[Talking about "16 bit image map pattern not working with isosurface"]

Christoph Hormann wrote:
>Note this is probably an extremely bad idea since it does not work with
>interpolation 
>
I dare to object...

I think I have the mathematical proof that there is no problem 
with interpolation even when using two separate 8 bit images instead 
of one 16 bit set. You just have to use bilinear interpolation on 
both of them: 

Consider we have 16 bit topography data split into two 8 bit sets 
with low and high byte of the values. 
Look at two pixels a,b next to each other: 
  a = a_lo/256 + a_hi
  b = b_lo/256 + b_hi

Normal linear interpolation would do just that: 
  val = p * a + (1-p) * b

But using two separate textures, they get interpolated separately: 
  val_lo = p * a_lo + (1-p) * b_lo
  val_hi = p * a_hi + (1-p) * b_hi
Then, these get composed to the real value as used in the isosurface: 
  val = val_lo/256 + val_hi 
      = [ p * a_lo + (1-p) * b_lo ]/256 + p * a_hi + (1-p) * b_hi
      = p * (a_lo/256 + a_hi) + (1-p) * (b_lo/256 + b_hi)
      = p * a + (1-p) * b      
...and that's just what we want.   -- q.e.d.

Bilinear interpolation is just the same in two directions. 

Incorrect? Any comments?

Cheers,
Wolfgang

Post a reply to this message