Automatic Segmentation of Receipts

I want to build a model to extract (item, price) pairs from pictures of receipts. There are two structure prediction problems that need to be solved. The first would be segmenting the part of the receipt that contains the item list from the part that has irrelevant information. The second problem would consider the relevant segment, and find the (item, price) mappings.

Comment from William

So, not neuroscience? I'm still checking, I think Tom Mitchell's group has some usable structure-prediction data, but I haven't nailed him down. At any rate, this seems like a nice problem as well, if you can find suitable data for it. I know Brendan has talked to you about it, make sure you stay in touch with one of us as the project evolves.

--Wcohen 20:51, 22 September 2011 (UTC)