Make app to use its own reading-order algorithm

[keighrim]

New Feature Summary

While evaluating the app in clamsproject/aapb-evaluations#33, we realized that passing cropped images from existing BoundingBox annotation doesn't help the overall recognition performance because of lack of functional "reading order" component. Hence I'm suggesting to alter the app behavior (or add an optional behavior) to use the union of all the BoundingBox annotations and pass a single crop that includes all the boxes.

Related

No response

Alternatives

No response

Additional context

Additional factor to consider: when we evaluated this app (clamsproject/aapb-evaluations#33), the BoundingBox annotations used in the evaluation data was from https://github.com/clamsproject/app-east-textdetection. However, as already raised clamsproject/app-east-textdetection#1, the boxes have their own timePoint properties instead of being anchored (via Alignment annotations) to separate TimePoint annotations. When we make a union of such boxes, it might not be very clear how we should create Alignment annotations between the newly created TextDocument annotation and existing BoundingBox objects.