Current Server Time: 23:31 (Central Europe)

#apertus IRC Channel Logs

2015/01/04

Timezone: UTC


00:09
davidak
joined the channel
00:28
troy_s
intracube: White point doesn't matter on a sensor as the filters are fixed.
00:29
danieel
troy_s: hi... that is a quite delayed answer :))
00:29
troy_s
It matters, of course, when mixing media or trying to deduce accurate colors after capture, but not in camera.
00:29
troy_s
danieel: Yeah. I am slow.
00:29
danieel
btw regarding white point... do you think it is enough to get 2 poitns and interpolate between them to cover others?
00:29
troy_s
danieel: TCP = Tin Cup Protocol
00:30
troy_s
danieel: Explain?
00:30
troy_s
I don't think I understand your query.
00:30
danieel
the normal approach (sorry for pulling dng here) is to define 2 wp calibration points. one for 3200 one for 5600 k...
00:30
danieel
anythign in between is a compromise/average
00:30
troy_s
Oh no.
00:30
intracube
troy_s: hmm, but for arguments sake, if you videoed a black/white photo with typical sensor
00:31
danieel
or one shall do every 100K calib matrix?
00:31
troy_s
danieel: Actual white point is a curve through CIE XYZ, so no.
00:31
intracube
if the colour temperature was very low, lik 1800k, then the RGB sensels wouldn't fill up evenly
00:31
intracube
in extreme cases you'd end up with more noise on blue channel
00:31
intracube
and some perceived resolution would be lost?
00:32
danieel
intracube: we had almost no blue on tungsten...
00:32
troy_s
danieel: The only correct method is to use CIE XYZ positions and adjust the channels in the LMS domain using Sharp, Bradford, or CIECAT02 etc.
00:32
intracube
danieel: what are you referring to?
00:32
danieel
that is high math again :) not up for that at this time of day
00:32
danieel
intracube: noise on blue
00:32
danieel
we got almost no blue...
00:32
troy_s
intracube: Except the sensor is agnostic, and the filling of the wells is balanced against the primaries and band capture of the filters
00:32
g3gg0
left the channel
00:33
intracube
right, but overcast daylight 6000K probably matches most sensors 'native' colour temp
00:33
troy_s
danieel: Not high math at all. Hell Bradford is a super simple matrix set. ;)
00:33
intracube
if there is such a thing?
00:33
troy_s
intracube: Nope. No such thing as white!
00:33
danieel
i can confirm that... the KAC is quite representative at skylight
00:33
troy_s
intracube: Hence my point, it is all arbitrary.
00:34
intracube
but what about filling the RGB sensels with equal voltage/charge?
00:34
troy_s
intracube: And the ability of the sensor to record levels is all hardware that ends up agnostic to the light
00:34
troy_s
intracube: doesn't matter
00:34
troy_s
intracube: Matters in imaging in post for sure
00:34
intracube
o.O
00:35
intracube
but there will be a point at which the RGB filters in the sensor treat as 'white'
00:35
intracube
and you'd get maximum dynamic range at that point?
00:35
troy_s
intracube: Nope
00:35
intracube
isn't sure what he's talking about at this hour
00:35
intracube
ok, I shut up now :)
00:36
troy_s
White doesn't exist nor does it matter at all if equi-channel is perceptually achromatic
00:36
troy_s
Where it _can_ matter is in post imaging where you need a neutral axis to cut through the space perfectly
00:37
troy_s
Elle wrote a decent article on that (wall o text sadly)
00:37
danieel
intracube: here is how it goes: sensor =1=1=1 what you mean by white is not really white, as there is a matrix coming on top of that, which leads to other RGB than equal
00:37
troy_s
Google "nine degrees well behaved color space"
00:37
troy_s
Exactly
00:38
troy_s
Think of the sensels as a map into 3D space
00:38
troy_s
No position in that entire space is magic
00:38
troy_s
Just as no position in CIE XYZ is neutral
00:38
danieel
oh, i have an idea... we should program the channel bot, that if somebody writes that white or black is a color, he/she should be kicked from the channel :)
00:38
troy_s
White (and black) do _not_ exist
00:39
troy_s
And only exist in the display referred domain
00:39
troy_s
Naw. It matters. Huge even. But we need to keep our scene referred and display referred subject to different evaluations.
00:40
troy_s
In manipulation of images, having R=G=B being achromatic is a much bigger deal
00:40
troy_s
In data representation and gathering or expression, no real deal.
00:40
danieel
yes - i have an algo for saturation, which depends on R=G=B being achromatic, otherwise it would not work
00:41
troy_s
intracube: Sense?
00:41
troy_s
intracube: Ask yourself, if you are sitting in a room, what is white?
00:41
danieel
tough question
00:42
troy_s
And the moment that becomes irrelevant is the moment you realize how a sensor works.
00:42
troy_s
Likewise, think about a massively intense color that perfectly matches the bands of say, the R filter in a sensel
00:43
intracube
troy_s: I meant white only as far as equal charge in each of the RGB channels
00:43
troy_s
It gets completely full, yet (theoretically only) doesn't fill the things beside it
00:43
danieel
so treating everything as a wavelength distribution curve is the best?
00:43
troy_s
intracube: Right. That isn't white :)
00:43
troy_s
danieel: Well it is just coordinates isn't it
00:44
danieel
i think the distribution curve is lot more than 3 scalars
00:44
troy_s
Just like a CIE xyY map
00:44
troy_s
Really just data
00:44
troy_s
We give it meaning
00:44
danieel
