Ad
Ad
Ad
Pages: « 1 [2]   Bottom of Page
Print
Author Topic: Camera Calibration  (Read 8727 times)
opgr
Sr. Member
****
Offline Offline

Posts: 1125


WWW
« Reply #20 on: August 16, 2012, 12:05:39 PM »
ReplyReply

Dollars to donuts you’ll see far, far higher dE values.

Speculation?
Logged

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

Posts: 9189



WWW
« Reply #21 on: August 16, 2012, 12:12:35 PM »
ReplyReply

Speculation?

Not at all. Think about it just a tad. Just subtract two dark captures from each other in Photoshop.

If you really must, I can capture two images and use the test described and come up with a dE value for a single camera. But really, you think a DSLR, the sensors and such from multiple camera systems are going to be as consistent as an Epson? Just what lighting should be used to capture the images and how consistent is it from capture to capture?

It seems others here have speculated about the quality of canned DNG profiles without an ounce of evidence to back it up. Speculation? I don’t hear the ACR/LR masses complaining about it do you?
Logged

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

Posts: 9189



WWW
« Reply #22 on: August 16, 2012, 12:38:33 PM »
ReplyReply

Oscar, what does this quick little test suggest? Two dark captures (lens cap on) from a 5DMII. Both exactly the same rendering in LR (default). Then use Calculations and subtract the two:



If the images were truly identical, every pixel in the image would be a solid level 128 gray. Pixels that aren't level 128 gray are different by the amount they depart from 128 gray which is useful. I’ve used Levels to exaggerate the difference, which makes patterns easier to see. With the Offset of 128, its far easier to see differences.

If I send R0/G0/B0 to the Epson, print two copies and measure, what do you think the results would be? Now this isn’t to suggest capturing a target of colors twice WITH THE SAME BODY let alone differing bodies wouldn’t be a better reality check. But this quick and dirty test kind of shows the difference in consistency between a capture device like a 5DMII and the existing hard data I have about the consistency of a pile of Epson printers no?
Logged

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

Posts: 1125


WWW
« Reply #23 on: August 16, 2012, 12:51:53 PM »
ReplyReply

What does dark noise have to do with the stability of your sensor? Does that dark noise influence your middle gray capture? By how much?

First of all:

There is nothing, absolutely nothing, in a DNG colorprofile that couldn't be stored in an ICC profile. An ICC profile is an open container, and as a container it is practically similar to a DNG file as a container. Any arguments set forth to promote DNG can be applied to ICC profiles. It's a standard, it's extendable, and private data can be stored there as well. This whole idea that DNG profiles are somehow better suited for the task at hand is absolute nonsense, and clearly shows that even Adobe doesn't take standards too seriously if push comes to shove.

Second:

The only thing relevant to color calibrating a camera is the stability of the sensor dyes and the stability of its linear response. All other variables are exactly similar in the printer calibration: if you think that contextual light has any relevance to the calibration process, than you should include that in the printer profiling as well. It isn't relevant, because it is divided out of the equation. In case of camera's at least we may assume that to be a linear device. Stupid printers have up to 12 non-linear colors and we calibrate them with only 3 primaries. Then you conclude that they have a stable delta E? Based on what? By measuring all possible combinations of the entire color cube?

Let me ask this:

You have custom profiled your camera and tell us that it is slightly better than the canned profiles.
What do you think causes the difference?
Logged

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

Posts: 9189



WWW
« Reply #24 on: August 16, 2012, 01:02:15 PM »
ReplyReply

What does dark noise have to do with the stability of your sensor? Does that dark noise influence your middle gray capture? By how much?

What it tells me quickly is that the capture device is far more variably and inconsistent than the Epson. You want me to capture an full color image, two in a row and subtract them now?

Quote
There is nothing, absolutely nothing, in a DNG colorprofile that couldn't be stored in an ICC profile.

That is a meaningless statement. The facts are, where in the imaging chain an ICC profile and a DNG profile are applied is vastly different (scene vs. output referred). The statement doesn’t bring anything useful to the discussion.

Quote
The only thing relevant to color calibrating a camera is the stability of the sensor dyes and the stability of its linear response. All other variables are exactly similar in the printer calibration:

Source, proof? Please provide some useful dE values of say, 4 DSLRs and I’ll supply the same from 4 Epson printers.

