I’ve said in the past that I thought OCR was pretty ridiculous for receipts as it isn’t 100% accurate and receipts have so many wacky formats. Well, after a lot of work and evaluation, I have managed to integrate the Tesseract OCR engine into ReceiptWallet. It is an open source engine that has OK results for the price. (I did investigate some commercial engines; there are 2 that are on the Mac, and the costs were outrageous and currently one doesn’t even work on PowerPC machines.) As Google continues to work on the engine, I’ll be integrating the new versions into ReceiptWallet.
So, how did I manage to get things to work and get OK results? How about I just say magic? :-). Turns out the hardest part is attempting to recognize merchant names; I cheated on this and simply recognize merchants that the user has already used. This works quite well as I find that I keep going to the same merchants, so while it won’t recognize a merchant the first time you visit it, it has half a chance on subsequent visits.
I’ve been using my ScanSnap for testing and it produces very clear images that the OCR engine seems to like. My DocketPORT, however, has had mixed results. I think I need to change the contrast/brightness settings for better results. I’ll leave this up to beta testers to try out.
My goal is to push out a beta this weekend and see what happens. There is no way that this will be 100% accurate, but it might help a little and then I get to say that it has OCR!