Ad
Ad
Ad
Pages: « 1 2 [3] 4 »   Bottom of Page
Print
Author Topic: Dual Illuminant Profiles  (Read 38661 times)
Schewe
Sr. Member
****
Offline Offline

Posts: 5496


WWW
« Reply #40 on: May 14, 2012, 02:29:44 PM »
ReplyReply

Finally, bear in mind that dual illuminant profiles are just an optimization - so far as I am aware, no other camera company/raw developer has found the improvement compelling enough to adopt the idea.

Do you mean other companies don't think it's worth it or don't know how to do it? As far as I can tell, the concept of using two profiles for two different illuminants is somewhat unique to ACR...other companies understand that different profiles are needed for daylight and tungsten though. For example Phase One provides separate daylight (actually multiple flavors) and tungsten (multiple flavors) of ICC camera profiles. Capture One simply doesn't try to tween between them and requires the user to select the correct single profile to use. The white balance comes AFTER the correct profile is selected.
Logged
sandymc
Sr. Member
****
Offline Offline

Posts: 269


« Reply #41 on: May 14, 2012, 02:42:26 PM »
ReplyReply

Do you mean other companies don't think it's worth it or don't know how to do it? As far as I can tell, the concept of using two profiles for two different illuminants is somewhat unique to ACR...other companies understand that different profiles are needed for daylight and tungsten though. For example Phase One provides separate daylight (actually multiple flavors) and tungsten (multiple flavors) of ICC camera profiles. Capture One simply doesn't try to tween between them and requires the user to select the correct single profile to use. The white balance comes AFTER the correct profile is selected.

Hi Jeff,

Given that everything anybody might ever want to know about dual illuminate profiles is in the DNG spec, I'd say its "don't think its worth it". Separate profiles are quite common - I don't think that anyone would argue with their place. Temperature based interpolation  - well, if it involves rewriting your core processing algos it would need to deliver something pretty substantial, and I think (maybe wrongly) that the industry consensus is that clever idea though it may be, the cost doesn't justify the difficulty.

But that's just my outside-looking-in perspective.

Regards,

Sandy

Logged
Tim Lookingbill
Sr. Member
****
Offline Offline

Posts: 1193



WWW
« Reply #42 on: May 14, 2012, 04:40:43 PM »
ReplyReply

Sandy, what is your take on the tweaks to neutrals and primaries in julianV's image gallery profile comparison I described above? Clearly the color tables are being manipulated separately from R=G=B white in the parade image.

I wonder what the engineers measured from or based their tweaks in building Camera Standard and other canned profiles. The thing is the changes between the profiles are so subtle I can't figure why the profiles were necessary if just to attempt to copy the manufacturer's color rendering. Or maybe it's just more evident in that particular parade image.
Logged
Schewe
Sr. Member
****
Offline Offline

Posts: 5496


WWW
« Reply #43 on: May 14, 2012, 05:11:31 PM »
ReplyReply

Temperature based interpolation  - well, if it involves rewriting your core processing algos it would need to deliver something pretty substantial, and I think (maybe wrongly) that the industry consensus is that clever idea though it may be, the cost doesn't justify the difficulty.

I don't disagree that it is/was difficult but when you have a guy like Thomas Knoll and his sidekick Eric Chan, "difficult" takes on a new meaning, ya know? What may be very difficult for some isn't always so difficult for others.

And I do agree that other app developers who didn't try temperature based profile interpolation when they were developed would find it very difficult to re-write their apps to do so. There was a lot of discussion at the time when ACR was released WHY Thomas didn't use ICC camera profiles and part of his reasoning (as I recall) was that it made temperature based profile interpolation very difficult :~)
Logged
opgr
Sr. Member
****
Offline Offline

Posts: 1125


WWW
« Reply #44 on: May 14, 2012, 06:17:02 PM »
ReplyReply

Finally, bear in mind that dual illuminant profiles are just an optimization - so far as I am aware, no other camera company/raw developer has found the improvement compelling enough to adopt the idea.

