Smoothing Algorithm Question

  1. Nope, I just was just using the word to help express the fact that while values are related, and you could probably fit a function to them, that they’re to be taken as individual discrete points in the grid. The values are positive, non-zero floating point values (Unsure of the precision).
  2. This is somewhat true, at least for our real world test set it. We typically run it at about 96x26x66 or something like that.

My only knowledge about signal processing is having written a parallel FFT. Which is to say I honestly don’t know much about anything that you’re currently talking about in that area.

One last shot on the basic sliding rectangle. Instead of smoothing the whole volume just apply the function when a cell’s value is requested. Because the rays will be continuous you can still use some of the simple optimisations and slide the rectangle down the ray.

This would require three copies of the data though: the original, the sparse smoothed version, and the result.

The efficiency would then depend on how many cells are typically hit by rays between smoothing operations.

Okay: not being an integer set mean the s-transform and related are not interesting. My thought on “close” is that for early iterations a classic low-pass is likely to be helpful in convergence as well as preventing too great of impedance changes. However when parts of the data set are close it will end up being counterproductive. A 1D example of a close is to visualize the signal has a height field, then you roll a circle of some given radius over the surface of the field. If the there are any sample point below the radius the samples value is increased to where it would touch the surface. A circle is just an example structure element…so it removes holes that are smaller than the structuring elements. (Don’t bother looking at the wikipedia page).