Casting GCP image coords to reflect the source image, not the split image --> IIIF support

[mradamcox]

From the outset, the georeferencing workflow was designed to be performed on individual files. These could be original scans of full pages, or if a page had been split, the smaller files that resulted from the splitting process. This means that all ground control point definitions have been created with pixel references to each individual file.

The issue is that this does not exactly hold up when trying to tie GCPs on split pieces back to the original image from which the split was generated. This is because the splitting process is powered by "divisions", which can be any shape. The division is then applied to the original image, all content outside of it erased, and then the output is cropped to a rectangle that encloses the division. GCP image coordinates are then made based on this cropped rectangle.

This discrepancy is currently the only theoretically blocker for the conversion of all existing data in the database into annotations that are compliant with the IIIF Georeference Extension spec, which is a long-term goal.

According to the spec, there are two ways to select sub regions from a full resource to be targeted in the transformation, 1) by using an SVG selector (passing a shape like a mask) or 2) by using the Image API Selector. It's likely the SVG selector will be the better fit for these regions, because the Image API Selector does not allow irregular shapes, and in that case, the original, full image pixel coordinates will be needed, not the image coords from within the selected region.

More links:

• https://iiif.io/api/registry/selectors/
• https://iiif.io/api/cookbook/recipe/0261-non-rectangular-commenting/
• https://iiif.io/api/image/3.0/#41-region