Enblend layers (was Re: [ptx] PTmender Version 0.2 "The enlightment")

[Daniel M. German]

 JD> Andrew basically said it is not.  There exists no mask which can be
 JD> applied to two unaltered layers to produce the Enblend output.  Unlike
 JD> the old PTStitcher, enblend does not simply "feather" a mask to
 JD> transition from one image to another: it actually intelligently selects
 JD> regions within all overlapping images to blend.  In order for this to
 JD> work, Photoshop would need some type of curve or mask layer that allows
 JD> you to specify, for each pixel, a surrounding neighborhood of pixels to
 JD> operate on, and the specific way in which to operate on them.  If this
 JD> were true, Enblend could probably be re-written as a Photoshop plugin.

Maybe I am wrongly expressing my ideas.

enblend takes part of image A and part of image B and creates the
output.

But how about, at the same time, to apply a similar Colour and Brigh.
transformation to each of the entire original images? So we have 1
transformed layer per photo, plus one transformed layer for the
original enblend result (even with no masks).


I have the following idea, which might sound very crazy, not quite
optimal, but which will require no modification to enblend.

1 layer is the current enblend result (no mask)
2 layers that are the original photos, also corrected. 


I haven't looked into enblend algorithms yet. But how about, say,
photo A (200x200) is blended with photo B (200 x200), and the photos
overlap at 50% of each (100 pixels overlap). Assume for simplicity
that the "seam" created by enblend is exactly in the middle.

So we have 100 pixels of A, 100 pixels overlap, and 100 pixels of
B. the 100 pixels of the overlap are divided 50 and 50 each.

We now have a resulting blended image.

Now, change the mask of B to make it effectively 150x200 pixels, so
the overlap is now only 50 pixels. Run enblend. Create a new layer,
and put in under the previous one.

Repeat the previous step, with original B and A with only 150x200
pixels

So now we have 3 layers, each with seams at different places. This
might allow some flexibility to select data from one layer or the
other, while still preserving the optimal seam. 

Once I get back home I'll try this idea in a hand held panorama that
gave me a lot of trouble.

 JD> Enblend can be told to ignore certain regions on input by masking them
 JD> out in the input image(s) alpha channel(s).

Yes, but if the masks are complex and the matching is not perfect it
has problems trying to put the seams in the perfect place.


--
Daniel M. German                  "That the only purpose for
                                   which power can be rightfully exercised
                                   over any member of a civilized
                                   community, against his will,
                                   is to prevent harm to others.
                                   His own good, either physical
   John Stuart Mill ->             or moral, is not a sufficient warrant."
http://turingmachine.org/
http://silvernegative.com/
dmg (at) uvic (dot) ca
replace (at) with @ and (dot) with .