Post by Peter OlcottPost by Peter OlcottPost by Claudio GrondiPost by Peter OlcottPost by Capt Ravikumar R ManianPost by Peter OlcottIs there any OCR software that can approximate 100% accuracy at 96 DPI? What is
the most accurate OCR available that can process text from the display screen?
I am not sure OCR will work at all on images of 96 DPI. 100% accuracy
will probably be obtained if you work on a scanned image at 300 dpi of
an original which is a first generation laser print out.
Ravi
That was my best estimate too. I had Nuance techs try OmniPage 15 on the
sample from the link below, and it completely failed to match any
characters. ABBYY FineReader 8.0 Professional Edition provided the same
results when their techs tested their software on this same sample.
http://www.seescreen.com/Unique.html
My best current estimate is the SeeScreen is unique in the world in its
ability to recognize
characters as small as 6 points at 96 DPI screen resolutions with 100% accuracy.
This appears trivial to me as the shape of screen fonts is 100%
predictable, so no OCR in the usual sense is required covering the huge
variety of possible patterns - for me the reason for the 100% accuracy.
Simultaneously processing this huge set of patterns was the tricky part.
Post by Claudio GrondiMay someone explain here what the patent behind SeeScreen is actually about?
I have trouble to get the core of the message from the published patent
This technology provides the means to simultaneously process millions of
(1) The accuracy rate for text recognition is much higher than any
alternative method.
(2) GUI scripting of every application is now possible.
This latter benefit is derived from the fact that the mouse can now always
"see" where it needs to click.
This claim (2) is in my eyes nothing more than a huge balloon of hot air as
there is much more to 'GUI scripting of every application' than recognition
of graphical objects and texts. If it were actually true we would have any AI
problems already solved by it ...
This is true, another review by another expert pointed out some of these
difficulties. The key problem that this technology solves is providing the
means for the GUI scripting language to always know where it needs to click
the mouse. This now permits applications requiring mouse input to be
automated. Another key problem that this technology solves better than ever
before, is providing a means of data interchange between applications where
none existed previously.
SeeScreen appears to me more as a test if the features described will find a
market as a matter of fact of existence of a technology capable of doing what
it pretends to be capable of.
Perfect reading of texts from screen is already a long time successfully on
the market with the right mouse click using Babylon translator software
displaying the text of the word under the mouse pointer along with its
translation.
Claudio Grondi
If a method using all of the elements of the claim below already exists, I
need to know this. SeeScreen technology does not translate from one language
to another. Also there are methods for obtaining text from the computer
display screen that do not involve processing pixels. These methods (when
they work at all) always produce 100% accuracy. If there are any other
methods that involve processing display screen pixels that can achieve 100%
accuracy with font size as small as 6 points at 96 DPI display screen
resolutions, I need to know this.
1. A method for constructing a deterministic finite automaton (DFA) for
recognizing machine generated character glyphs in a graphic image,
comprising: collecting character glyphs of a font instance, including all
combinations of overlapping glyphs; merging the identical prefixes of these
character glyphs together; generating the DFA from the merged prefixes; and
storing the DFA in a sparse matrix representation.
The term "Deterministic Finite Automaton (DFA)" is taken to mean any
automated system whereby a next state is entirely determined by a current
input and a current state.
The term "Sparse Matrix" is taken to have the common meaning of the term
"Sparse" combined with the common computer science meaning of the term
"Matrix", a two dimensional array of elements. Thus a "Sparse Matrix" is to
include every means of storing any two dimensional array of elements such
that the number of elements stored is significantly less than the number of
columns times the number of rows in the matrix. This includes, but, is not
limited to storing a two dimensional array as a single dimensional array.
Now I am exactly at the point at which I have tried to get some hints about
how and what the method described in the patent does - up to now in vain.
Sorry, I don't understand the claims - maybe because the language used sounds
alien to me being not able to grasp what DFA or a 'Sparse Matrix' is for in
we take _one_ CPU and process the entire graphical information of a screenshot
in parallel (on multiple CPUs? or what does it mean in parallel) what does
speed up the process of recognition of screen elements letting all other
already known methods miles behind.
I would be glad to hear here what exactly is, what has not been done before
and is covered by the patent - best from experts in the area of OCR which seem
to stay away from this thread yet.
From reading the comp.compression I know how many US patents exist on methods
without any sense and practical use granted often because of lack of expertize
to reject them or knowledge about prior art.
So the only way to see oneself if a patented method makes sense is to
understand it - because it is not enough to know that a patent was granted to
assume it is real - as it can happen that the only purpose of a patent is to
get fundings for further research.
An online demonstration of the technology would sure help here as anyone could
see with own eyes if it works and how long does it take to get a response
after a screenshot was submitted.
Claudio Grondi
The working prototype is not yet ready for formal review. Two things that are
screen resolution.