RAW vs White balance

Started Nov 20, 2006 | Discussions thread
GordonBGood Veteran Member • Posts: 6,308
Re: Unequivocal evidence - here too now

CurtisR wrote:

GordonBGood wrote:

Jonas B wrote:

Now I have to think for a while about this. In what situations can
one benefit from this? NB: The metering doesn't change. With the
pink towel exposed with a manual white balance the red channel is
clearly held back.

This does confirm what was stated in the Russian article that WB
does change the relative gains of the colour channels, but as per
your earlier tests and with these, only manual white balance
affects the gains. There is an advantage in this in that using
manual WB will avoid clipping of the red channel in cases with
extremely warm light, as in tungsten or candlelight. One might
assume it might also avoid blue clipping in cases with extremely
cool light as in extreme shade, etc., although this would be less
often a problem. Not bad, just interesting.

Um.. no. The WB is done pre-raw for preset WB as well. Go back
and look again. I just did it with manual to show it very clearly
(and demonstrate one potential practical use.) If you test with a
brightly exposed target, the difference in preset WBs is clear as
well - the russian guys show it well (and if you look closely at
JonasB's shots it's clear too.)

Sorry, I missed that, as Jonas himself did. Well, maybe that's good in that at least the camera is consistent, and it does avoid the chance of clipping the blue and red channels when the WB is extremely warm or cool if the preset is set close to being right. It would be a switch in thinking for those who just leave it set to Sunny all the time, I guess.

The second part of the observations by the OP was that there is a
discrepency in gain for all channels between ISO 200 and ISO 400.
As per your earlier post with the multitude of charts, it looks
like this is true in every setting for WB (I would assume for
Custom WB as well). With my knowledge of the Pentax raw format for
the K100D, I assume that there is the same offset on all channels
in raw of about 128 counts, so what we seem to be seeing here is
that the ISO 400 charts are just reduced by a factor of 2 which is
just another form of bit clipping - not using the upper half of the
range. I'm sure that if one looks in the raw file, they will find
that the stated maximum range reflects that the gain has been
reduced. In other range, there is just one ISO sensitivity setting
that is the real gain of the pre-amplifier and ADC: ISO 200. It
looks like all of the others are just software manipulated,
first just amplified by the maximum range setting for ISO 400, then
just bit shifted for each subsequent increase in ISO by a factor of

I'm not sure what you're saying, but it seems clear to me that
sensitivity up to, and including 800 are done without bit shifting
(or you'd see black stripes in the histograms as you do at 1600 and
3200). I think you're misreading something here. And the "offset"
that it seems you're talking about are just that there's no black
in the target, and with the lack of a log function on the
horizontal axis, its consistency gets exagerated when viewing these
histograms. It does appear that whatever is done to achieve ISO
400 and above changes the response curve somewhat, but it's hard to
say exactly how it is changed with only this data - something with
two sets of peaks would be more informative.

I'm not looking at the missing codes, which are clear indications of bit shifting for ISO 800 and above, including ISO 800 as Jonas B himself sees, too. I'm referring to the double width of the ISO 200 histogram as compared to all of the other histograms even though the exposure values were changed to compensate for the higher differences in ISO by using half/double the shutter speed with a fixed aperture. This was also the one of the observations in the OP. It looks like ISO 400 may be accomplished as compared to ISO 200 not by using a bit shift but merely be keeping the same gain and changing the maximum white point value so that the raw values will be scaled by a factor of two in the raw convertor (which is basically the same effect as a bit shift).

The offset in the data by 128 counts is real, at least if the DS works like the K100D, which is likely.

Regards, GordonBGood

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