EANYCODE_CHAR . confidence > 100?

25 views
Skip to first unread message

Albert

unread,
Jan 29, 2009, 1:19:55 PM1/29/09
to tesseract-ocr
Hi,

I have found that some EANYCODE_CHAR.confidence's are over 100. Does
that make sense?


Albert

Todd

unread,
Mar 9, 2009, 3:59:27 PM3/9/09
to tesseract-ocr
Albert,

It doesn't look like you received an answer here. Were you ever able
to figure this out?

In ocrclass.h, EANYCODE_CHAR.confidence is defined as "0=perfect,
100=reject (0/100)".

Thanks,
Todd

On Jan 29, 12:19 pm, Albert <a...@snowbound.com> wrote:
> Hi,
>
> I have found that some EANYCODE_CHAR.confidence'sare over 100.  Does
> that make sense?
>
> Albert

Ray Smith

unread,
Mar 9, 2009, 11:41:06 PM3/9/09
to tesser...@googlegroups.com
Thanks for bringing this back to the top of my inbox. The scaling of the certainty to create the confidence in EANYCODE_CHAR is inappropriate. The value can easily get over 100, but is clipped to 255. In fact the scaling is so far out that the 0.035 in output.cpp would be better off being 0.19. That would bring down the values a bit and make the 100 level more like a reject.
Ray.
Reply all
Reply to author
Forward
0 new messages