[Mageia-discuss] Reading payment forms with a scanner

[Juergen Harms]

A positive remark about swiss banks! But errorproofness for swiss 
payment slips is also quite relative: (a) the long reference digit 
chains only have a very simple longitudinal 1-digit checksum that 
catches many, but not all errors that happen if you key them in manually 
and (b) all the OCR pograms that I have tested are far from perfect. And 
yes, for some time I tried to use one of these portable devices - they 
accumulate limited quality of scanning with limited quality of OCR 
conversion: between handling and checking I spent more time than I need 
for manual input and verification.

My recent error scenario illustrates this: my incorrectly keyed in digit 
string happened to not trigger a checksum error, hence was accepted by 
the ebanking system of my bank, but was then rejected by the destination 
bank because the string did not make sense to them.

In the meantime, I have hacked a kind of tool which both shortens the 
time I need to prepare my payments and diminuishes errorproneness,
- it selects the area to read and adjusts parameters for xsane,
- increases the output of xsane by a factor of 1.5 (amazing how that 
reduces the errors of OCR conversion),
- submits the result to tesseract,
- does some string processing,
- displays the corrected and the original strings (see attachment - bad 
quality to limit the size of this email), ready for copy paste.

Looks very promising (and can be improved, I only spent half an 
afternoon) - needs now to be assessed by really using it in 
"production". Talking about generalisation - I see your arguments - does 
only make sense once that provides positive results.

Generalisation: maybe the concepts I put to work allow to create similar 
tools / adjust the one I have made. I live part of my time in Austria - 
Austrian slips should not be too difficult to handle along such lines. 
And, maybe the mandatory introduction of IBAN input will create an 
incentive to banks to improve the readability of their slips.

Juergen
-------------- next part --------------
A non-text attachment was scrubbed...
Name: convert.jpeg
Type: image/jpeg
Size: 14108 bytes
Desc: not available
URL: </pipermail/mageia-discuss/attachments/20130215/36752468/attachment-0001.jpeg>