compress a 3D volume into a 2D image

Feature extraction for classification purposes is more complex in 3D compared with 2D images. The corresponding functions work with 2D images mostly. That is why I was wondering if that would be possible to use a (non)linear tranformation of the 3 intensity values at each (x,y) coordinates of a 3D volume to find a single intensity value. Then, the question is can this sort of 2D projection be used by feature extraction functions to develop accurate classifiers.

[Image Analyst]

Perhaps, if your objects are cylinders (don't depend on one of the directions).

Answer 1

[Neuropragmatist]

Could you project the data along each axis (using mean(data,dim) for instance) to get 3 two dimensional images and then do the training on them?

M.