3D images: A human way to create Captchas

The new 3D image-based design for Captchas.

(Credit: Taylor Hayward)

If you have registered a new e-mail account or tried to retrieve your, say, Gmail password, you'd be familiar with the use of Captchas, the challenge-response method to verify that the input is not generated by a computer.

The problem is that most existing Captchas, or Completely Automated Public Turing tests to tell Computers and Humans Apart, are actually quite inhuman. They are often made up of wavy lines, letters, and numbers shown in high contrast, different colors, or strike-throughs. Generally, they are distorted sometimes to the point that it's really hard for people to quickly make anything out of them. Personally, I've rarely been able to pass a Captcha the first time.

As the use of Captchas has been getting more and more popular, it's about time that somebody thought of a better test. And somebody sure has.

A blogger named Taylor Hayward has come up with a brilliant idea: using 3D images as Captchas.

The challenge is a 3D image of an animal, say of a rabbit's face. The list of answers would display different common animals from different angles, including a photo of the rabbit, this time of its side. Only a human brain would be able to quickly see that the challenge image and the second image on the answer list are of the same animal. Now you just need to click on the correct second image to pass the challenge.

This would make a great method to create Captchas that are fun, easy for humans to read, and, at least for now, close to impossible for computers to quickly pass. This is because humans' vision is so much more sophisticated than any existing computing algorithms. Something our eyes can see right away would take a lot of computing power to figure out.

For example, you can recognize a pair of shorts on top of a pile of clothing instantly. A computer wold take a long time and might even not be able to do that at all.

A computer might be able to circumvent such a Captcha method by clicking on all of the images, one after another. However, this outcome could be prevented by making the list longer and repeating the challenge from the beginning, if the first click was incorrect.

I really hope that Web sites would pick up this new design. In the end, I'd rather have to prove that I am human than prove that I am not a machine. And yes, they are two different things.

An example of Captchas currently used by Gmail

(Credit: Dong Ngo/CNET)

[michaelportent]

I think this is cute and creative. I utterly hate captchas and there needs to be a better, faster, more user-friendly method to judge humans against robots.

[mcrosby78]

Of course this would take a lot of processing power, especially for very large sites, as each image is rendered by the server. Computers are also very near to cracking this method.

[karpenterskids]

I can't believe I didn't invent this myself! :0

As a Myspace user who adds a lot of people (and writes a lot of messages) from my band's Myspace page, I'm constantly annoyed every three or four actions I do to prove to the system that I'm actually a human being.

This is so much easier!
Every website should adopt this system, or at least something similar.
If animals are too "unprofessional" for certain websites/organizations, it could be exchanged for shapes instead.

THIS is the future.
I hope it comes quickly.

[c|net Reader]

The human has to be able to recognize the projection of the original, 3D object. Many humans have a difficult time doing that with shapes, whereas they recognize animals easily from having seen so many images from so many angles. In other words, 3D images of real world objects are relatively easy for humans to project, but geometric shapes are not. The images also have to be relatively complex to prevent rapid image scanning and projection.

[JRKhoury]

On ubuntuforums.org, word math problems are used and i find that it is a FAR simpler than this approach though this is possibly more failsafe.

[c|net Reader]

OCR and math are well within the capabilities of a computer.

[rapier1]

The problem with this method is that you still need a fallback to handle visually impaired users. If auditory cues prove to be more hackable then you really haven't solved the problem. You really need to address the entire verification chain at once.

[jc364]

A computer would have a hard time getting it correct the first time. But with a 1 in 9 chance, it wouldn't take long before the catcha is defeated. I would suggest picking 3 related images from a group of 16 or so. The chances of a computer guessing that would be much greater than the 1 in 9 suggested here.

[c|net Reader]

Note that there is a 7s penalty for guessing wrong and then a new set of images is presented. Thus, the chances of guessing are much lower than 1 in 9 and will take longer than you assumed.

[virgilp]

yep, that was my first reaction too: a bot trying to spam a forum (for instance) doesn't need a 100% success ration at cracking captchas. 1% is quite ok. It's not feasible to pick from prefabricated answers.
However, the idea is not entirely bad: what if the captcha was an embedded (interactive) flash, with 3D text that you could rotate? assuming that it is initially in an "unreadable position", now the captcha-breaking programs would have to guess the required rotation, too (or would have to try reading at many different rotations; also, interacting automatically with the flash movie may prove to be quite difficult).

[StefRush]

I guess you are not aware of Numenta 3D images won't work for long or at all http://www.numenta.com/ ;-)

[usbethjim]

Forget the rest, use - ReCaptchas. Keystrokes with a purpose...

http://recaptcha.net/

[c|net Reader]

I would think that the projections of the original image should be required to include information not in the projection shown in the top image. The example does this, but I didn't see any mention of this requirement. The idea is to prevent the software from being able to identify the 3D information from the top image and then try various projections from it.

[hippypink]

I have already hacked this using simply geometry. Doesn't anyone here remember symmetry? Just make an outline of these and its easy to plot which two images are most alike.

Heck, I will even use it in its 3d form and graph the line points, and compare it with my 3d animal database. Sooo lame.

[marquinhocb]

hippypink: It's not as easy as you'd think. There's more to it than just simple geometry. You can see a captcha like this in action here:
http://www.yuniti.com/register.php
I challenge you to come up with this supposed hack that would so easily break this captcha.

[marquinhocb]

My brother and I discussed a similar idea on Skype on the 12th of March - and implemented it. You can see it in action here:
http://www.yuniti.com/register.php

[nu7i]

Don't forget about the needs of the blind and visually impaired for equal accessibility. As long as we're not talking about an accessible alternative, this 3D CAPTCHA scheme is discrimination and should not be used anywhere. We are holding web site operators accountable for accessibility, sometimes with civil lawsuits, and CAPTCHA is no exception.