They are not an "optimization", they are a "convenience". They are specific to Adobe's contorted way of processing, and should therefore never have been part of the DNG specs.
Logged

Regards,
Oscar Rysdyk
theimagingfactory
opgr
Sr. Member
****
Offline Offline

Posts: 1125


WWW
« Reply #45 on: May 14, 2012, 06:19:11 PM »
ReplyReply

There was a lot of discussion at the time when ACR was released WHY Thomas didn't use ICC camera profiles and part of his reasoning (as I recall) was that it made temperature based profile interpolation very difficult :~)

Would be interesting to know what the exact logic was behind that thinking. If you could ask him, that would be nice. As stated, it makes no sense.
Logged

Regards,
Oscar Rysdyk
theimagingfactory
opgr
Sr. Member
****
Offline Offline

Posts: 1125


WWW
« Reply #46 on: May 14, 2012, 06:27:13 PM »
ReplyReply

I would additionally like to add:

For very rough reference in order of temp:

tungsten or sunset: <3000K (very warm red tint)
sunny: ~5500K
cloudy (white): ~6500K
cloudy (gray): ~7500K
shade: as high as 10000K (very cold blue tint)

Therefore, if "open shade" means Cloudy with white clouds, then, yes, that would be the preferred condition for shooting the target.
if "open shade" means something else, then it is not the best condition. Neither is sunny…

And also: the Original Gretag MacBeth used to be "directionally sensitive". The color changed based on the direction of the incident light. It would therefore be useful to rotate the target and combine several shots. I don't know whether the current crop of color checkers still suffer this problem. I also don't know in how far this affects the results of any profiling solution…


Logged

Regards,
Oscar Rysdyk
theimagingfactory
Tim Lookingbill
Sr. Member
****
Offline Offline

Posts: 1193



WWW
« Reply #47 on: May 14, 2012, 06:53:55 PM »
ReplyReply

Quote
Therefore, if "open shade" means Cloudy with white clouds, then, yes, that would be the preferred condition for shooting the target.
if "open shade" means something else, then it is not the best condition. Neither is sunny…

Can you prove (with images) what happens to the color rendering applying a camera profile built from a CC chart lit from sunny, open shade or suboptimal lighting condition compared to optimal white cloudy conditions?
Logged
opgr
Sr. Member
****
Offline Offline

Posts: 1125


WWW
« Reply #48 on: May 14, 2012, 07:08:20 PM »
ReplyReply

Can you prove (with images) what happens to the color rendering applying a camera profile built from a CC chart lit from sunny, open shade or suboptimal lighting condition compared to optimal white cloudy conditions?

Yes, but only for my own profiling solution for my own RAW converter. I obviously have no idea what other applications are doing in this respect. Given that there are at least 3 aspects involved:

1. the profiling application which creates the profiles
2. the raw converter which uses the profiles
3. the raw conversion procedure which applies the profiles

Since 1 en 3 are usually different applications, from different manufacturers, I have no idea whether it even makes sense to try and mess with the profiles as provided by the manufacturer. But, if people here ask what the preferred lighting condition is for D65, and the answer is "open shade" then the previously mentioned table may help limit any confusion.
Logged

Regards,
Oscar Rysdyk
theimagingfactory
julianv
Newbie
*
Offline Offline

Posts: 28


« Reply #49 on: May 14, 2012, 07:52:21 PM »
ReplyReply

Sandy, what is your take on the tweaks to neutrals and primaries in julianV's image gallery profile comparison I described above? Clearly the color tables are being manipulated separately from R=G=B white in the parade image.

I wonder what the engineers measured from or based their tweaks in building Camera Standard and other canned profiles. The thing is the changes between the profiles are so subtle I can't figure why the profiles were necessary if just to attempt to copy the manufacturer's color rendering. Or maybe it's just more evident in that particular parade image.

Let me add a brief comment, just in case some people are wondering why I adjusted exposure when converting the files in that gallery.  This was a year ago, so my recollections are a bit fuzzy.  There was a point in time when Adobe's Camera Standard profile for the D700 was optimized to work best with a +3ev adjustment to exposure.  So I fiddled with exposure to get the images matched at a neutral mid tone.  I was only interested in comparing color effects.  Unfortunately, I don't have my notes on this anymore, and I don't remember which ones were adjusted which way.

