** Next:** Noise as strong as
** Up:** Solution by weighting functions
** Previous:** Estimating the noise variance

I made the noise in Figure 2
and 3
from random numbers that I filtered spatially
to give a lateral coherence on a scale
something like the size of a letter--which is somewhat larger
than a line (which makes up the letter) width.
The noise looks like paper mottling.
The spectral **color** (spatial coherence) of the noise
does not affect the results much, if at all.
In other words, independent random numbers of the same amplitude
yield results that are about the same.
I chose this particular noise color
to maximize the chance that noise can be recognized on a poor reproduction.
We can see on Figure 2 that the noise amplitude
is roughly one-third of the signal amplitude.
This data thus has a significant amount of noise,
but since the signal is bigger than the noise,
we should really call this ``good'' data.
Next we will make the noise bigger than the signal
and see that we can still solve the problem.
We will need more powerful techniques, however.

** Next:** Noise as strong as
** Up:** Solution by weighting functions
** Previous:** Estimating the noise variance
Stanford Exploration Project

10/21/1998