the 3 coordinates are rather the three integrals of the sensel sensitivity x original distribution curve
00:44
troy_s
And again, if we are careful in imaging, we align our work for manipulation to our colorspaces
00:45
troy_s
danieel: Sure. Math nerd.
00:46
troy_s
My point is really that our contexts determine what matters, and R=G=B=achromatic is truly something that is an issue in imaging and manipulation, not in the sensel world
00:46
troy_s
intracube: Remember too that _no_ sensor is narrow band
00:47
danieel
that is not the biggest issue
00:47
troy_s
So it is impossible for R=G=B=achromatic in a sensor
00:47
troy_s
(Ever)
00:47
danieel
i see the worst, that the peaks of the sensel filters do not match the primaries of common color spaces
00:47
troy_s
Doesn't mattee
00:47
troy_s
We map the sensor to our needs.
00:48
troy_s
Ideally via a 3D LUT of course
00:50
troy_s
(So nice to see pixel nerd talk here)
00:51
intracube
troy_s: sorry, it's too late for me to follow
00:51
danieel
i think the pixel nerd is one who paints its logos pixel by pixel :)
00:51
troy_s
intracube: Think on it tomorrow
00:52
troy_s
intracube: It really isn't nearly as complex as it seems. Just remember that the sensel is _not_ an RGB system really
00:53
troy_s
It has red, green, and blue filters, but how those filter is not a perfectly single band like a true RGB color space.
00:53
troy_s
And even more important, the raw data really has no clue what white or black is in our perceptual sense
00:53
troy_s
It is just an ignorant beast
00:54
danieel
i have a question
00:54
fsteinel_
joined the channel
00:54
troy_s
If we had three holes that we filled with a shovel of dirt while listening to music, hole 1 could get it on a and d, hole two on b,c,e, and hole three on fg
00:54
danieel
an ideal rgb space - shall be treated like a 3 narrow spikes? or as three gauss distributed curves (think of 3 phase AC current)
00:55
troy_s
You can see having equally filled holes really doesn't matter when capturing Moonlight Sonata over time
00:55
troy_s
Well almost but not quite as I understand it
00:56
troy_s
Wavelengths do not correlate to color, due to psycho physical magic
00:56
troy_s
But R / G / B shoot upwards and out in straight lines withing X / Y / Z, which is a bent nonorthogonal system
00:57
danieel
(yeah, i was fighting with my sister over what is red and pink-purple... i do cameras and she paints)
00:57
troy_s
(Bent to isolate Y as luminance)
00:58
troy_s
So RGB (in imaging perfect model terms, not sensels obviously) traverse through XYZ that hold their primary color vectors perfectly
00:58
danieel
i found my answer... the narrowness just increases saturation
00:58
fsteinel
left the channel
00:58
troy_s
?
00:58
danieel
having a more flat/distributed primary results in "faded" / reduced color space
00:59
danieel
narrow is more pure color
00:59
troy_s
Ah
00:59
troy_s
Yes makes sense
01:00
troy_s
Non-narrow would fill up two or three channels
01:00
troy_s
And be less saturated of course
01:00
troy_s
But it still isn't quite a perfect analogy because wavelengths aren't color
01:00
danieel
why?
01:01
danieel
they map pretty nicely
01:01
intracube
night all
01:01
danieel
night
01:02
troy_s
danieel: Because, as Fairchild says, color isn't physical
01:02
troy_s
It is psychological and physical
01:02
danieel
he was a person?
01:02
troy_s
Hence psychophysical
01:02
danieel
we call the company unfairchild these days... :)
01:03
troy_s
Fairchild is arguably one of the most well known color researchers out there currently
01:03
troy_s
Mark Fairchild. :)
01:03
troy_s
Developed CIECAM 02
01:03
troy_s
And led Munsell labs for a while etc
01:03
troy_s
Heavyweight color peep
01:03
danieel
I should move to US... so much is going on there
01:04
danieel
we have only the ICC guy in switzerland :)
01:04
troy_s
ICC
01:04
troy_s
Great stuff for printing, not great for photographic stuffs. :)
01:10
danieel
time to rest...
01:10
danieel
gn troy_s, nice talk
01:10
troy_s
Back at you
01:10
troy_s
Let's have more
05:54
Bertl
off to bed now ... have a good one everyone!
05:54
Bertl
changed nick to: Bertl_zZ
10:28
davidak
left the channel
11:44
ItsMeLenny
joined the channel
12:00
g3gg0
joined the channel
12:10
ItsMeLenny
left the channel
12:12
ItsMeLenny
joined the channel
13:03
se6astian|away
changed nick to: se6astian
13:22
lab-bot
sebastian created T244: image pipeline: highlight recovery. http://lab.apertus.org/T244
13:31
Bertl_zZ
changed nick to: Bertl
13:31
Bertl
morning folks!
13:36
se6astian
good morning!
13:40
ItsMeLenny
left the channel
15:17
ctag
left the channel
15:46
rhavan
left the channel
16:59
rhavan
joined the channel
18:31
jucar
left the channel
18:52
davidak
joined the channel
19:05
lab-bot
sebastian closed T71: invite all crowd funding backers to join axiom-community mailing list as "Wontfix". http://lab.apertus.org/T71
19:13
lab-bot
sebastian closed T171: add cheeseplate concepts to wiki as "Resolved". http://lab.apertus.org/T171
19:26
lab-bot
sebastian closed T219: Bluetooth as "Wontfix". http://lab.apertus.org/T219
20:16
Bertl
off for a nap ... bbl
20:16
Bertl
changed nick to: Bertl_zZ
20:17
aombk
joined the channel
20:20
davidak
left the channel
20:20
davidak
joined the channel
20:40
davidak
left the channel
20:49
aombk
left the channel
21:29
se6astian
good night
21:29
se6astian
changed nick to: se6astian|away
22:52
Bertl_zZ
changed nick to: Bertl
22:52
Bertl
back now ...