extragear/graphics/digikam/libs/dimg/filters

[Marcel Wiesweg]

SVN commit 512021 by mwiesweg:

digikam from trunk: rounding error of BCG modifier

With the current algorithm, even setting a gamma of 1.0 would create
changes, (x / 255.0) * 255.0 might be slightly less than x, so casting
this value to int would result in an error of 1.
I hope using the round() function is a correct fix for this,
at least the 1.0 value now creates a proper x=y mapping.



 M  +2 -2      bcgmodifier.cpp


--- trunk/extragear/graphics/digikam/libs/dimg/filters/bcgmodifier.cpp
 #512020:512021 @@ -152,7 +152,7 @@

     for (int i=0; i<65536; i++)
     {
-        val2 = (int)(pow(((double)d->map16[i] / 65535), (1 / val)) * 65535);
+        val2 = (int)round(pow(((double)d->map16[i] / 65535), (1 / val)) *
 65535); if (d->overIndicator && val2 > 65535)
             d->map16[i] = -1;
         else
@@ -161,7 +161,7 @@

     for (int i=0; i<256; i++)
     {
-        val2 = (int)(pow(((double)d->map[i] / 255), (1 / val)) * 255);
+        val2 = (int)round(pow(((double)d->map[i] / 255), (1 / val)) * 255);
         if (d->overIndicator && val2 > 255)
             d->map[i] = -1;
         else

-------------------------------------------------------
_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel

[Gilles Caulier-2]

Le Mardi 21 Février 2006 16:37, Marcel Wiesweg a écrit :

> SVN commit 512021 by mwiesweg:
>
> digikam from trunk: rounding error of BCG modifier
>
> With the current algorithm, even setting a gamma of 1.0 would create
> changes, (x / 255.0) * 255.0 might be slightly less than x, so casting
> this value to int would result in an error of 1.
> I hope using the round() function is a correct fix for this,
> at least the 1.0 value now creates a proper x=y mapping.
>
>

Fine for me.

If you have any time, HSLModifier don't work properly with 16 bits image (i
suspect a problem too with 8 bits image).

Can you take a look my comments in source code and give me your viewpoint ?

Thanks in advance

Gilles

Nota : HSL tool is the last one from core plugin to port in 16 bits (:=)))...

_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel

[Marcel Wiesweg]

> > digikam from trunk: rounding error of BCG modifier
> >
> > With the current algorithm, even setting a gamma of 1.0 would create
> > changes, (x / 255.0) * 255.0 might be slightly less than x, so casting
> > this value to int would result in an error of 1.
> > I hope using the round() function is a correct fix for this,
> > at least the 1.0 value now creates a proper x=y mapping.
>
> Fine for me.
>
> If you have any time, HSLModifier don't work properly with 16 bits image (i
> suspect a problem too with 8 bits image).
>
> Can you take a look my comments in source code and give me your viewpoint ?

As to 16 bit, there is an integer overflow in setLightness and setSaturation,
65535^2 = 2^32, so something like (i * (65535 + value) will wrap around for
larger values of i.
I think I have a fix for this, but now I have managed to break it even more
thoroughly, for 8bit as well :-)
I will work on this tomorrow.

Marcel

>
> Thanks in advance
>
> Gilles
>
> Nota : HSL tool is the last one from core plugin to port in 16 bits
> (:=)))...
>
> _______________________________________________
> Digikam-devel mailing list
> [hidden email]
> https://mail.kde.org/mailman/listinfo/digikam-devel

_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel

[Gilles Caulier-2]

Le Mercredi 22 Février 2006 23:58, Marcel Wiesweg a écrit :

> > > digikam from trunk: rounding error of BCG modifier
> > >
> > > With the current algorithm, even setting a gamma of 1.0 would create
> > > changes, (x / 255.0) * 255.0 might be slightly less than x, so casting
> > > this value to int would result in an error of 1.
> > > I hope using the round() function is a correct fix for this,
> > > at least the 1.0 value now creates a proper x=y mapping.
> >
> > Fine for me.
> >
> > If you have any time, HSLModifier don't work properly with 16 bits image
> > (i suspect a problem too with 8 bits image).
> >
> > Can you take a look my comments in source code and give me your viewpoint
> > ?
>
> As to 16 bit, there is an integer overflow in setLightness and
> setSaturation, 65535^2 = 2^32, so something like (i * (65535 + value) will
> wrap around for larger values of i.
> I think I have a fix for this, but now I have managed to break it even more
> thoroughly, for 8bit as well :-)
> I will work on this tomorrow.

Ok. thanks (:=)))

Gilles
_______________________________________________
Digikam-devel mailing list
[hidden email]
https://mail.kde.org/mailman/listinfo/digikam-devel