Simplifying voxel graphics

Hey /prog/

Decided to start working on a new project and I'm kind of stumped. Hoped someone might have some ideas or links to papers or something dealing with this sort of thing.

Writing a program for creating voxel graphics and I want to be able to export optimized meshes from this program. I'm having a hard time wrapping my head around how to group up voxels to eliminate redundant faces, however. Baking the texture and triangulating the results will be simple enough, but that first step of optimizing out all the redundant faces just has me stumped.

I've got an example of what I mean here:
http://i.imgur.com/O3bq4.png

Anyone have any ideas? :)

>>1
Sounds like you're trying to do voxel rendering by pushing them through your typical polygonal rasterization pipeline, which is what Minecraft and various clones do. Marching cubes is the algorithm Minecraft uses to convert voxel data to a triangle mesh.

It should be noted that this isn't the only way to render voxel data. Looking forward, the future of voxel rendering on GPUs is directly ray-casting sparse voxel octrees (SVOs) into a framebuffer or geometry-buffer (for deferred shading renderers) using a compute shader/kernel, such as with OpenCL or DirectCompute.

The cool thing with deferred shading approach is that you can seamlessly support both voxels and conventional triangle mesh geometry in the same pipeline, just have a geometry-buffer pass for voxels, and a second geometry-buffer pass for triangle meshes. After that, the rest of your pipeline does deferred shading and lighting composition as per usual, without having to worry about the topology of the original data.

nVidia has put together a demo with source code that does this, but it uses Cuda + Direct3D.

http://code.google.com/p/efficient-sparse-voxel-octrees/