HOCR driven by Annotations

[DiegoPino]

What?

Imagine you have this:

If you blindly run OCR, Tesseract will detect each building as a non-sense word (Lot's of L and I). What if you could use annotations to mark what is worth OCRing?

I like this idea

[dmer]

I like this idea! This also seems like a pretty great candidate for a crowd-sourcing option - allow visitors to annotate/suggest areas of images that ought to be OCR'd and indexed for search.
Also a batch application capability might be handy - I've worked with plenty of digitized materials where the images always have some text/label in the same area (bottom left-hand corner for example) and if you could specify that area and apply it to all ADO's in a set...

[DiegoPino]

@dmer I wonder how we could pass the in batch mode the areas to be OCRed. What I plan to do is to extend the currently available options for the "type" of annotations to some admin managed ones. Basically new tags (e.g HOCR process) which will in turn generate an annotation like the ones we already generate that can be caught by the OCR processor to do a "crop image around the annotation -> generate OCR, merge all generated Ocrs into a single Flavor Document. Tagging @giancarlobi @alliomeria and @patdunlavey here. But it seems like users will for sure have trouble doing this upfront without the tools so a batch (e.g AMI update/ingest) would require basically documentation on how to write annotations manually OR a template that takes values and generates those for them

[DiegoPino]

Another approach I could try (not for batch but for the actual OCR process) is to take all annotations, generate a Blank Image with the same original size and position the cropped elements in their original places on the blank. Then do a single time HOCR.

[dmer]

Right - I imagine at least initially that any batch work would have to involve adding an annotation manually to the JSON - maybe through the JSON patch find and replace?