Uses for Screen OCR Technology ???

[Peter Olcott]

I have technology that is similar to OCR technology except that it determines
rather than estimates its results. The main advantage over conventional OCR
technology is that it is much more accurate with very small fonts. Current
testing indicates that it can achieve 100% accuracy with fonts as small as 6
point. It also works with both types of font smoothing.

I have thought of a few different uses for this technology, I want to see if
anyone here can come up with any more uses.

 

[Oliver Wong]

I have technology that is similar to OCR technology except that it
determines
rather than estimates its results.

The difference being that the former is an exact value with certainty,
and the latter is a guess or approximation?

The main advantage over conventional OCR
technology is that it is much more accurate with very small fonts. Current
testing indicates that it can achieve 100% accuracy with fonts as small as
6
point.

Does it work with the fonts "WingDing", "Symbols", or my specially
crafted "all characters are invisible" font?

It also works with both types of font smoothing.

Are there only 2 types?

I have thought of a few different uses for this technology, I want to see
if
anyone here can come up with any more uses.

Might help if you posted the uses you came up with, so we'll know which
ones are "more", and which ones are "already thought of those".

Might be useful for turning a scanned image of a document into a text
file.

- Oliver

 

[Peter Olcott]

The difference being that the former is an exact value with certainty, and
the latter is a guess or approximation?


Does it work with the fonts "WingDing", "Symbols", or my specially crafted
"all characters are invisible" font?

It works with any machine generated character glyphs that have visible pixels on
the screen. I don't know what your {"all characters are invisible" font} is, but
if the foreground and background colors are identical, then it will not work. No
character recognition technology can possibly work if there is nothing to
recognize.

 

[Stefan Ram]

I have technology that is similar to OCR technology except that
it determines rather than estimates its results. The main
advantage over conventional OCR technology is that it is much
more accurate with very small fonts.

See

http://www.structurise.com/kleptomania/

 

[Peter Olcott]

See

http://www.structurise.com/kleptomania/

I have extensively reviewed this technology, and it is the next best technology
in the world to www.SeeScreen.com technology. It achieved 44% accuracy when
other technologies such as
http://www.nuance.com/omnipage/ had essentially zero accuracy. Preliminary
tests tentatively indicate that SeeScreen will be able to achieve 100% accuracy
on this same sample. The following is a link that contains the actual sample
http://www.seescreen.com/Unique.html

 

[Peter Olcott]

See

http://www.structurise.com/kleptomania/

Basically to boil it down in a nutshell, SeeScreen is very similar to
Kleptomania for text recognition, except that SeeScreen is much more accurate.
Additionally SeeScreen is much more versatile, able to recognize thousands of
arbitrary graphical objects (not just text) from millions of alternatives in a
fraction of a second. This latter capability is very useful for GUI scripting.

 

[Oliver Wong]

It works with any machine generated character glyphs that have visible
pixels on the screen. I don't know what your {"all characters are
invisible" font} is, but if the foreground and background colors are
identical, then it will not work. No character recognition technology can
possibly work if there is nothing to recognize.

I thought I recall you exploring peering directly into RAM, so perhaps
under some conditions you could have determine what text had been entered
into a GUI control, even if such text were invisible. I'll now assume that
you don't do that, and base your recognition entirely on the set of pixels
captured.

In my WingDing font, the character for capital A looks like a right hand
forming the V sign (http://en.wikipedia.org/wiki/V-sign). In some other
font, that exact same glyph might be used for the character lowercase zeta.
Without peering into RAM, how would you know which of the two fonts are
being used to tell wether to recognize such as glyph as A or zeta?

- Oliver

 

[Peter Olcott]

I thought I recall you exploring peering directly into RAM, so perhaps
under some conditions you could have determine what text had been entered into
a GUI control, even if such text were invisible. I'll now assume that you
don't do that, and base your recognition entirely on the set of pixels
captured.

Yes there are methods where this works some of the time. The biggest case where
this never works is when the text is first written to a memory bitmap, and the
bitmap is then copied to the screen.

In my WingDing font, the character for capital A looks like a right hand
forming the V sign (http://en.wikipedia.org/wiki/V-sign). In some other font,
that exact same glyph might be used for the character lowercase zeta. Without
peering into RAM, how would you know which of the two fonts are being used to
tell wether to recognize such as glyph as A or zeta?

- Oliver

If there is even a single pixel that is much as a single shade of difference, my
technology can know with certainty which is which. There are various heuristics
that work with great reliability if two glyphs are identical. The heuristics
generally maintain the reliability above 99%. The only case where accuracy may
possibly drop below 100% is where there are two different glyphs in the same
font instance that are identical. If two glyphs are identical, yet, the between
glyph pixel spacing is different, the SeeScreen can still determine which is
which with 100% certainty.