[ptx] Re: Enblend suggestion: focus score weighting

[Sebastian Nowozin]

Hi Andrew,


On Fri, Jun 18, 2004 at 12:15:03AM -0700, Andrew C Mihal wrote:

> > Maybe such intermediate tool could produce an additional alpha-mask for
> > each image and enblend would use that alpha mask as "focus weighting"
> > and adjust the blending to prefer pixels that are in focus, but falling
> > back to unfocused ones if no other is available.
 
> I worry that the focus weight and enblend's blend mask weight are mutually
> exclusive metrics. Imagine a region where the blend mask says image A 
> should be mostly faded out. What if image B is mostly out-of-focus there? 
> If you give A extra weight, there will be a splotch.
 
I do not understand clearly what you mean. See, for the two images case we have
to use both anyway, so we shall encourage a soft blending over any focus issue.
Two images will be no differently blended than the current enblend does, just
because we have no other choice (besides "shifting the blending into the
direction of the more focused image", which does not seem to be a good idea to
me).

The really interesting case is for areas where there are three or more images.
Then, instead of choosing the pixel to be blended with by random or by the
order of images given on the commandline, one could make this decision based on
the quality of its focus.

Do you think that makes sense?

(I have some free time after the next week and this problem surely sounds
interesting enough to hack something up. Maybe I can write some simple
prototype from one of the papers named on the list.)


On a more abstract level I think the steps of smooth blending (enblend) and
focus based image merging are not separable into two passes. Both steps produce
one image from many, based on some blending metric. You cannot prepend one step
in front of the other because the resulting image will have only one layer.
Therefore, I think its the most logical thing to integrate part of such focus
score method in enblend. However, the computation of the focus scores could be
done externally, to minimize modifications in enblend. Do you think that makes
sense?


> Andrew

bye,
Sebastian

-- 
nowozin at cs.tu-berlin.de --- http://user.cs.tu-berlin.de/~nowozin/