Quote
You have custom profiled your camera and tell us that it is slightly better than the canned profiles.
What do you think causes the difference?

I can’t say what the difference result from, but there are differences or the two profiles would be the same and produce the same results considering I used the same target and same DNG profile generation as Adobe. So let me present the question to you: If I use the same Macbeth as Thomas and the same DNG profile creation software, why are there differences in the profiles? My guess is my camera and Thomas is not identical. The differences while not huge don’t point to our 24 patch Macbeth’s. And if what you say about variably in capture devices is so low, why would the anyone need to build their own ICC camera profiles and please explain the vast vacuum in the ICC camera profile solutions.

Quote
Stupid printers have up to 12 non-linear colors and we calibrate them with only 3 primaries.

In the case of my data for Epson and the resulting profiles, there was zero calibration involved. Nor is it needed based on the consistency of the data. This isn’t to say the Epson driver could not be calibrated to produce a better (more linear) result. That isn’t possible with the native driver. We didn’t calibrate, we simply profiled the behavior. And as I’ve pointed out, the differences in consistency here is invisible to a human! Multiple Epson printers from across the US produced output that was half a dE from one another.
Logged

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

Posts: 1125


WWW
« Reply #25 on: August 16, 2012, 01:23:00 PM »
ReplyReply

What it tells me quickly is that the capture device is far more variably and inconsistent than the Epson. You want me to capture an full color image, two in a row and subtract them now?

Yes, and then I want you to tell me what delta E is represented by the difference. But it is hardly relevant, because you're simply showing the stability of a single pixel. We are interested in the stability of a group of pixels, which is exactly what you're measuring on a printer as well. If I single out a dot on the Epson, it will be either on or off, by random diffusion chance.

Also, I am not arguing against the stability of the Epsons. I am fine with those. I am arguing that the instability of camera profiling as mentioned in this thread is not based on facts.
Logged

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

Posts: 1125


WWW
« Reply #26 on: August 16, 2012, 01:36:09 PM »
ReplyReply

I can’t say what the difference result from, but there are differences or the two profiles would be the same and produce the same results considering I used the same target and same DNG profile generation as Adobe. So let me present the question to you: If I use the same Macbeth as Thomas and the same DNG profile creation software, why are there differences in the profiles? My guess is my camera and Thomas is not identical. The differences while not huge don’t point to our 24 patch Macbeth’s. And if what you say about variably in capture devices is so low, why would the anyone need to build their own ICC camera profiles and please explain the vast vacuum in the ICC camera profile solutions.

Considering that I have no access to the profiling software and the underlying technology, I am obviously at a disadvantage here. Clearly the differences shouldn't be in the MacBeth, they might be in the lens colorcast, but in my experience the largest differences come from selecting the blackpoint base line or multiplication reference.

The reason that most ICC profiling solutions don't work is exactly as you mention, but applying ICC profiles in the wrong spot of the workflow column is hardly a reason to conclude camera's to be inherently instable.
Logged

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

Posts: 9189



WWW
« Reply #27 on: August 16, 2012, 01:46:19 PM »
ReplyReply

Yes, and then I want you to tell me what delta E is represented by the difference.

Best I can do until you send me another 5DMII body (please) is tell you the dE differences from two captures of the same source. Popped camera on tripod, shot a color target twice, render, size down, load two images into ColorThink Pro, extract all colors, save out two color lists (166500 colors, very taxing on CT):

--------------------------------------------------

dE Report

Number of Samples: 166500

Delta-E Formula dE2000

Overall - (166500 colors)
--------------------------------------------------
Average dE:   1.63
    Max dE:  44.51
    Min dE:   0.00
 StdDev dE:   2.62

Best 90% - (149849 colors)
--------------------------------------------------
Average dE:   0.97
    Max dE:   3.69
    Min dE:   0.00
 StdDev dE:   0.74

Worst 10% - (16651 colors)
--------------------------------------------------
Average dE:   7.59
    Max dE:  44.51
    Min dE:   3.69
 StdDev dE:   4.91

--------------------------------------------------
--------------------------------------------------

Visually you can see this as well:



I suspect there could be SOME registration issues here (ideally I’d have a camera stand, mirror lock up etc). These two captures were shot with available Fluorescent light under my booth, that light source ALONE could be a factor.
Quote
Also, I am not arguing against the stability of the Epsons. I am fine with those. I am arguing that the instability of camera profiling as mentioned in this thread is not based on facts.

