[bksvol-discuss] Re: Question on Scan quality for blind vs other types of disabled bookshare readers
- From: Guido Corona <guidoc@xxxxxxxxxx>
- To: bksvol-discuss@xxxxxxxxxxxxx
- Date: Fri, 8 Feb 2008 14:27:50 -0600
Evan, what version of Kurzweil do you use, which engine?
G.
Guido Dante Corona
IBM Research,
Human Ability & Accessibility Center, (HA&AC)
Austin Tx.
Phone: 512. 838. 9735.
Email: guidoc@xxxxxxxxxxx
Web: http://www.ibm.com/able
". . . Maybe it was only those who were most certain they were right who
were guaranteed to be wrong. And that maybe, just maybe, those who
questioned the most were in the end those who came closest to being wise."
[David Poyer, The Command]
"EVAN REESE" <mentat3@xxxxxxxxxxx>
Sent by: bksvol-discuss-bounce@xxxxxxxxxxxxx
02/07/2008 04:10 PM
Please respond to
bksvol-discuss@xxxxxxxxxxxxx
To
<bksvol-discuss@xxxxxxxxxxxxx>
cc
Subject
[bksvol-discuss] Re: Question on Scan quality for blind vs other types of
disabled bookshare readers
I also agree with this.
Besides, when Pratik advocates not changing fonts in the book, he is also
making a rather large, unstated assumption, namely that OCR software is
accurately recognizing font style, type and size. This may be a valid
assumption in many cases, but in many others, it is clearly not valid.
I've seen - and heard of - examples of K1000 changing its mind about the
font size right in the middle of text, or after a rescanning of the same
text with different settings. How could I possibly know which is the more
accurate rendering? And what should I do if I am told that many of my
page numbers are a size 2 font? which happened with one of my first scans
with K1000. (I actually posted about this, and was told that this is an
unreadably small font size, and I could at least tell with my Optacon that
the page numbers were not really that small. Should I have left them as
they were?) I often get page numbers being recognized at different sizes
on adjacent pages when they are clearly the same size in the book. The
same with the size of the font with other text, if I should close a scan
session before finishing the book, or change scanner or OCR settings.
I'm sure many others could amplify these examples by a very large number.
In fact, this thread was begun by a sighted person, pointing out
readability problems that were
almost certainly caused by OCR Error.
So leaving the book as is with respect to font is a great idea in theory,
but it assumes that the OCR software renders the book as it really
appears, which it often does not. So I can do one of two things: I can
either change the book to one font size, except for things like section
headings, which may improve readability for sighted persons and may also
very well return the book to something more like it's original appearance,
or I can let the chips fall where they may when it comes to font
recognition and hope for the best. In many cases, probably, doing the
latter will result in accurate recognition of font size and style, making
for a good reading experience for sighted readers. But in many other
cases, it may well make reading more difficult for sighted readers for
reasons that have little or nothing to do with the appearance of the
printed book but are due to misrecognition of the font size or style. I
tend to do the latter anyway, just because I read through every word of
each book I submit or validate and just don't want to spend the time
bothering with fonts after having spent all that time trying to make sure
that the text is first rate. But I have done the former on at least one
occasion, and have increasing sympathy for doing it in the future, as I
have become more aware of how sometimes apparently whimsically changeable
font recognition seems to be. But I - like Monica - am not going to check
every font line by line and fix it. I read through my submissions and
validations on a braille display, and don't get that info on the Pac Mate
without manual checking, and have no desire to read them through again
with font messages enabled with Speech on the PC, just to make sure every
font is properly rendered. And besides, having no vision and only the use
of an Optacon - something most people with no vision do not have access to
- except in cases like the absurdly small font size I mentioned, how could
I possibly know if the font I get from K1000 is really the one in the
book? If the font size suddenly changes by a few points for no apparent
reason, which it sometimes does, which should I take as the one that is
really in the book?
In the case of the absurdly small page numbers, I did fix those. I went
through this 500 plus page book and checked the font of every page number
by hand and made the very small ones a size 9 or 10 or something like
that. There weren't a huge number of size 2 font page numbers, but how
could I know if I didn't check them all? (I hadn't thought of selecting
the whole book and making it all one reasonable font size at that time.)
But I have never done that kind of checking or fixing again, and don't
plan to any time soon. I will either leave them entirely alone - except
for section headings - and hope for the best, or make them all one size -
except for section headings.
All of the above applies to formatting as well, which also can create
readability problems as I believe Judy mentioned in her original message.
OCR software - including K1000 - does not always accurately render the
formatting of a book, and sometimes makes changes for no apparent reason -
which, I believe someone else mentioned here. In addition to occasionally
fixing fonts, I have also left justified the text in at least one or two
books that I can recall, except for section headings, which I centered if
they were centered in the book. Or I just let the formatting come out as
the OCR software renders it, and again, hope for the best. But here again,
I am in increasing sympathy with the notion of left justifying everything,
except for section headings, and making some reasonable indent for the
paragraphs, at least for fiction, which is usually just paragraphs of text
anyway; but I have not made a habit of this kind of changing as of yet.
However, once again, leaving the formatting as it is may also detract from
the readability of a book by sighted readers, not because of the
formatting of the original book, but because of inaccuracy of the
recognition of that formatting by the software, or of its changing its
mind as to what that formatting is for no apparent reason.
I recognize that what to do about these kinds of issues is something that
reasonable people can disagree on. But merely saying that books should be
left as they are is not adequate if they are not necessarily rendered as
they are by even the best OCR software available.
Evan
----- Original Message -----
From: Jamie Yates, CPhT
To: bksvol-discuss@xxxxxxxxxxxxx
Sent: Wednesday, February 06, 2008 10:36 PM
Subject: [bksvol-discuss] Re: Question on Scan quality for blind vs other
types of disabled bookshare readers
Lora, I agree with you. And here is the thing, I can see. So I know that
the font might be different than the font in the book. The majority of
volunteers are blind. How do they know if what they scanned isn't
identical to the print book? So how can it matter so much? Because I can
see, it is wrong for me to KNOWINGLY change the font of a book I scan or
validate. But if I couldn't see, it wouldn't be wrong for me to
UNKNOWINGLY change the font or size.
Right?
But, isn't the whole point of bookshare to make books ACCESSIBLE to
everybody? In a large print book the publisher changes the font size. In
an audio book the publisher changes many things about the book including
eliminating headers and page numbers. It seems to me that Bookshare should
have some leniency here. But it also seems a simple solution for the
sighted print disabled Bookshare member would be for that member to change
the font size in his or her own digital copy of the book to a workable
font for that user.
Jamie in Michigan
Currently Reading - When death comes stealing : a Tamara Hayle mystery /
Valerie Wilson Wesley
I'm an eBay affiliate, click here before you bid!
Click here for eBay!
Other related posts:
