From: cpd@cat.pan.net
"Tim McGaughy" wrote in message
news:AeednVd8WMTvSjnQnZ2dnUVZ_sGdnZ2d@posted.toastnet...
> catpandaddy wrote:
>>
>> "Tim McGaughy" wrote in message
>> news:L8ydnT7-MfuilALQnZ2dnUVZ_ridnZ2d@posted.toastnet...
>>> catpandaddy wrote:
>>>
>>>>>> Look, even simple optical-character-recognition scanners don't
>>>>>> perform as well as humans.
>>>>>
>>>>> Maybe you should actually try it sometime. Not counting the time it
>>>>> took to do the actual scan, it just took my Mac about 10 seconds to
>>>>> visually recognize characters and write the results to a window in
>>>>> TextEdit.
>>>>>
>>>>> Even counting the scan, I can't copy a page of text nearly as fast as
>>>>> my computer can.
>>>>>
>>>>> And that's just basic commercial software with an off-the-shelf
>>>>> scanner. Not some robot system somebody just spent hundreds of
>>>>> thousands of dollars developing.
>>>>>
>>>>>> It took decades for OCR to get to where it is today, and even now it
>>>>>> ranges from no higher than 98% accuracy to below as 75% accuracy, and
>>>>>> that's with clear imaging of typewritten text as the source.
>>>>>
>>>>> You really should try it for yourself before you continue. I saw no
>>>>> mistakes in what the computer cranked out, and I frankly think you
>>>>> just pulled those numbers from your own ass
>>>>
>>>> Not necessary.
>>>
>>> You mean you have nothing to say. I gave you a real-life example of an
>>> experiment I conducted myself, and that you can also conduct yourself to
>>> make sure I'm not fibbing.
>>>
>>> You said optical character recognition doesn't perform as well as
>>> humans. I proved you wrong, as far as typewritten text, and all you have
>>> to offer up is, "Not necessary".
>>
>> I was referring to the cussing.
>
> Sure, you were.
"Not necessary" did refer to the cussing. There's nothing else it /could/
have referred to.
>
>>> I'll ask you straight out... Where did you get the 75% statistic from?
>>
>> We crossed each other in traffic. I was putting together my response and
>> posted it before seeing this one, so it will show up in the thread as
>> another reply to your previous message. It will meet the burden of
>> proof.
>
====
> Bullshit.
====
^^^^^^^^
Again, *not necessary*, and if you still need to cuss me even though this
is not some kind of personal contest, that's not good on your part.
> Your cite talks about the error rates for ARCHIVED NEWSPAPERS, not clear
> images of typewritten text. In fact, it disagrees with you about 'clear
> imaging of typewritten text'. Its conclusion is that the older and more
> damaged newsprint got more errors. Well, duh. Imagine that... If it's hard
> to read, the computer has trouble reading it.
In other words, it can't read those in situations where a human /can/ read
them. Again -- where a *human* CAN read it just fine, the OCR fails by a
huge margin. Even when the text itself is undamaged and only the background
paper isn't the perfect pristine shade of contrasting pale grey or white.
Your very own reasoning would suggest, that it must be time for OCR to pull
the plug on any further R&D and forfeit all funding and/or grants.
And it bears repeating: we have both stated our claims and given our
rationales, and it all rests on a difference of opinion on what a
computerized recognition system ought to be able to do, by what metrics, and
by what point in time. I have no problem with us being in disagreement
there. You can believe such cognitive research projects should be scrapped,
that they shouldn't be worth anyone's interest, that nothing of utility will
ever develop from their trials on the timetable you demand of them, whatever
you want to believe. The world will go on just fine regardless of what
either of us think.
You really needn't get this worked up about things. We disagree on how
cut-and-dried this fancy research stuff is, and that's that. Chill out.
--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)
|
