie De Zutter (head of photo collection, KIK-IRPA)
• Joris Pockelé (master student in computer science, UGent)

Description

Before digital photography became mainstream, photographs were commonly captured on sheet film. Many manufacturers included notch codes on their films: small notches cut into the edge of the film whose number, shape and position along the short side identify the film brand and emulsion type. They were originally designed as a tactile aid for use in the darkroom.

The Royal Institute for Cultural Heritage (KIK-IRPA) maintains a photographic archive of over one million negatives, a portion of which has been digitised at high resolution. Determining the film type for all negatives through manual inspection is practically impossible. Since the notch code is visible in the scanned image, it opens the possibility of automating film identification at scale, enabling the collection to be characterised by substrate type, which is essential for prioritising conservation and assessing storage risks (nitrate film, for instance, being particularly flammable).

In this project, we develop a computer vision pipeline to automatically detect and classify notch codes in digitised sheet film scans. The pipeline combines object detection to localise notch candidates along the film border and deep feature-based classification to distinguish true notches from border damage and other artefacts, before mapping the detected pattern to a known film type using a reference chart.