Recognizing form regions
At the moment we don't have features like recognizing checkboxes, radio buttons, signatures, etc.
Can form recognizer recognize a region of the form, so we can cut that part out and put for further processing?
For example, I would like to know where on the document are the radio button questions, so that I know in which area I need to do custom processing.
Hi! For radio boxes specifically, we’ve added support for selection marks in our v2.1 preview. For region specification more broadly, this is only “started” status.
Jernej Kavka commented
I think the new update gave us a bit of the ability I'm looking for but not fully.
I'm currently investigating if I can determine where on the image checkboxes are, so I can crop them and push them into custom vision, to determine if they are checked or not. I can do this because there is text before checkboxes and I can do some statistics based on boundary boxes of OCR and relative distances between elements to determine mathematically the rough location of the checkboxes. A lot of work but if checkboxes aren't coming in the next few weeks, I'll at least have some fun doing this and let people know how I did it.
Training custom vision and then trying to detect them failed with spectacularly bad confidence levels when doing it against the entire document.
Signatures, say you have a box, where you generally have a signature and you want to crop it out. At the moment, if you have an OCR like "Signature" you can assume the start of the signature box but knowing the boundary of the signature box would be difficult.
Being able to teach, you want the raw content in "that line" or "that box" would be awesome.
Checkbox support will be coming soon.
also, could you explain a bit more about recognizing signatures?