Why bother changing base ISO?

Started Jul 18, 2007 | Discussions thread
Tom Christiansen Senior Member • Posts: 2,239
Re: Adobe problems

Julia Borg wrote:

Julia, might you happen to have any idea why Adobe converters
seem to do worse on Nikon noise than Nikon's converters do?

I tried to figure that out. it seems that applying of white balance
to RGB data through matrix transform is introducing more noise then
applying multiplication to individual channel data before
demosaicing. btw using multiplication only 2 channels are affected,
while with chromatic adaptation over RGB data all three channels are
affected and introducing additional cross-noise (noise from weak
channel affects stronger channels during demosaicin, and then,
additionally, during chromatic adaptation).

That makes some sense.

It's a bit frustrating to see people wave ACR on NEFs at you
as "proof" that Nikon is lousy at dealing with noise, rather
than saying ACR is.

RML demosaicing acts separately on luma and chroma; and then restores
exact values of channel data where possible. Adobe way does not
follow yellow brick road, accuracy is sacrificed for speed. repeating
demosaicing each time user changed WB is of course time-consuming.

In NC it drives me crazy on a crawling machine to change the
noise reduction or digital lighting, especially at zoom. Sometimes
I just walk away from the computer. They're pretty clearly trying
to compute the whole matrix, at least for display resolution, every
time you change white balance or contrast or mode. I understand
the tradeoffs. But I still think some caching would help. You should
be able to flip back and forth through history without so much pain.

Does everyone reverse engineer the exact
multipliers needed due to Nikon's idiosyncratic dye-color choices
between bodies by dissassembling Nikon's own converters, or do
they derive these number empirically, and potentially inaccurately?

I used disassembly for my table. however it turned out there are much
simpler methods

Set color temperature. Shoot known neutral. Use spectrograph
if necessary to measure how far from neutral resultant values appear
to be. Now see how far a given white-balance moves those
numbers, and in what direction. Now just apply inverse/reciprocal factors
and call those WB coefficients.

Something like that?

-- hide signature --

tom

Post (hide subjects) Posted by
3LX
tko
tko
RAL
RAL
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow