Has anyone found a robust solution for recognising and converting handwriting in scanned documents to text - or at least making it searchable?
I’m looking to solve a problem with septic tank inspection forms in NSW local government. They are tremendously labour intensive to process. The inspections are carried out by licensed independent contractors who all use their own forms. There is a certain amount of required information, but details like address etc. are handwritten, and the original is provided to the property owner and a secondary carbon copy provided to the council. Each form then needs to be checked for compliance, sent for enforcement action if risk factors are identified, and tagged against a property file, or as a worst case, at least have an OCRd rendition saved so that it’s searchable if there is a need to find it later. It’s massively labour intensive at the moment for no improvement in service level delivered, and it is very hard to identify non-inspectors.
Robust handwriting recognition could provide a good bottom up solution that would solve the handling problem.
I’m also talking to the legislative authority to see if there’s a top down way to fix it, but these are rural areas so web forms and those sorts of technologies aren’t really usable.
I’ve spoken to vendors who might traditionally provide this technology - abbyy, ezescan, docscorp - however most approaches require each letter to be contained in a box individually. These forms are lines on paper (I’ve attached particularly horrendous sample). We are also about to start testing the Google Vision API and have seen some preliminary announcements of Microsoft Cortex that look promising - but unavailable for a significant period of time.
Has anyone round a reliable and robust solution to handwriting recognition?!