Using very basic techniques (zone segmentation + OPTICS) I was able to organize a set of around 10^4 business documents (invoices, receipts) into hierarchy of clusters of documents of similar layout. Now, for each cluster, I would like to extract a template. A template consists of:
- labels: text boxes whose position and content (text) remains fixed
- inputs fields: text boxes whose position is fixed but content varies.
I have already OCRed my documents. As a result for each document I already have list of boxes with their position, size and content.
What would be the best approach to find labels and input fields for each cluster?
First thing that comes to my mind is to group boxes by their position, size and content (with some tolerance for position shifting and small spellig errors coming from OCR).