On 8/22/07, Thomas Lytje <[EMAIL PROTECTED]> wrote:
> I am not sure you take feature requests like this, - but try to take a look.
> It seems quite cool.
> I don't know enough about image processing (but I am a software engineer)
> but to me it looks like it wouldn't be to hard to implement. Hopefully there
> isn't a lot of patens making it impossible

Looks like a resizer in which the amount of source pixels per output
pixel varies spatially, rather than being roughly constant. It seems
to have a few requirements:
a) Resizing can only be done on one axis at once
b) two scalers would be needed, one iterating over X axis, one over Y.

Basically the only change relative to normal accumulators is that the
number of pixels resulting in a final pixel would need to be inversely
weighted by the significance mask (that is, read more input pixels to
produce an output pixel in insignificant areas.) There is also one
coefficient involved that would need some experimentation with to get
right: the exact ratio of scaling between completely significant
pixels and completely insignificant pixels. (I mean, when you shrink
that image, the significant features must also shrink somewhat -- at
least once they begin to push up against each other.)

Anyway, if you give a link to a paper describing the exact workings of
the algorithym, then it's much more likely that something will get
done in relation to it.
Gimp-developer mailing list

Reply via email to