3d image compression

[m^2]

There's been a lot of research on 2D rasters, but I haven't seen anything about 3D. I am thinking about doing some research on it with the main area of interest being tomography images, but also dive into films which are basically 3D images too. Is anybody aware of any results in the field?

[willvarfar]

Both precomputed mipmaps and 3D images are common in games; decent compression (with decent decompression speed) could be increasingly important for games as assets get more extensive at the same time as moving away from physical media distribution

Also, spectual maps being compressed in terms of the corresponding colour maps and be considered 3D too

[thorfdbg]

There's been a lot of research on 2D rasters, but I haven't seen anything about 3D. I am thinking about doing some research on it with the main area of interest being tomography images, but also dive into films which are basically 3D images too. Is anybody aware of any results in the field?

Do you mean compression of 3D models/objects as used in computer graphics and 3D animation (i.e. meshes and textures), or do you mean volumetric images? For the latter case, such images appear in medical imaging, and there JPEG 2000 part 2 is applied, using a wavelet transformation in Z direction - actually a "disguised" color transformation. Additionally, JPEG 2000 had a part 10 that describes a dedicated 3D compression, but it never became very popular because it doesn't work much better than part 2. Improvements over slice-by-slice compression are typically not so large as most of the energy is already in the smallest LL band and decorrelation over the high-passes in Z direction does not deliver high returns.

[m^2]

Both precomputed mipmaps and 3D images are common in games; decent compression (with decent decompression speed) could be increasingly important for games as assets get more extensive at the same time as moving away from physical media distribution

I guess that specialised compression is not used for them ATM?

Also, spectual maps being compressed in terms of the corresponding colour maps and be considered 3D too

Spectual maps? Isn't it a typo? At least I can't find any information of them.

Do you mean compression of 3D models/objects as used in computer graphics and 3D animation (i.e. meshes and textures), or do you mean volumetric images? For the latter case, such images appear in medical imaging, and there JPEG 2000 part 2 is applied, using a wavelet transformation in Z direction - actually a "disguised" color transformation. Additionally, JPEG 2000 had a part 10 that describes a dedicated 3D compression, but it never became very popular because it doesn't work much better than part 2. Improvements over slice-by-slice compression are typically not so large as most of the energy is already in the smallest LL band and decorrelation over the high-passes in Z direction does not deliver high returns.

Volumetric images, that's the keyword, thanks a lot for it.
Now I can find quite a lot of research. I knew I was just searching wrong.
As to the standard extension...hmmm...I feel like 100 euro is a lot for a pdf. Anyway I found the reference implementation with a claim that the authors give it for free to researchers, so I'll look into it as well as to other things that the new keyword reveals.

[willvarfar]

yeah, typo: specular

Often there's a colour map, a specular map, a normal map and even an ambient map. These are all just three or four component images. And they all correspond to one another.

Look a collage of them overlayed here: http://blenderartists.org/forum/show...=1#post1161619

[thorfdbg]

As to the standard extension...hmmm...I feel like 100 euro is a lot for a pdf. Anyway I found the reference implementation with a claim that the authors give it for free to researchers, so I'll look into it as well as to other things that the new keyword reveals.

It's the usual ISO pricing, however, you might be able to find the FCD text which is the most advanced version we are able to release. Besides, there is not so much to say about volumetric compression in JPEG 2000. Within part-2, the RCT/ICT color transformation can be replaced by a multi-stage decomposition, consisting of a RGB to YUV transformation (for example), followed by a wavelet-decorrelation in component direction. If you represent each slice of the volume by one (or a set of) components, this will implement a wavelet transformation in Z direction. Part 10 works similar, except that it uses a more canonical approach to inidicate the Z direction transformation, i.e. slices are there not represented by components - the overall definition is cleaner. In essence, it just changes the order in which the transformations are applied. Codeblocks become code-volumes, but the encoding procedure remains two-dimensional because any extension of the context model into the third dimension did not help to improve the coding efficiency. The problem is here that you get too many contexts, and the context dilution stops the improvement.