Fwd: Convolution

[Boudewijn Rempt]

On Thursday 26 May 2005 10:44, Bart Coppens wrote:
> On Thursday 26 May 2005 07:12, Casper Boemann wrote:
> > I wuld like to warn against per channel kernels.
> >
> > I actualy changed it from 4 to one a while back.
> >
> > The reason is colorspace independence.
>
> You have a point. However, I would still like to see at least a seperate
> kernel for the alpha channel. This is because normal filters should not
> affect the alpha channel, but some filters do require that (for example,
> feathering the selection applies a gaussian blur to the selection paint
> device, which has only an alpha channel)

The curious thing is, I've tried to use

depth = src -> colorSpace() -> nColorChannels()
instead of
depth = src -> colorSpace() -> nChannels()

and the emboss/edge detection filters stopped working correctly, although the 
sharpen and blur filters worked fine. I wonder what I am doing wrong -- my 
experiments with convolution have been marred by integer overflow, 
off-by-ones and a complete disregard for correct coordinate calculation so it 
could be anything :-).

As for the selection: I guess nothing much changes there. You need just the 
one kernel, and have it operate on all the channels in the paint device, i.e, 
just the alpha channel.

I'm just wondering what's going to happen currently with paint devices with 
some funny things going on in the alpha channel...
-- 
Boudewijn Rempt 
http://www.valdyas.org/fading/index.cgi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://mail.kde.org/pipermail/kimageshop/attachments/20050526/58aed974/attachment.pgp