Source? Facts?
Logged

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

Posts: 1125


WWW
« Reply #28 on: August 16, 2012, 02:02:38 PM »
ReplyReply

Those numbers seem rather good.

1. Does colorthink compensate for uneven light over the target?

2. Is your lightsource possibly flickering @ 60Hz, could the exposure time pick up that difference?

3. Could you also measure the chroma difference alone?

4. Can ColorThink also compensate for whitepoint differences so we could test what happens for wildly different lightsources.

And less relevant, but interesting:

5. Does the Epson do better for the entire colorcube?
Logged

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

Posts: 9189



WWW
« Reply #29 on: August 16, 2012, 02:10:51 PM »
ReplyReply

Those numbers seem rather good.

1. Does colorthink compensate for uneven light over the target?

2. Is your lightsource possibly flickering @ 60Hz, could the exposure time pick up that difference?

3. Could you also measure the chroma difference alone?

4. Can ColorThink also compensate for whitepoint differences so we could test what happens for wildly different lightsources.

And less relevant, but interesting:

5. Does the Epson do better for the entire colorcube?


1. No. Do you think that such unevenness doesn’t ever exist even when capturing a target for creation of a profile? It certainly isn’t an issue for our Epson output.
2. It is possible. Do you think that such flickering doesn’t ever exist even when capturing a target for creation of a profile? It certainly isn’t an issue for our Epson output.
3. Not sure, I’d have to look further into the CT data. Bottom line is, one camera, two captures, even using the average dE, a heck of a lot higher than my Epson data!
4. Yes.
5. Don’t understand the question.

This is where we started down this rabbit hole:

Quote
ME: But there is a huge difference in consistency between an Epson printer and a DSLR

YOUR: Source?

So there is little question about the Epson. In terms of huge difference, well the Epson was around .5dE for multiple devices and the Canon, just one sample is 1.6dE so I’d say that is significant. Until you send me another 5DMII body, the differences in multiple devices can’t be answered. But 1.6dE from ONE sample is telling (you say it looks rather good). You still want to ‘argue’ that a number of DSLR’s will be anywhere as consistent in producing captures as multiple Epson’s as I originally guessed and if so, I’m ready for YOUR data.
Logged

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

Posts: 1125


WWW
« Reply #30 on: August 16, 2012, 02:30:37 PM »
ReplyReply

Oh, I just realized, you are measuring all the pixels in the image. Not the separate patches?

Kind of completely invalidates that test, because we don't know the direction of the delta E. The delta E for an entire patch could be 0 while the average per pixel delta E for pixels in that patch could be anything.

Again the instability of single pixels is not what we are interested in.


Logged

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

Posts: 1125


WWW
« Reply #31 on: August 16, 2012, 02:32:50 PM »
ReplyReply

So there is little question about the Epson. In terms of huge difference, well the Epson was around .5dE for multiple devices and the Canon, just one sample is 1.6dE so I’d say that is significant. Until you send me another 5DMII body, the differences in multiple devices can’t be answered. But 1.6dE from ONE sample is telling (you say it looks rather good). You still want to ‘argue’ that a number of DSLR’s will be anywhere as consistent in producing captures as multiple Epson’s as I originally guessed and if so, I’m ready for YOUR data.

Of course, that 1.6dE is a result of 10% of the pixels being out-of-wack. Any explanation as to why those particular pixels are out-of-wack? You're not suggestion that be the result of some instability in your camera, are you?
Logged

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

Posts: 9189



WWW
« Reply #32 on: August 16, 2012, 03:01:42 PM »
ReplyReply

Of course, that 1.6dE is a result of 10% of the pixels being out-of-wack. Any explanation as to why those particular pixels are out-of-wack? You're not suggestion that be the result of some instability in your camera, are you?

