A great idea. These would be super useful for reuse in typesetting my own books using LaTeX, if only these ornaments were high quality SVG's instead of low quality jpegs.
The originals seem to be inherently low quality. Possibly an interesting tech project here: try to automatically convert the JPGs into clean SVGs? Might or might not require DL like GANs.
From the project's About page [0], it seems a fair amount of ML/DL was involved in creating the database.
"We developed a program to recognise printers’ ornaments and extract them from the full page images into a new database."
"The approach adopted here is a morphological one. A series of morphological operations (e.g. filtering, dilution, erosion, etc) is applied on each image followed by a series of heuristics to filter out those connected components that are deemed to be printers ornaments."
I followed the trail to the lead software developer, an expert in machine learning and data science. In his past projects page [1], he described this project under the heading "Image Processing":
"the problem was to automatically detect and extract printers ornaments from millions of scanned 16th, 17th, and 18th century books"
If they would release the dataset of these rough-quality images of printers' ornaments, perhaps someone from the public (i.e., open source) would be daring enough to try generating SVGs. I know practically nothing about deep learning (but curious and willing to learn :), but I'm supposing this wouldn't count as "second-order machine learning"..
Edit: Further down the rabbit hole, there do seem to be some promising/interesting work in this direction, such as "Conversion of Bitmaps into Scalable Function Graphics using a Pre-trained Neural Network Database" [2].
