[Digikam-users] Digikam internal precision?

[Greg Kennedy]

On Wed, Jan 20, 2010 at 8:22 AM, gerlos <gerlosgm at gmail.com> wrote:
>
> Il giorno 19/gen/2010, alle ore 08.40, Gilles Caulier ha scritto:
>
>> 2010/1/19 Greg Kennedy <kennedy.greg at gmail.com>:
>>> But would it not help, say if you were doing repeatedly lots of
>>> operations, to prevent loss of quality due to imprecision in 8-bit?  If
>>> the 8-bit numbers were expanded to 16 bit (say mul. by 256) then you
>>> could do all kinds of operations on them, then save the result back as
>>> 8-bit rounded or truncated.
>>
>> With current code to convert 8 to 16, no, because expanded histogram
>> has holes everywhere. Color informations are missing.
>
> This is true, after we convert to 16 bits there are holes everywhere, but after a little tinkering with the image (for example denoising, blurring, curves adjust, ...) don't we get a more uniform histogram?
> Image editing don't spread that color informations around?
>
> From this point of view, don't you think that working in 16 bits and truncating them back to 8 bits could be useful?
>
> These are only my hypothesis. I'm sure you can give us better explanations. Or maybe we need some experiments with some images...
>
> bye
> gerlos
>
> _______________________________________________
> Digikam-users mailing list
> Digikam-users at kde.org
> https://mail.kde.org/mailman/listinfo/digikam-users
>

Right, that was my thinking: convert 8 -> 16 (or 24, 32, etc) gives no
benefit until you start doing image manipulations.  Sometimes I adjust
white balance, then further do brightness / contrast adjust, then
change the saturation.  All these repeated manips done in 8-bit surely
have a loss in precision, while 16+ bit would help to prevent that to
some degree.

(Yes I could probably do it all in one go with curves, but that's
unfamiliar territory for me...)