Do you have a DSLR and something like ColorThink or a similar tool? Because at this point, I’ve exhausted the testing for today and so far, you’ve made a lot of statements without anything to back them up. Rather than knock my admitted quick and dirty methodology to show dE between two kinds of capture, why don’t you provide your own methodology and results for us. Ideally using TWO or more cameras of the same make and model, the crux of the question you asked me in terms of consistency between multiple devices. Or just Fed EX that extra camera body to me...
Logged

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

Posts: 9189



WWW
« Reply #33 on: August 16, 2012, 03:05:57 PM »
ReplyReply

Oh, I just realized, you are measuring all the pixels in the image. Not the separate patches?

No. There was a Nearest Neighbor resample DOWN from the 20MP capture (5616 pixels) to something like 900 pixels along the long axis. There is NO way ColorThink can open a document that large.
Logged

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

Posts: 965



« Reply #34 on: August 16, 2012, 05:47:15 PM »
ReplyReply

Oh they can! I’m sure that Thomas and Eric produced stellar profiles for the camera bodies they used.

And what are those, I am truly curious...


Be very careful about making such large generalizations. The canned profiles for my 5DMII isn’t at all bad. The custom profile I build is a bit better.

You are not an average Joe ! that was the point.
Logged
deejjjaaaa
Sr. Member
****
Offline Offline

Posts: 965



« Reply #35 on: August 16, 2012, 05:50:24 PM »
ReplyReply

The facts are, where in the imaging chain an ICC profile and a DNG profile are applied is vastly different (scene vs. output referred).

not at all  - nobody prevents a raw converter to apply the data stored in icc container not exactly as ICC was/is thinking it shall be.
Logged
deejjjaaaa
Sr. Member
****
Offline Offline

Posts: 965



« Reply #36 on: August 16, 2012, 05:57:10 PM »
ReplyReply

So let me present the question to you: If I use the same Macbeth as Thomas and the same DNG profile creation software, why are there differences in the profiles? My guess is my camera and Thomas is not identical. The differences while not huge don’t point to our 24 patch Macbeth’s.

the speculation is that they are not using the same custom level (non individually measured) targets and not the same software...

And if what you say about variably in capture devices is so low, why would the anyone need to build their own ICC camera profiles and please explain the vast vacuum in the ICC camera profile solutions.

which vast vacuum ? there are enough such solutions...  the reason for sudden increase in dcp profiling is simple - it is an easy way to make some money for manufacturer because LR/ACR is widely used...
Logged
digitaldog
Sr. Member
****
Offline Offline

Posts: 9189



WWW
« Reply #37 on: August 16, 2012, 05:59:35 PM »
ReplyReply

not at all  - nobody prevents a raw converter to apply the data stored in icc container not exactly as ICC was/is thinking it shall be.

And then end users would build their own ICC profiles how?
Logged

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

Posts: 9189



WWW
« Reply #38 on: August 16, 2012, 06:08:08 PM »
ReplyReply

the speculation is that they are not using the same custom level (non individually measured) targets and not the same software...

Lets see if Eric comments. My understanding is they use a good old 24 patch Macbeth target. So nothing usual here.

Quote
which vast vacuum ? there are enough such solutions...  the reason for sudden increase in dcp profiling is simple - it is an easy way to make some money for manufacturer because LR/ACR is widely used...

IF it didn’t work well, was easy to use and affordable, it wouldn’t matter. Point of fact, creating ICC camera profiles are not easy, the targets can be expensive as can the software and either the market rejected them or the folks making the solutions rejected them or we’d see modern solutions for this usage.
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
nkpoulsen
Newbie
*
Offline Offline

Posts: 47


« Reply #39 on: November 17, 2012, 04:19:26 PM »
ReplyReply

Have any of you tried profiling your camera as if it were a scanner? I do art reproductions and I've been doing this for a short while now and its giving me good results. . .

I do this, and I've obtained quite good results.  I use the "old" PM5 software for creating the profiles.  My process is a little different; I convert using ACR.  I'm careful not to make any adjustments after converting the target.

After converting the photo of the artwork to which I've applied the resulting profile, I find I need to use some curves/levels adjustment to optimize the digital image.  I do this, while observing the artwork.
Logged
Pages: « 1 [2]   Top of Page
Print
Jump to:  

Ad
Ad
Ad