3.124 optical scanning, cont. (90)
Willard McCarty (MCCARTY@VM.EPAS.UTORONTO.CA)
Thu, 15 Jun 89 22:05:43 EDT
Humanist Discussion Group, Vol. 3, No. 124. Thursday, 15 Jun 1989.
Date: Thu, 15 Jun 89 12:19:39 EDT
From: jonathan@eleazar.dartmouth.edu (Jonathan Altman)
Subject: Re: 3.121 Kurzweil 5100 scanner (70)
[This is a very long article. If you are not interested in scanning,
you may want to skip this as it deals to some extent with the technical
side of scanning.]
I would like to point out one flaw in Terry Erdt's logic that the
ability of new OCR software to save the IMAGE of the page and work
on these saved images will allow people to save the images and
re-interpret them later with newer, better OCR software when it
becomes available.
I do NOT wish to state that the ability to save images is useless.
It is not. There is some benefit to Terry's scheme. But it is NOT
a universal panacea. It addresses the first advancement which OCR
has made, in the quality of the software written to interpret
scanned images into text. But, it does not address the second
advancement in OCR, which is the increasing level of quality of
scanned images and the resolution (number of dots-per-inch) which
new scanners can produce. By increasing the resolution which
scanners can work at, we provide more detailed information about the
images on a page. So, Calera Truescan, which is designed to deal
with 300 dpi (dot-per-inch) input is feasible only because scanning
technology allows an image with the detail of 300 dpi to be
produced and interpreted by Truescan. Let's take a state-of-the-art
image scanned, say, ten years ago. Considering the first scanners I
know of for PC's (the Thunderscan for a Macintosh) worked at 72dpi
in 1986, I'd be willing to say that in 1979 scanners offered
something less than 300 dpi resolution. The images I might have
saved with that scanner in 1979 do not have the level of detail
needed to re-interpret with OCR software to any much greater degree
of accuracy, because the newer OCR software probably depends on the
greater information which 300 dpi offers.
Will this be an issue? Isn't 300 dpi certainly high enough
resolution? Well, for those who heard Ted Brunner speak at the
Dynamic Text conference, 300 dpi is 1989 technology. And 1989
technology will be obsolete and archaic in the future just as
teletype machines were outdated once Ted got his Tektronix terminal
that could display Greek characters if one wrote the vectors for it.
And that Tektronix is certainly out of date now.
As an example,
let's take a 300 dpi image scanned with Calera Truescan now, and
save it to a file. What happens when 400 dpi scanning becomes
de-facto standard as 300 dpi is now(and this day is NOT far off.
400 dpi will be it in about 2 years I estimate, since scanners
capable of 400 dpi are already relatively common)? What happens is
that OCR software will work with 400 dpi images instead of 300 dpi.
Since scanned images are two dimensional, meaning that they have
height and width, that means that each square inch of page scanned
has a 400x400 dot image per square inch because scanner resolution
is measured in dots per linear inch. A 1 inch square of text
scanned at 400 dpi has 400X400 dots of information, or 160,000 dots
of information. At 300 dpi, there are only 90,000 dots of
information. So, scanning software that works at 400 dpi will have
over 43% (43.75%) MORE information on which to attempt to recognize
characters. And without those extra 70,000 dots of information, the
new scanning software may have trouble with what is now very
granular 300 dpi images.
Further, the concept of bitmaps themselves may be archaic soon.
Postscript and Bezier curves may be generatable from a scanner in the
near future. The fact that these images are represented as numerical
functions means that we could have OCR software that could do more
advanced numerical modelling of images based on more advanced
statistical and mathematical calculations than how well a particular
bitmap matches another bitmap in memory.
So, keeping images around, even to retouch them later, may not be as
big a win in the future as Terry feels it is. This is not to say
that it is no win at all. Keeping old scanned images of pages
around for later manipulation will have some benefit. I just do not
see the benefit being very large, without having to do a great deal
of later image-enhancement.
Jonathan Altman jonathan@eleazar.Dartmouth.edu
Database Consultant jonathan.altman@Dartmouth.edu
Dartmouth Dante Project