3 Dec
2019
3 Dec
'19
7:15 a.m.
From: Guy Dunphy
JBIG2 .. introduces so many actual factual errors
(typically
substituted letters and numbers)
It's probably worth noting that there are often errors _in the original
documents_, too - so even a perfect image doesn't guarantee no errors.
The most recent one (of many) which I found (although I only had a PDF to
work from, so maybe it's a 'scanning induced error') is described at the
bottom here:
https://gunkies.org/wiki/KS10
Although looking again at the PDF, the two digits in question are quite clear
and crisp, and don't seem like they could be scanning errors.
Noel
3 Dec
3 Dec
12:22 p.m.
> JBIG2 .. introduces so many actual factual
errors (typically
> substituted letters and numbers)
On Tue, 3 Dec 2019, Noel Chiappa via
cctalk wrote:
It's probably worth noting that there are often
errors _in the original
documents_, too - so even a perfect image doesn't guarantee no errors.
. . . and how often will the randomizing corruption actually result in
changing an error to what it SHOULD HAVE BEEN? :-)
Although looking again at the PDF, the two digits in
question are quite
clear and crisp, and don't seem like they could be scanning errors.
Watch out. PDF with OCR can show you a clear and crisp [possibly wrong]
interpretation of the scan, not what the actual scan looked like.
3:46 p.m.
On 03/12/2019 20:22, Fred Cisin via cctalk wrote:
Watch out.? PDF with OCR can show you a clear and crisp? [possibly
wrong] interpretation of the scan, not what the actual scan looked like.
The OCR may well say "0" where the printing says "8" but what your
eyes
will see will be the representation of the printing. So if you rely only
on OCR you may well miss something, but if you fall back to the way
you'd have towork without OCR (or even the way you'd have to work if you
had the original paper copy) then you have to rely on your eyesight to
fail to find what you are looking for ...
Unless, that is, you discard the graphical representation and keep only
the OCR result. In which case all bets are off.
Antonio
--
Antonio Carlini
antonio at acarlini.com
1851
days inactive
1851
days old
2 comments
3 participants
participants (3)
-
a.carlini＠ntlworld.com -
cisin＠xenosoft.com -
jnc＠mercury.lcs.mit.edu