[daip] CONVL question

[Eric Greisen]

You have tested this pretty thoroughly and it certainly seems that
with small enough Gaussians that there is no need to double the input
image size.  CONVL actually does some doubling anyway internally.

When one convolves an image I with a second image I2, you can
understand that the second image would have to be twice as big as the
first in order to do all pixels correctly - thinking of the problem in
the image domain (forgetting about the FFT actually used).  That is
the origin of the statement in the help file.  Perhaps that statement
should be modified.

Eric Greisen