I believe that the recent Camera Standard profiles do not require an exposure adjustment.
Logged
julianv
Newbie
*
Offline Offline

Posts: 28


« Reply #50 on: May 15, 2012, 02:44:05 AM »
ReplyReply

I wonder what the engineers measured from or based their tweaks in building Camera Standard and other canned profiles. The thing is the changes between the profiles are so subtle I can't figure why the profiles were necessary if just to attempt to copy the manufacturer's color rendering. Or maybe it's just more evident in that particular parade image.

Just a heads-up, in case you did not read the notes in the intro page of that gallery.  Those images are tagged with AdobeRGB.  To get the best view of the color differences, you need to be using a color-managed browser, and a calibrated wide-gamut display.  Better yet, download the files (use the "Share" link) and view offline.  Web sites and browsers sometimes mess with the colors.  Anyway, some of those files differ to a degree that I would not describe as "subtle."  It's particularly obvious when comparing Adobe's Camera Standard and the CC Passport "open shade" profiles (although some of that difference may be due to different tone curves).

By the way, I have another D700 custom profile from CC Passport that was made under direct mid-day sunlight.  I did not put images made with that profile up on the web gallery.  What I see is that the reds and blues become even more saturated than they appear when converted with the "open shade" profile.
Logged
sandymc
Sr. Member
****
Offline Offline

Posts: 269


« Reply #51 on: May 15, 2012, 04:44:42 AM »
ReplyReply

Sandy, what is your take on the tweaks to neutrals and primaries in julianV's image gallery profile comparison I described above? Clearly the color tables are being manipulated separately from R=G=B white in the parade image.

I wonder what the engineers measured from or based their tweaks in building Camera Standard and other canned profiles. The thing is the changes between the profiles are so subtle I can't figure why the profiles were necessary if just to attempt to copy the manufacturer's color rendering. Or maybe it's just more evident in that particular parade image.

I would guess, and I think that what Julian has said supports, that the differences are more in the color tables and small differences in exposure than in the dual illuminant stuff. The issue with LR3 and later is that there are hue twists in the Adobe profiles, and content-aware processing going on later in the chain. All of those mean that quite small changes in exposure, black level, etc can have an out of proportion effect, so its really difficult to isolate what exactly caused a change without a whole lot of painfully detailed experimentation.

Sandy
Logged
sandymc
Sr. Member
****
Offline Offline

Posts: 269


« Reply #52 on: May 15, 2012, 04:52:33 AM »
ReplyReply

There was a lot of discussion at the time when ACR was released WHY Thomas didn't use ICC camera profiles and part of his reasoning (as I recall) was that it made temperature based profile interpolation very difficult :~)

Yes. Although I think there were also other considerations. E.g, ICC profiles have no clue about CFAs, sensor calibration, etc. So in order to encapsulate all the information you needed to decode an image, you'd need an ICC profile plus something else. Something else being either another file, or embedding private data in the ICC profile. But if standard parts of the ICC profile can't represent everything you need, then why bother with it? Better to a custom container that's designed for purpose. DNG Camera Profiles do have the advantage over ICC profiles that they use EXIF formatting, and every raw converter already can decode EXIF, so they're easy to add.

Sandy
Logged
opgr
Sr. Member
****
Offline Offline

Posts: 1125


WWW
« Reply #53 on: May 15, 2012, 10:35:22 AM »
ReplyReply

Yes. Although I think there were also other considerations. E.g, ICC profiles have no clue about CFAs, sensor calibration, etc. So in order to encapsulate all the information you needed to decode an image, you'd need an ICC profile plus something else.

No, you don't.

Something else being either another file, or embedding private data in the ICC profile. But if standard parts of the ICC profile can't represent everything you need, then why bother with it? Better to a custom container that's designed for purpose.

That's the kind of thinking that results in tons of proprietary RAW formats no?

You should be aware that there have been a lot of very smart people working on the ICC standard, smarter than Thomas and Eric combined, dare I say? And there is a lot of solid color science behind the technology and thinking in ICC profiles. It is useful to stick to that standard initially in the same way that it is useful to consider that most color-correction requirements are attributable to white-balancing in the vast majority of the cases.

Stick to the basics initially, then further down the pipeline you can always give users the opportunity to eff it up again.

DNG Camera Profiles do have the advantage over ICC profiles that they use EXIF formatting, and every raw converter already can decode EXIF, so they're easy to add.

