Ask HN: Paper form OCR options

4 points by peatmoss ↗ HN
Hi all,

I'm inheriting a process where people record numbers by hand in a grid and those numbers are later manually entered in on a computer. I'd like to figure out how to automate this to the greatest degree possible.

On the one hand, I could potentially farm this process out to Mechanical Turk. But, I feel like someone has probably developed some sort of paper form OCR app for this sort of purpose. I can even imagine some sort of clever number encoding / bubble filling sort of thing to reduce the error.

Does anyone know of such a thing, or better yet, have personal experience with a particular product? I feel like this is a machine learning business application that someone must have already capitalized on.

EDIT: The ideal workflow would involve snapping a picture of the filled form with a smart phone, perhaps with remediation of low confidence numbers right on the device.

2 comments

[ 5.0 ms ] story [ 18.8 ms ] thread
You could do this quite easily with Amazon rekognition or Google image APIs.

Both I believe give you the position of text in the page.

If you place 3-4 known words on a page you can determine what manual input fields other text relates to.

I have been playing around with this for paper order forms as I sell to an industry which is sometimes somewhat backwards...

I've been working on a prototype that does exactly this. Sent you an email.