Re: Parsing images

At 1:01 PM -0400 5/12/06, Robert Cummings wrote:

The original poster asked the question -- can it be done. And of

course, the answer is Yes.

Yes "it can be done". No, it's not trivial in the general.

Well... generally it's of little value to me. :-)

Visial recognition is what we are talking about when you say it's trivial and can be done easily. Because visual recognition is the necessary branch of analysis required to be able to analyse any captcha and come up with the answer.

I say it's trivial because it's not important. I also say it's trivial because what I have seen and worked on makes it pale in comparison.

Granted, the more complex the image, the more difficult for a program to decipher it, but a CAPTCHA has to be, by definition, detectable by a visually unimpaired human.

And you're making what point here? That you can write code to match the capabilities of a visually unimpaired human's visual processing and recognition? Wow!!

No the point I am making here is that CAPTCHA is used weed out bots from persons who ARE NOT visually impaired -- and thus, the objection to it's use because CAPTCHA is a device that goes against the ADA and all those who care about access for persons with visual disabilities. As for visual processing, we do have screen readers -- they do work. You say otherwise? Could I write that software? Yes I could. I even wrote a Macintosh blind browser that would capture the content of a web page and read it aloud -- however -- that was not image recognition, but rather text to speech processing. Have I written tools to analyze signals and detect anomalies? Yes. Have I written AI programs? Yes. But, what's the point here? It isn't IF I could do it, but rather IF it can be done -- and it HAS been done. The only issue you are arguing here is IF it is trivial or not. I say it's trivial and you say otherwise. Okay, you're entitled to your opinion -- as am I. Everyone has the right to be wrong. :-)

Image analysis, enhancement, alteration, and such are better

performed by computers than by humans -- that's the reason we developed the software in the first place.

WRONG! A small subset of these are better performed by computers.

Bull--toot! You say that you're better than PhotoShop at doing frequency analysis, bright/contrast, color levels and such -- now that's funny. I know you didn't actually mean that.

And in almost every case, the computer's results are checked by humans afterwards. Most computers just flag things as interesting, then a human goes and makes the executive decision.

As it should be -- BUT -- the computer is still making a decision to show something, is it not? It's still a conditional outcome, right? It flags stuff for a reason.

The step between analysis and detection is simply meeting an

acceptable error threshold.

Maybe so, and I guess if you considering .00000001 success rate ok, then writing a captcha breaker might be trivial since you can probably just generate a random string.

It's all relative, isn't it? Like my success rate at winning an argument with my wife, that figure you gave above would be pretty high for that. :-)

As for me being naive -- well... either one of us could be -- but

that's probably what I get for working in signal analysis (seismic data) since 1975.

Someone once posted the following to the list, I feel it appropriate to quote it at this time:

    "Locus ab auctoritate est infirmissimus"

LOL -- You got me on that one! I almost posted that quote as a framework for me using my "authoritative argument". I hate it when my words come back to bite me in the butt. But, in my defense, 31 years of working with signals (granted, it long enough ago that the first ones were smoke), has given me a perspective that everything can be distilled down to components and those components can be analyzed. And from that analysis, one can make a more informed "final" decision. However, you must also realize that the computer (via the programmer) has helped you arrive at the final decision -- and that took countless smaller decisions. It's not that much different than programming. I gave a talk one time for the AAPG regarding AI in the Geosciences and someone asked me to provide an example of a computer being smarter than me. I answered that many times I have been able to research and write a routine that would record what I learned about a specific algorithm at that point in my life (my apex of knowledge for that subject). After writing the routine, I could then dummy-up and go on about my life researching other routines in similar fashion. However, my routines would retain the apex of my knowledge about a specific topic and if I placed two of these routines in concert to generate an outcome, then the outcome would surpass what I could have done. In that fashion, my computer generated intelligence surpassing my own. Granted, some say that a computer unplugged would still do that.

As for the bleeding edge, that's obvious, just look to medical

imaging. They wish that the detection of their problems were as

simple as CAPTCHA. Just because they may wish they were detecting something simpler such as captcha (maybe, since I could render my captcha to look like medical imagery and so it would be just as difficult to detect), doesn't mean captcha is trivial. I'm not currently aware of too much medical imaging processing that occurs without human intervention. Much of it requires that a human view the results and make an informed decision based on the computers analysis. I think part of your naivety is thinking that the captcha you see right now is as hard as it gets. You are very mistaken.

Mistaken? I've been there before -- and will be again. But, you still haven't persuaded me that deciphering a CAPTCHA ranks in any fashion whatsoever to the problems that computers are currently solving. It's still a trivial problem. Thanks for the exchange. tedd -- -------------------------------------------------------------------------------- http://sperling.com