New CAPTCHA

[Cheater512]

That logic puzzle would take me mere minutes to crack.
Ditto for the curved text.
The 3d one is the only real challenge but it too has its flaws.

The logic puzzle's flaw is how it tells the user which ones are the code.
Green circle with a white number in it? Thats a pushover. There is nothing like that green in the image.
Even with random colours it would take 30 seconds longer to code a workaround.

The curved text is way too easy. Find two straight lines close to each other and assume that they are parallel.
Use some maths to unwarp based on that assumption.

The 3d one's flaw is that its a nice smooth gradient but the letters disrupt it.
Its easy to find the edges of the 3d bit and unrotate it.
Then all you have to do is compare a proper gradient with the unrotated result and 'plot' the disruptions.
It stands out like a sore thumb and very easy for standard OCR to read after the disruptions are highlighted on a new image.

[NeoThermic]

The logic puzzle's flaw is how it tells the user which ones are the code.
Green circle with a white number in it? Thats a pushover. There is nothing like that green in the image.
Even with random colours it would take 30 seconds longer to code a workaround.

If it didn't tell the user which ones are in the code, it would be impossible to solve, no?

Can you back this up with any kind of concept code?

The 3d one's flaw is that its a nice smooth gradient but the letters disrupt it.
Its easy to find the edges of the 3d bit and unrotate it.
Then all you have to do is compare a proper gradient with the unrotated result and 'plot' the disruptions.
It stands out like a sore thumb and very easy for standard OCR to read after the disruptions are highlighted on a new image.

Again, any kind of concept? I've tried exactly what you detail here, but the biggest problem has been that you can't edge find on the 3D. You end up with either not enough edges, or too many.

NeoThermic

[Balint]

I had a nearly same idea some minutes ago, _before_ reading your post... here are a few of my suggestions. Feel free to add noise, shadows and all the other nasty stuff 8)

Another one is coming in a minute...


Bálint

[Balint]

Here's the last one...

Flame or OCR please *g*


Bálint

[Lieutenant Clone]

See this is the thing with the 3D ones. There isnt accually any letters in the images. It is a 2D images with some disruptions in the pattern, but through the magic of the human mind, we can decipher it as a 3D object, even though its not. Computers however have quite a bit more trouble doing what takes our brains a mere second.

Dont think so? Please, you and the other 0 people that have cracked a decent 3D CAPTCHA prove me wrong.

[Yawnster]

There is one called tEABAG_3D that looks promising (its a whole series accually)
http://www.ocr-research.org.ua/_teabag101.php (autogenerated, refresh page)
- very easily readable for being 3D (I can read them all easily and fast)

Had a little go at cracking this baby.. This is the product of around 20 lines of code..

solve4.png (4.71 KiB) Viewed 8599 times

teabag4.png (22.71 KiB) Viewed 8495 times

My other attempts are not as successful and require slight adjustments, but im sure a bit of sampling of the image will be able to make those changes.. I dont know if this would successfully go through an OCR program, I am guessing someone else can hazard a guess on this one, I think that i am probably experiencing the same problem as Neo that I am getting far too much..

Yawnster

[Lieutenant Clone]

and how well does the computer pick up that as a 'w'?

hmmm interesting at any rate....
Im sticking with the third one I posted then (care to try and crack it?)

[Yawnster]

and how well does the computer pick up that as a 'w'?

hmmm interesting at any rate....
Im sticking with the third one I posted then (care to try and crack it?)

Not now, I find it rather pointless to break a single image, a bit of rooting around the site reveals that the image you have produced is the only one of this kind, no biggie, but I will give it a go once a live one is done, thus it will actually be broken and not just image matched...

The one above is by no means comprehensive, far from it, im just showing that the edges can be found, but their clarity is questionable..

Yawnster

[Cheater512]

If it didn't tell the user which ones are in the code, it would be impossible to solve, no?

lol. Thats the problem with logic puzzles. Trying to tell the human what to do without letting the computer find out how to do it too.

Again, any kind of concept? I've tried exactly what you detail here, but the biggest problem has been that you can't edge find on the 3D. You end up with either not enough edges, or too many.

You dont edge find at all. You basically do a diff between a proper gradient and that and then do a edge find on the result.
It should highlight the letters rather nicely.

See this is the thing with the 3D ones. There isnt accually any letters in the images. It is a 2D images with some disruptions in the pattern, but through the magic of the human mind, we can decipher it as a 3D object, even though its not. Computers however have quite a bit more trouble doing what takes our brains a mere second.

A bit more trouble yes but if you derive a image with just those disruptions in it and not the rest and unrotate it it wont be too difficult for a computer to crack.

Can you back this up with any kind of concept code?

If I'm bored I'll whip some up today.

[DavidMJ]

See this is the thing with the 3D ones. There isnt accually any letters in the images. It is a 2D images with some disruptions in the pattern, but through the magic of the human mind, we can decipher it as a 3D object, even though its not. Computers however have quite a bit more trouble doing what takes our brains a mere second.

I am pretty sure Lieutenant Clone said that, not NeoThermic...