[Digikam-devel] [Bug 188573] RW2 converter broken for low light images

[Matthias Welwarsky]

https://bugs.kde.org/show_bug.cgi?id=188573





--- Comment #18 from Matthias Welwarsky <matze welwarsky de>  2009-04-02 17:00:46 ---
(In reply to comment #17)
> First. You *need* to autoscale because RAW data is limited to some non-round
> value (i.e. 14735 for most ISO on Canon 5D mkII) and full-scale is 16 bit
> (0-65535).

Yes :-) That is due to the sensors quantization limit.

> Second. Autoscaling in linear space (before gamma conversion) works very well
> for *all* images (due to linear nature of RAW image). Autoscaling is virtually
> lossless (we need to round to nearest integer, but in 16 bit rounding effect is
> very tiny).

> The only question is autoscaling size detector. If you will use real maximum
> value from RAW, autoscaling will be affected by hot pixels. If you ignore too
> many bright pixels, you'll have too scaled shadows (as in your eclipse shot).

Exactly, that's it. But the hot pixel count is much lower than 1%, normally you
only have a few and they're not really 'stuck' pixels like on a LCD, they show
up under long time exposures only.

> May be, that 1% for highlights (clipped pixels) is too optimistic. It works
> well enough for most pictures, but photographers use different styles of
> exposure (exposure by meter, ETTR and so), so different setting for automatic
> scaling may be useful.

> I'll think about adding some autoscaling control in LibRaw postprocessing
> routines.

Cameras nowadays have 12bit to 13bit of sensor range (including noise). And
there is always noise, so maybe that's a valid criterion? You compute a
histogram, check what comes out as zero on the high end, then just scale to
fill that gap to 16bit?

-- 
Configure bugmail: https://bugs.kde.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.