The code for hue is:
hmix = CLIP(in[0] * data->red[CHANNEL_HUE]) + (in[1] * data->green[CHANNEL_HUE])
+ (in[2] * data->blue[CHANNEL_HUE]);
'in' is the input pixel array. The code basically just multiply the
red, green and blue channels by the weights in 'data' and sum them up.
As you can see, the CLIP is applied only to the red transformation. If
the green or blue calculations results are out of range (< 0 or > 1),
the final result will be out of range. It is the same for the
saturation (smix) and lightness (lmix).
Further in the code we have this line:
rmix = CLIP((out[0] * data->red[CHANNEL_RED]) + (out[1] *
data->green[CHANNEL_RED])
+ (out[2] * data->blue[CHANNEL_RED]));
Here a similar calculation is done, but this time the CLIP macro is
applied to the final result, not only to the red piece. It is the same
for the gray (graymix), green(gmix) and blue (bmix).
So was it on purpose to clip only part of the calculation?
Cheers
--
Caio S. Souza
Laboratório de Biologia Teórica e Computacional
Universidade de Brasília
www.lbtc.unb.br
2015-11-27 8:30 GMT-02:00 Tobias Ellinghaus <m...@houz.org>:
> Am Donnerstag, 26. November 2015, 22:48:26 schrieb Caio S. Souza:
>> Hello!
>
> Hi.
>
>> I'm almost done implementing this feature, but I have two questions.
>> 1) I had to add one entry to the struct dt_iop_channelmixer_params_t
>> to store where the correction must be applied. Should I increment the
>> version in the DT_MODULE_INTROSPECTION macro?
>
> Yes.
>
>> 2) The final outputs of red, green, blue and gray channels are clipped
>> to avoid out of range values. But the clip is applied only partially
>> for the hue, lightness and saturation (on both CPU and opencl
>> versions). Was it on purpose or is it a bug?
>
> I haven't looked at the code, but in general we should try to avoid clipping
> wherever possible.
>
>> Cheers
>
> Tobias
>
> [...]
___________________________________________________________________________
darktable developer mailing list
to unsubscribe send a mail to darktable-dev+unsubscr...@lists.darktable.org
