In the context of this paper *image synthesis* is the process of
transforming an uncorrelated image to one with the same textural
qualities as a known ``traning image'' (**TI** for short).
The traditional applications of image synthesis have been in computer
graphics, but same ideas will prove useful for earth scientists, due to
the nature of their experimental measurements.

To create the synthesized image (**SI** for short),
we first compute some of the TI's key one-point and two-point statistics
and then impose these statistics on the SI.
To overcome the difficulties in estimation induced by scale variance, our
method utilizes the ``Laplacian Pyramid'' decomposition, a simple
and well known tool for multiscale image analysis.

1/25/2002