Christian Iconography: The Émile Mâle pipeline

From Wikipedia: "Émile Mâle 2 June 1862 – 6 October 1954) was a French art historian, one of the first to study medieval, mostly sacral French art and the influence of Eastern European iconography thereon. He was a member of the Académie française, and a director of the Académie de France à Rome."

Christian iconography is a system of visually identifying agents and situations through attributes, or icons. For example, the attribute of Saint Olaf, the Rex Perpetuus Norvegiae, is the axe, his instrument of martyrdom. Someone untutored in this system will have great difficulty interpreting a work of art or media that deploys the system of Christian iconography. Art historians (e.g. Émile Mâle) spend years, decades, exploring and learning it. But much of the relevant art is digitized, often with accompanying notes that identify the details of Christian iconography deployed in the work of art. With some ingenuity, these described digitized images could be transformed into "ground truth" datasets of tagged data—the already-existing catalog notes provide the tags, if they could be harvested as such. Such ground-truth datasets could be ideal for creating a machine learning recognition system that then could robotically provide metadata for art works that are not already so tagged. The Vatican Library has a vast digital archive. To be sure, the imagined Émile Mâle robot will never substitute for the historian, but would cut down on the historian's tedious repetitive tasks, and would be a great assistant for the student who does not happen to have a trained art historian handy. We might also explore how much recognition could be achieved by unsupervised machine learning.

Would you like to work on the computer-vision task of creating a robotic Émile Mâle?

If so, write to

and we will try to connect you with a mentor.

Related Scrolls

Related Links

More Information

The dataset collected uptil now is here.

To Do

  1. Harvest a ground truth dataset.

  2. Make classifiers that pick out the saints from the visual presentation of their attributes and icons.

  3. Make classifiers that recognize co-speech gestures (e.g. pointing) in paintings?


Progress so far is recorded here. See also