There is already a "tickytacky" extractor in dockerhub that supports this workflow by identifying the significant vertical and horizontal lines in an image. A clustering algorithm running after the extractions can group images into like forms. This is a workflow that I'm developing in DRAS-TIC/Clowder combination.

I've tested clustering locally with image files, but now working to implement a work flow in DRAS-TIC CI-BER cluster. Loading a collection from NARA, running extractor, trying various clustering approaches, both parallel (Spark over DRAS-TIC) and as a separate process (export metrics for the job).

This task is to build a test collection for this workflow, using NARA image collections. Specifically, we will use WWII records collections related to the restitution of seized artworks.