Letter about cut-canceled dies - San Francisco

[RWB]

.

 

[RWB]

.

[thebeav]

That's interesting. I've always had multiple errors in scanning OCR and I just thought the process is what it is. I didn't know the software was tuneable.

I can't talk now, I've got a 1900 Liberty nickel die I'm trying to repair. I'm going to over-run the world with 1900 nickels !!!

 

Paul

[bstrauss3]

Great job! You don't know how much OCR has improved until you compare 10 years ago (99% or 1 error per 100 characters) vs. today with several 9s... but it always helps to tweak it to the specific input.

[RWB]

.

[jgenn]

The OCR software was modified to better recognize manual typewriter text, and to shrink the "halo" surrounding carbon copy and press copy typescript. Some of the subroutines come from astronomy software toolkits. These are very good at separating real from spurious image forms. Another aspect of the SW is comparison of image forms with restricted character sets.

 

Ingenious! Is Tesseract the OCR software?

[RWB]

.

[jgenn]

Thanks for that information. I often work with non-Latin OCRed text that has an accuracy of about 90%. Is the source for any of that old SW posted on-line?

 

And more importantly, I think your application of this technology for searching old mint records is fascinating! Please keep posting your new insights.

[RWB]

.

[RWB]

.