[Mageia-discuss] Reading payment forms with a scanner

[Morgan Leijström]

With the risk of sounding grumpy i suggest to hammer your payment provider to 
fix their bug.

I think no solution that include scanning is less time consuming than proof 
read manually typed number.
Besises, you need to proof read the scan anyway ;)

( Apart from that it would be interesting from a technical point to get OCR 
working nicely ;)  )

/Morgan

torsdagen den 14 februari 2013 09.21.36 skrev  Juergen Harms:
> I just had a nasty experience with an ebanking bill that got rejected
> (without sending me a corresponding note) due to a typo.
> 
> Does anybody have experience/advice on using a normal scanner for
> reading the essential fields of payment forms to make ebanking more
> efficient and less error-prone?
> 
> I just did some googling and quick checks along the following lines:
> - tile the payment forms on the scanner (I have an Epson 1260), so that
> only the reading zone at the bottom is visible of each,
> - scan with xsane (selecting adequate settings - different from those I
> ordinarily use)
> - if necessary us gimp to cut away zones with garbage that upset the OCR
> conversion (i.e. tesseract, can be avoided by properly setting the
> reading area in xsane)
> - use tesseract to do OCR
> - filter the output to throw away garbage lines, and to correct
> characters that frequently get mis-interpreted (e.g. B->8, Z->2, O->0,
> D->0 etc.)
> - output the data thus produced, formatted for copy-paste into the
> ebanking form
> 
> That works surprisingly well, but is excessivly complicated to handle
> (easy to make handling mistakes, not fit to give it to my wife). Are
> there tools that help automating these steps and integrating them into a
> single tool? - if not, it should not be too difficult to do some
> scripting (but I dont want to re-invent things). (And yes, I had tried
> some years ago these small reading sticks that you slide over the form -
> I ditched it: only works on windows, and produces an excessive amount of
> errors).
> 
> Juergen

-- 
Morgan Leijström