3D engine to create simple 3D objects

[stampede]

Is it simple?

Automatic generation of 3d models using images ? I think its definitely not.

3D modelers use a technique to create objects using blueprints placed on the "planes of reference", you can read about it here (just to see that its rather complex task):
modeling from guide images
Never heard about a tool that does this job automatically.

Please share if you find something.

[SixDegrees]

MRIs, CAT scans and the like build up 3D images from a series of 2D slices. But the key here is "series" - they use a LOT of slices, with very small differences between them, to build up their 3D model. If you only have a few images (front/left/top, for example) you are limited in what you'll be able to reconstruct. And some structures, like hollows that are partially enclosed, fall between very difficult to impossible to reconstruct this way; you simply never get a clean silhouette to work with.

In some cases - mapping surface images onto a known spherical surface, for example - this sort of thing becomes much simpler, but even here errors in sensor positioning can cause lots of problems in the reconstructed surface.

Short answer: unless you're dealing with very simple, predictable shapes, this is a problem that falls between very hard and impossible.

[been_1990]

Let me clarify, It should only be " dealing with very simple, predictable shapes". Example: Fonts, plain vector shapes,etc

[wysota]

How is that even related to programming? Moving to General...

[tbscope]

There is a technique where you apply some basic filters (don't know the names of these filters by head) and get an image with all the important contour lines. This way, from a basic 2D image, you get another 2D image with the contour lines.

Then, transform these contours into vectors. This should not be too difficult and there even exist libraries that do this. I know that Inkscape can do this.

Then, use these 2D vector images to compose a 3D image.

And as a last step, you can try and sample some colors or textures and apply them to the 3D image.

I think if you go look for "non photo realistic" rendering or "occlusion" rendering with OpenGl, you'll find some interesting filter techniques.

[stampede]

There is a technique where you apply some basic filters (don't know the names of these filters by head) and get an image with all the important contour lines. This way, from a basic 2D image, you get another 2D image with the contour lines.

Then, transform these contours into vectors. This should not be too difficult and there even exist libraries that do this. I know that Inkscape can do this.

You can do something like this with OpenCV ( feature detection ). Try to search for "edge detection" and "finding contours" with OpenCV.

[wysota]

You don't need OpenCV for edge detection. Such algorithms (like the Canny, Sobel and others) are easy to implement.

[stampede]

You don't need OpenCV for edge detection. Such algorithms (like the Canny, Sobel and others) are easy to implement.

You're right.
But what OpenCV can do for you is to represent detected edges in convenient data structures, restore full hierarchy of nested contours, use different approximations ect.
As long as it's about something more than displaying the image with detected edges, I would consider using image processing library.

Anyway, Been_1990 wanted a tool in a first place, so I doubt that our discussion here is really helpful for him.

[been_1990]

How is that even related to programming? Moving to General...

Sorry Wysota,I thought it was appropriate.

Anyway, Been_1990 wanted a tool in a first place, so I doubt that our discussion here is really helpful for him.

I don't need a "tool", a program, I need a library that simplifies 3D creation. Example:
Create 3d square using vector image X that has a square drawn in it.
Or if I give it a O shaped vector, it would calculate the 3D version of it.

Kind of what Qt does with 2D shape creation. Just that it will do the same with 3D.
I want a 10px 3D square, and then it's created.

[tbscope]

Just use OpenGL ?

[stampede]

Create 3d square using vector image X that has a square drawn in it.
Or if I give it a O shaped vector, it would calculate the 3D version of it

What do you mean by "3d version of a square" ? Or "3d version of O shaped vector" ?
You expect a cube in the first and sphere in the second example ? I'm just little confused.

But, if you just want the shapes to be represented as flat shapes in three dimensions, then its quite easy. First you just need to detect edges ( contours ) of such objects and represent them (edges) as a collection of points in 2D space, then foreach point p=(x,y) from this collection you can simply create a point in 3D:
p' = (x,y,0)
This way you'll create a flat representation of your vector shapes in 3d space.

As I've mentioned before, OpenCV can do the contour detection job for you.