If that would be even remotely a consideration in development, then it's useful to know that the ICC profiles use TIFF formatting…
Logged

Regards,
Oscar Rysdyk
theimagingfactory
Tim Lookingbill
Sr. Member
****
Offline Offline

Posts: 1193



WWW
« Reply #54 on: May 15, 2012, 10:42:59 AM »
ReplyReply

Just a heads-up, in case you did not read the notes in the intro page of that gallery.  Those images are tagged with AdobeRGB.  To get the best view of the color differences, you need to be using a color-managed browser, and a calibrated wide-gamut display.  Better yet, download the files (use the "Share" link) and view offline.  Web sites and browsers sometimes mess with the colors.  Anyway, some of those files differ to a degree that I would not describe as "subtle."  It's particularly obvious when comparing Adobe's Camera Standard and the CC Passport "open shade" profiles (although some of that difference may be due to different tone curves).

By the way, I have another D700 custom profile from CC Passport that was made under direct mid-day sunlight.  I did not put images made with that profile up on the web gallery.  What I see is that the reds and blues become even more saturated than they appear when converted with the "open shade" profile.

julian, I viewed the parade images in color managed Safari. That's not the issue anyway. When I say subtle differences I'm referring to the lack of extreme overall off balance of color appearance. You have no blooming and loss of detail in the saturated primaries.

My tomato image is a demonstration of these extremes showing the dual illuminant's deepening and added magenta hue over the single illuminant's original orangish red tomato rendering which I find both aesthetically pleasing over whether they're accurate. My line of questioning to be to the point in this discussion is finding out if these are intended hue shifts through the software's interpretation of the spectral characteristics of the scene recorded by the sensor or possible metameric failure caused by tungsten light's weak blue spectrum as Jeff pointed out.

The level of saturation you should be gauging as in the parade image should be based on whether the profile causes detail to be lost in the form of posterization through "blooming". I don't see that happening in any version of your parade image. I get much worse blooming using a dual illuminant profile but easily fix by switching to single illuminant in similarly lit shots as yours. See the orange flower demo below.

But the one thing about dual illuminant profiles I can't live without is the way they maintain hue/sat levels over a wider range of colors shot under all kinds of oddball lights which I can easily fix pushing the WB slider to extreme ends as I demonstrated in the bluish WB of the tomato image. If that doesn't work I switch to Adobe Standard which neutralizes the entire tonal scale of the image where I have to adjust WB, usually the Green/Magenta tint slider.
Logged
digitaldog
Sr. Member
****
Offline Offline

Posts: 9185



WWW
« Reply #55 on: May 15, 2012, 10:48:06 AM »
ReplyReply

You should be aware that there have been a lot of very smart people working on the ICC standard, smarter than Thomas and Eric combined, dare I say?

I don’t want to say one is ‘smarter’ than the other but a really smart fellow, Eric Walowit who happens to be on the ICC digital photo committee has for years had a very interesting idea of how to do this with current ICC profile technology. It isn’t easy in that we’d need a small Spectrophotometer of similar device built into the camera to measure the illuminant of the scene and we’d need the spectral characteristics of the sensor and we’d have to write this as metadata into the raw data to build an on the fly profile.

This public post by Eric on the ColorSync user list in 2008 might be useful (I find any post by Eric has some useful information):

Quote
Subject:    Camera profiling with ICC et al
Date:    September 11, 2008 9:07:42 AM MDT


