Since Recovered Light progressed through some 115 stained glass panels, many of surprising compositional complexity, we needed code that could automatically segment the constituent parts of the images.

For example, given an image like this (itself a detail of the over-all stained glass panel), how could our code make sense of its parts?

 

Boundaries

First, it could fashion one kind of network of boundaries that it discovered in the image.

 

Centers

Then it could make a second kind of network by joining the centers of bounded areas.

 

Recomposition

These two kinds of analyses allowed the code to select and transform given parts of the image, as in the examples above. And each time the artwork encountered a given panel, it could choose from among many such parts as it partially recomposed the whole.