Hi
I am searching using OCR and getting a result i was not expecting. Looking at the tuner we can see that 2 instances of the same string are being interpreted differently, one having it’s space stripped out
Space removed:
Space preserved
I have no non-default settings on and cant see why the 1st image reads without the very obvious space.
many thanks
Hi Paul,
That seems peculiar to me as well, but not out of character for what I have seen before. Because I am unable to look at the underlying code, the best that I can do is offer an educated guess as to why it is happening along with a potential solution to your issue.
I think that why it is happening is related to how kerning is being interpreted by the OCR engine. I believe that the engine is expecting the capital R and capital C to be kerned more tightly, and they are not kerned at all.
If you are trying to locate “NCIR Code” in the SUT window (e.g. ImageFound, ImageLocation), I would recommend adding the ignoreSpaces property. I expect that you are trying to read the code adjacent to the field, and this solution will resolve that problem. If, however, you are trying to capture the field name, you might use validWords:"*" instead if the phrase is already expected.
Hope this helps,
Dave
Hi Dave
Thanks for that - setting the ignore spaces for that one check seemed to do the trick.
I am however seeing additional issues:
Here “feedback” is being detected as “feecback” whith the DPI at a default 72dpi
If i change the DPI to 144 it gets correctly processed:
The display is not set as anything specific in windows, but the resolution is set to 1024x768, which is a dpi of 82. Given this should i be setting all our OCR attempts to be 144, given that we cannot select a resolution-specific DPI?
many thanks
This is interesting, Paul, and I will need to do some experimenting. “In days of old,” I learned that changing the DPI was only necessary if you were establishing a connection to a device that had a higher pixel density.
Dave
Is there any update on this, as I have been noticing some inconsistencies within OCR and close fitting table cell bounds.
Not sure if there is enough to merit raising a support case / defect.
