5 replies to this topic

[Juraj Rolko]

hello,

i have created a new captcha technology concept. this captcha is based on 3D model rotation and human imagination. i am trying to present this idea to public worldwide. you can see the project description at 3D captcha. i will be grateful for any feedback.

thank you.

[WingedPanther]

I hate to say it, but I can see two attack vectors that could work well.

1) Using auto-it to capture all possible positions. Pass the images off to an image processing algorithm that attempts to match the center image to the surrounding images. (or process smoothness of boundaries if there are normally no surrounding images).

2) Detect the position that has the minimal amount of white-space. Since each triangle blocks less of the background white as it is rotated away from the correct position, the angle with the least white is probably correct. This could be mitigated by converting each triangle into a tetrahedron and/or adding additional tetrahedra/triangles at different levels that will be blocked by the correct positioning.

[Juraj Rolko]

thank you for your reply.
- you are right that the smoothness of boundaries detection technique is quite a big problem
- adding some "noise" triangles to the model is one of the additional techniques which was not published yet. so you are right again
- converting triangle to tetrahedron is probably not the way. for the demonstration purpose, 2D image was decomposed to the equilateral triangles, but in real use, it has to be decomposed to non-equilateral triangles. i think that using irregular tetrahedrons will cause that the original shape of tirangle can be calculated with some mathematical method. but i am not sure. i am not good mathematical theorist

[WingedPanther]

I'm a mathematician as well as a programmer. Perhaps noise triangles plus irregular orientations of ALL triangles would be helpful.

[Sinipull]

Interesting idea, but seems a bit easy to break. Even machines could tell when the the dots make a complete picture. They would just use density.

You know, machines can't say very well which side of the image is "up", if there's no horizon. Perhaps you should look into that.

[Juraj Rolko]

thanks sinipull for the reply.

- you are right that the example is pretty easy to break. but I wanted to make it as simple as possible. i am working on more complex examples now.
- captcha based on image's upright orientation already exists. it is called what's up captcha and it was created by google. I wanted to create something different.
- in this case, the basic idea is to solve a 3D puzzle. I still think that this idea is good and there is some potential for future work