Unable to identify simple 6 digit numbers

[Rob Shanks]

I am trying to use Tesseract to recognise a 6 digit number from scanned documents.  Because they are scanned the numbers can be faded but I know that they are 6 digit.  The scanned documents have been destroyed long ago so I am not able to get them rescanned.

I have tried whitelisting 0-9 and using a user patern file \d\d\d\d\d\d and setting psm(8) to say it is looking for one word but nothing seems to improve detection.

Does anyone have any suggestions?

Thanks

Rob

[Tom Morris]

I'd play with modifying the images by hand to see what types of operations improve performance. The first two things I'd try would be:

1. Line removal - see if cropping/removing the thing horizontal and vertical lines improves performance

2. Dilation or other image operators to "fill in" the dropouts (ie white areas) in the numerals 

Tom