Comment 1 for bug 685919

acmihal (acmihal) wrote :

Logged In: YES
user_id=1152911

The general idea is that the width of the blending mask
grows proportionally to the spatial frequency of the image
features. The standard Laplacian pyramids the spatial
frequencies are divided into an exponential series. So there
is plenty of room to change the constant of proportionality,
or come up with a more sophisticated relationship.

The only caveat I can think of off-hand is that the width of
the blending mask can exceed the size of the image overlap
region. This happens even with blending mask generation in
the current version of Enblend. The result is that Enblend
identifies that certain pixels are close to the seam, and it
knows that it has to blend them against something in order
to prevent a visible seam, but there is nothing to blend
them against. This all stems from the fact that we have
images that overlap in irregular ways. It is an open
research question.

If the mask and the pyramids both use the same filter size,
and the seam line never leaves the overlap region, then this
problem does not occur. The image filter can automatically
perform a multiresolution extrapolation which fixes the
problem. Enblend does this now.

If you want the blend mask to grow a little faster, then we
first need to solve the problem of what to do when the mask
extends beyond the image data.