[sane-devel] ANNOUNCEMENT: dust removal: please test

Peter Seiderer Peter.Seiderer at ciselant.de
Tue Sep 10 10:56:23 BST 2002


Hello Andras,
sorry for the delayed answer....

No better idea as working with a fixed ir threshold....but I do not know if
different films need different thresholds, did not much tests yet with
different material......

Fixed a minor bug in my code (missed some pixels of the border around the
ir threshold area). Now the interpolation of the sky region is a little
bit improved.....

Peter

On Wed, Sep 04, 2002 at 05:22:10PM +0100, Major A wrote:
> 
> > Hello Andras,
> > 
> > did some test with your 'duster' and with a very similar own algorithm, see
> >  http://www.ciselant.de/projects/antidust/
> > 
> > Your duster seems to miss some pixels and/or areas of dust.
> > Maybe test if you hit enough random choosen pixel to calcuate the
> > interpolation and iterate some times over the nx,ny loops (rembering the
> > already interpolated pixels).
> 
> Great! Yes, I should make the radius larger as well. To be honest,
> I've never actually tested the code with slides taken out of a dog's
> mouth... the defects I have are usually quite small.
> 
> It seems mine gives fewer artefacts in smooth regions (sky) why yours
> looks better in detailed areas (buildings), where mine tends to create
> random noise.
> 
> Have you got any ideas as to how we can detect obscured pixels more
> reliably?
> 
>   Andras
> 
> ===========================================================================
> Major Andras
>     e-mail: andras at users.sourceforge.net
>     www:    http://andras.webhop.org/
> ===========================================================================
> _______________________________________________
> Sane-devel mailing list
> Sane-devel at www.mostang.com
> http://www.mostang.com/mailman/listinfo/sane-devel

-- 
------------------------------------------------------------------------
Peter Seiderer                     E-Mail:  Peter.Seiderer at ciselant.de



More information about the sane-devel mailing list