Chris, Edmund, Uli, Eric, and others,

Forgive me for coming into this thread late. I've been very remote and will
be again till next week. A couple comments:
1) I have the new camSpecs monochromator from Image Engineering designed
specifically for profiling by measuring camera spectral sensitivities out to the
near IR. Compared to lab-grade monos, its fast, inexpensive, repeatably, and
easy. I have no financial affiliation with IE except that I am a happy customer
and helped design it. I have measured all my cameras, compared the results to
lab-grade monos, and obtained very similar results. I've not seen near IR
issues with any recent cameras.
2) Once the camera spectral sensitivities, linearity, training data, etc. are
known then it is a simple and very accurate matter to determine the optimal
profile for any given illumination.
3) However, an ICC profile can only handle one illumination condition at a
time, though this does not preclude mixed (though not spatially varying in terms
of SPD) illumination if it is adequately characterized. This is a limitation
of ICC, though it is easily circumvented by building and embedding profiles
on-the-fly based on the scene illumination metadata or precomputing and storing
multiple profiles.
4) In my understanding, fundamentally, it is the latter approach that
underpins and is an advantage of Adobe's method compared with a
single-illumination-only ICC profile. Though Adobe doesn't use the ICC-wrapper, the Adobe profile
data itself is fundamentally similar (and probably not incompatible with some
massaging) to using multiple well-constructed ICC profiles. Adobe should be
commended for their approach, not berated.
5) Assuming that rendering from scene-referred to output-referred is turned
off (frequently not an easy assumption to check) it takes a very carefully
designed, controlled, and executed experiment to objectively evaluate the profile,
Adobe, ICC or otherwise. Much of what I have seen described here is simply
inadequate for validating profile quality.
Thanks,
Eric Walowit
Tahoe
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
sandymc
Sr. Member
****
Offline Offline

Posts: 269


« Reply #56 on: May 15, 2012, 10:57:56 AM »
ReplyReply

No, you don't.

Ok, I'll bite on that bait Grin

Where would you put e.g., the bayer green split, to randomly select my least favorite part of the DNG camera profile?

Sandy
Logged
opgr
Sr. Member
****
Offline Offline

Posts: 1125


WWW
« Reply #57 on: May 15, 2012, 12:32:06 PM »
ReplyReply

This public post by Eric on the ColorSync user list in 2008 might be useful (I find any post by Eric has some useful information):


Yes, but what is described there seems like overkill for most users. As a photographer, we generally don't want to *correct* the on-scene lighting conditions, instead we want to capture the mood of the on-scene lighting.

If I am in a forest with a lot of green light surrounding me, I generally want to preserve the green light, not completely correct it out of the image. Or if I capture a sunset scene, I generally want to preserve the warm colors dominating the atmosphere. While it would be useful to know the character of the light so I can correct it, perhaps just partially, we already know that that can be done quite adequately using tri-color profiles, and mostly very basic whitebalance tricks like a graycard or a white plastic lenscap/cup .

And the sunset example would be my preferred example to show that dual-illuminant profiles are merely a convenience, and may be confusing the actual capture process: i.e. we want to capture the atmosphere of a low temp scene by selecting the normal daylight response and only correcting the graybalance partially.

Whether this "normal daylight" response + graybalance correction is adequate under most circumstances, even tungsten, is indeed an interesting question. IIRC Magne Nilssen was the original creator of the capture one profiles, and he used to mention something to that effect in posts at the time.
Logged

Regards,
Oscar Rysdyk
theimagingfactory
digitaldog
Sr. Member
****
Offline Offline

Posts: 9185



WWW
« Reply #58 on: May 15, 2012, 12:34:23 PM »
ReplyReply

Yes, but what is described there seems like overkill for most users. As a photographer, we generally don't want to *correct* the on-scene lighting conditions, instead we want to capture the mood of the on-scene lighting.

I don’t know Eric is trying to correct anything but rather initially describe. This is supposed to work in a raw converter which doesn’t circumvent rendering to taste.
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
opgr
Sr. Member
****
Offline Offline

Posts: 1125


WWW
« Reply #59 on: May 15, 2012, 12:56:24 PM »
ReplyReply

Can you prove (with images) what happens to the color rendering applying a camera profile built from a CC chart lit from sunny, open shade or suboptimal lighting condition compared to optimal white cloudy conditions?

Just for illustrative purposes, I have attached a profiling example for Sunny daylight condition in my own calibrator.

Image 1 shows the result of combining 4 shots of the color checker rotating it 90degr. each time.
Image 2 show a single rotation sample, where the both the skin tone patch as well as the dark gray patch show significant deviation.

I'll try to find the cloudy daylight version as well, or some other version for comparative purposes. I have to immediately add that the resulting profiles were calculated based on variable blackpoint, which is not a useful method for Canon camera's. (One of the very important steps to consider in the entire processing pipeline, with significant effect on the end result).
Logged

Regards,
Oscar Rysdyk
theimagingfactory
Pages: « 1 2 [3] 4 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad