[bksvol-discuss] Re: text quality
- From: Guido Corona <guidoc@xxxxxxxxxx>
- To: bksvol-discuss@xxxxxxxxxxxxx
- Date: Thu, 29 Apr 2004 15:51:00 -0500
Yes, the ranked spelling feature will yield a spelling accuracy rating
similar to the error rate used by Bookshare. The recognition stats simply
indicate how tedious it was for the OCR engine to complete the job.
Guido
Guido D. Corona
IBM Accessibility Center, Austin Tx.
IBM Research,
Phone: (512) 838-9735
Email: guidoc@xxxxxxxxxxx
Visit my weekly Accessibility WebLog at:
http://www-3.ibm.com/able/weblog/corona_weblog.html
"Pratik Patel" <pratikp1@xxxxxxxxx>
Sent by: bksvol-discuss-bounce@xxxxxxxxxxxxx
04/29/2004 03:41 PM
Please respond to
bksvol-discuss
To
<bksvol-discuss@xxxxxxxxxxxxx>
cc
Subject
[bksvol-discuss] Re: text quality
Paul and Noel,
Since you've both expressed the same questions, let me ask something in
return. How do you characterize your results. Do you generally gage the
accuracy of your results via the self-reported recognition stats or by
looking at amount of spelling mistakes per page when you change various
settings. The recognition stats presented by Kurzweil are often
misleading. What is given to us often done by getting it from the OCR
engine in question. With those types of self reports, there is always a
matter of accuracy and reliability. There is also a matter of validity. I
have experimented with this issue a bit and have found that even with the
same settings, if you keep on scanning a page a few times the supplied
stats are different each time, in some cases by a large margin. However,
when this issue is looked at from the actual accuracy perspective, it's
quite reliable to look at the page from different scans with same
settings. Even when the Optimize Scanning feature comes up with different
settings for the same page when using that function several times, the
accuracy is not effected too often. The gray scale with 400DPI does make
a large difference. Even when I use optimize scanning, I make sure that
at the end I compare results by using the Gray scale with the 400 DPI.
Pratik
Pratik Patel
Managing Director
CUNY Assistive Technology Services
the City University of New York
(718) 997-3775
ppatel@xxxxxx
-----Original Message-----
From: bksvol-discuss-bounce@xxxxxxxxxxxxx
[mailto:bksvol-discuss-bounce@xxxxxxxxxxxxx] On Behalf Of Edwards, Paul
Sent: Thursday, April 29, 2004 4:25 PM
To: bksvol-discuss@xxxxxxxxxxxxx
Subject: [bksvol-discuss] Re: text quality
Actually, I knew that. So of course, the question becomes: if it makes no
difference, why do different values turn up when you do "optimize scan"?
Perhaps there is a Kurzweil guru lurking. Would you care to emerge from
the lurk and answer the question?
Paul
Paul Edwards, Director
Access Services, North Campus
Phone: (305) 237-1146
Fax: (305-237-1831
TTY: (305) 237-1413
Email: pedwards@xxxxxxxx
home email: edwpaul@xxxxxxxxxxx
-----Original Message-----
From: Guido Corona [mailto:guidoc@xxxxxxxxxx]
Sent: Thursday, April 29, 2004 4:02 PM
To: bksvol-discuss@xxxxxxxxxxxxx
Subject: [bksvol-discuss] Re: text quality
Paul, I also use grayscale at 400 DPI most of the time with Kurzweil 8.0.
If you find it is rather slow, scan images only, then turn on pure
recognition before going to sleep. Your book will be ready when you wake
in the morning, no matter how large it is. By the way, with grayscale
brightness makes no difference.
Guido
Guido D. Corona
IBM Accessibility Center, Austin Tx.
IBM Research,
Phone: (512) 838-9735
Email: guidoc@xxxxxxxxxxx
Visit my weekly Accessibility WebLog at:
http://www-3.ibm.com/able/weblog/corona_weblog.html
"Edwards, Paul" <pedwards@xxxxxxxx>
Sent by: bksvol-discuss-bounce@xxxxxxxxxxxxx
04/29/2004 02:42 PM
Please respond to
bksvol-discuss
To
<bksvol-discuss@xxxxxxxxxxxxx>
cc
Subject
[bksvol-discuss] Re: text quality
This is a difficult issue. I take the approach of carefully checking the
first few pages at the beginning of a scan. If there are errors I can
adjust for, I do that. I also rescan pages whose value in Kurzweil comes
back lower than ninety. I do not tend to scan ninety to ninety-five
because I can usually not make much of a difference and we are often
dealing with a screwed over heading or something.
However, I scanned a book recently which was a hard cover and which should
have scanned like a dream and came out as pure druck.
I have found that optimizing scanning is, for the most part, worth doing.
The results do not always make me happy in that I am now scanning a book
using gray scale and sixty which takes forever to scan. By the way, it is
legends two edited by Robert Silverberg.
Paul
Paul Edwards, Director
Access Services, North Campus
Phone: (305) 237-1146
Fax: (305-237-1831
TTY: (305) 237-1413
Email: pedwards@xxxxxxxx
home email: edwpaul@xxxxxxxxxxx
-----Original Message-----
From: Kellie Hartmann [mailto:kellhart@xxxxxxxxxx]
Sent: Thursday, April 29, 2004 1:04 AM
To: bksvol-discuss@xxxxxxxxxxxxx
Subject: [bksvol-discuss] text quality
Hi all.
Even with the wonderful new scanning software available there are a few
kinds of things that are very difficult to get a good scan from. For
example, linguistics books are often very graphical in nature and contain
symbols that the OCR packages don't recognize; things like r-underring and
turned V etc. Also some cheap paperbacks do have places where they seem to
be blurred. I scanned a novel that I was assigned to read in French class,
and when I found illegible passages I tried rescanning them. I rescanned
several times changing various settings, but certain passages absolutely
refused to scan. I don't really plan to submit it to Bookshare anyway, but
I
would prefer this scan, with a couple of blurred lines every 20 pages or
so,
to no scan. I'm able to use this in class with no problems, so in my
opinion
this is far better than nothing. Finally, I have another French book which
has very glossy pages and lots of flashy graphical design. Again, even
with
a lot of work on experimenting with different settings my results were not
encouraging. This I definitely won't submit to Bookshare because I can't
get
it in good enough shape; the effort required would be far beyond the
benefits. I agree that careless scanning is unreasonable, and think that
validating is important. It always takes me much longer to validate
something than to scan it because I read the whole book and fix every
error
that can possibly be fixed. Not every validator is going to do that, and
certain books, such as enormous textbooks, really would require a great
investment in time to proof thoroughly. So it isn't realistic to expect
every book to be flawless. What I would really like eventually, and I know
this isn't realistic either, would be to have all the fair-quality books
rescanned.
Kellie
Other related posts:
