X-Rite Colormunki Photo and Profile Optimization Option

[jtoolman]

I just watched a webcast from a Ilford sponsored guy on youtube! Very informative for those looking to learn about general profile creation!

http://www.youtube.com/watch?v=n71OJIBac5E&list=PL2fVzV-M_S-0ZgpD3Fw6sSV8BOSZKEJR1

It basically clarified my question on what the CM software actually does when one optimizes an existing CM generated profile.
Assuming that the 100 patches total that the CM software generates is covering a sufficient range of separate colors, it is my thinking that the software basically is forced to interpolate between the relatively small number of colors ( though they may actually cover a wide range of color ) when compared to the much higher number of patches one can print and scan with the I-One for example.

So when you load a secondary image in the Colormunki software, if it contains colors that were not found within the 100 that were originally used, will it "ADD" that new data to the optimized profile?

If that is yes, you could also load say a 20 or so Black to White step wedge, and the system will look at those grays and generate a third set of patches representing the range of grayscale values that were in the step wedge but not found in the original 100 colors.
This "NEW" data is supposedly added to the original profile. Does it allow to print a more linearly neutral B&W? According to the video, it does.

So what if I load a tiff with hundreds of color patches such as those meant to be used with a high end profiler? Theoretically this type of image would contain many other colors which would have not been DIRECTLY present within the 100 patches on the two basic test prints that the CM software generates.

Would more data, obtained from those colors be added to the existing profile being optimized? Is it actually an improvement? If were to visualize the two profiles on a profile viewer, would the optimized one show a larger volume when compared directly to the un-optimized one? Or is this a total waste of time? I would tend to think that it is not. Why would X-Rite include the Optimization Option to their software if it was a waste of time.

If optimization is an additive process, which adds more and more color info with each iteration, then is there a limit? Can a profile be optimized several times with different images containing a large enough variety of colors which are not part of the original 100? Or does the original 100 colors, when properly interpolated encompass the a sufficiently large gamut?

I know that I compared a RR Aurora White profile from RR and the one I made and optimized and the most graphic difference was in the ability to be able to print deeper blacks. If you view the video I linked above, I came up with basically similar improvement in my own optimized profile. Total Color gamut was about the same, with the only improvement being that mine had more information on the bottom ( deep colors ) of the profile when viewed in 3D.

In an actual print, the visual difference is hard to see unless you view two prints side by side under very strong lighting. One has slightly more tonal separation in the very darkest regions where the other sort of blends together as a single dark not as defined dark shade. Of course you have to have the right image with a wide enough tonal range to test the optimized profile. Again, you have to look at the two together.

So I would love to hear more discussion from those you who unlike me, are profiling experts.
I have very basic knowledge but I am not a color engineer so much of what I know is only from what I've read here, elsewhere, videos and plenty of speculation and assumptions on my part. Often incorrect.

Thanks in advance!

 

[rodbam]

So what if I load a tiff with hundreds of color patches such as those meant to be used with a high end profiler? Theoretically this type of image would contain many other colors which would have not been DIRECTLY present within the 100 patches on the two basic test prints that the CM software generates.

Would more data, obtained from those colors be added to the existing profile being optimized? Is it actually an improvement? If were to visualize the two profiles on a profile viewer, would the optimized one show a larger volume when compared directly to the un-optimized one? Or is this a total waste of time? I would tend to think that it is not. Why would X-Rite include the Optimization Option to their software if it was a waste of time.
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Interesting post Jose, I'm not knowledgeable about colour management but I think I read somewhere that the law of diminishing returns sets in in after a couple of optimisations. So I don't know how many colours would go to improve the profile from a Tiff file with 100s of colours. Well worth a try though.

 

[jtoolman]

Someone in the Dpreview printer forum verified it! Yes he viewed the original and the optimized and he say more gamut in the darker regions when viewed with Color Synk ( MAC ).
In fact the Colormunki profile after optamization showed a slightly larger gamut in the dark areas as well as some green and yewllow colors that were beyond what the Manufacture's profile was able to show.
Apparently there is something positive to this process. I have a 1013 color patch tiff that I've been using to optimize the colormunki profiles. You can only optimize colormunki profiles by the way. Not others!
I don't have too much time right to do really thorough testing as I am working on some money jobs. WHen you are retired you need all the $$ you can get! All this research cost money!

Joe

 

[Emulator]

Joe your Youtube link in post #1 does not work, needs re-copying I think.
Ian

 

[rodbam]

Jose is there any chance we could use your colour patch or can you tell us how to make one?

 

[mikling]

Whether it is better will vary depending on what the situation is. A lot of math goes into these programs. The perfect profile (good or not good would require an infinite number of points). If you're into digital music and come from the analog world of recordings you'll kind of understand what the issue is. The number of points is essentially the number of samples.

Next you need to understand what interpolation is all about. If you've not been exposed to this, then try to understand estimating, because that is what it is all about.

Let's try the first thing. If a real line was straight and perfect, then you took two samples or points, then you can interpolate or draw from that the real line perfectly. No more samples or points are necessary.

now imagine a real jagged line with all sorts of kinks. How can you recreate that line from two sample of points? You cannot. No math in the world or anywhere will ever be able to do it.
Now imagine a line that is a simple curve. How many points do you need? To be perfect, an infinite number. To be "good enough" that is up to what "good enough" means to the user.

Now to make things even more complicated, think about how many points you need to define a "surface" and what if that surface is rough and has holes. Remember the surface would be indicative of what the printer paper ink can actually put out. You're essentially exploring all the possibilities of what the printer can do.

Darn, it spins your head now. Now there is some advanced math involved.

Now think about how do you explore what the total potential of the printer is........you have to make some guesses. This is where the differences between the packages come in.

 

[jtoolman]

Exactly right!

I do know how interpolation works, I worked for 28 years in a bio lab plotting data curves all say! Never sufficient enough time points from which to plot from! YUK!

So since the Colormunki basic profile is based on only 100 colors which supposedly cover a large enough range of colors with very large gaps in between each one, the software is forced to interpolate and "figure out" what the in between values should supposed to be.

I don't like guessing!

So if we can add images during optimization containing colors that physically fill in some of those gaps, regardless of the "Math" involved ( that's why we have software )the program should not have to Guess or Interpolate since it has new real information to compute new data from. Home many times can we do this process? Not sure!

Now back to the CM.
If you have ever done an actual optimization of a CM profile, you know you have to load an image. The idea behind this was that the image that you load would be the actual image you were attempting to print but one that printer had some problems reproducing certain colors and values.

It then checks for colors values it did not have real data on, creates a tertiary set of patches and after printing and drying, they are scanned into the software, which then adds those NEW bits of data to the existing profile to fill in some of the interpolated data with real data.

So now that new but real data has been added it would not have to guess how to best attempt to reproduce those new colors that it would have had to guess on if just printing through the original non optimized profile.

If my explanation is completely wrong please feel free to correct me. I need to learn this stuff.

So as Mikling explained with his analog to digital comparison, one could assume that for a profile to be "Perfect" you would need an infinite number of data points covering every single variation of ALL colors and values.

How many sampling points are enough? At what point to our eyes fail to identify differences between these points?

So yes there is a basic limit to this. The so called theory of diminishing returns kicks in.

So I would think that doing a two step optimization of the basic CM profile would be PLENTY.

1st one with a 1000 plus color patch TIF ( I have a 1013 patch image generated by the i-ONE )

2nd one with a 20 black to white neutral step wedge TIF

Oh and if we use The Pharmacist's method for printing combined charts on a single sheet of A4 / Letter paper we can save on material as well.

Joe

 

[mikling]

The other assumption here is that the printer and ink combination is linear and behaves the way the designers of the printer had wanted. If for example OEM ink and paper was used, you could get away with very little points because the way the printer behaves fits the math that the profile equation is expecting. Now if for example you used substituted cyan instead of magenta and used the same amount of points, then you're going to be in trouble. It becomes obvious when you take it to the extreme.
Let's do it.
So suppose you take OEM ink and then only use 100 samples and compare this to a bad set of inks, it would not matter how many samples you took with the mismatched inks, you could not correct the problems even with 2000 points.

That is why it is not necessarily how many samples you take but whether or not the shape that is produced is smooth and easily defined. This cannot be predetermined. To me it is obvious that you get the printer printing as close to the real outcome as possible...then profile. If the output has a lot of problems without a profile..the profile is going to need a lot of work/data to define and correct the problems...and the results even with a profile might be unacceptable.

What is obvious is that you want to get the printer operating as close to what the designers wanted first then go create the profile.

What is not obvious is that greater gamut volume does not necessarily mean better...not if the shape is ragged and irregular. This is again similar to audio speakers. It is not necessarily that true that the speakers having the widest range will sound better but that the speakers be smooth in particular ranges might indeed be much better especially if the extended ranges are inconsequential anyways. So would you take the same analogy to gamut volume? Is it the total volume or is it that it is where it is needed and it is predictable where it is that is important?

I think that is where the CM takes it. It says, you don't like this range of colors? Let's try to fix it. Then it takes more data to see if it can better define it. But wait. If the range is indeed rough due to the print combination itself. Can anything fix it? Hmm...more food for thought.

 

[jtoolman]

I realize everything you said, so does CM Optimization help or not????
My profiles are slightly larger, extremely smooth, no spikes, broken areas, and the resulting prints when compared to those printed with an unoptimized profile are consistantly visually better.
So unless I and others are seeing things, it could be bacause maybe there is indeed an improovement?
Specially when compared to profiles from paper manufacturers.
They will often try to cover all variabilities on a partucular model printer so the resulting profiles that are more conservative and often come a bit short when it comes to permitting what a particular printer is really capable of outputting!
I get much better shadow detail and often less compression of saturated colors after I create my own profile, optimize then compare it to the print done with the manufacturer's version.

If it is an useless feature then why is it included in the software?

Joe

 

[Grandad35]

A few comments that don't answer any questions, but are just meant as food for thought:

With R/G/B independent inputs, if we generate a cube with 5 uniformly spaced inputs across all 3 colors, we need 125 color patches. This would allow 4 line segments in each direction for interpolation. Obviously, 100 patches leave some rather large holes.

Similarly, 1000 patches allow 10 uniformly spaced samples, giving 9 line segments across each color. As was already stated, if the responses were linear across all 3 colors the results for both sets of patches would be the same. But since the responses are not linear more segments give a better fit.

As was also already covered, some areas of the cube need more closely spaced samples than other areas. For example, the areas close to gray need more samples. The histograms for the .tif images with 918 patches used by my I-1 system are shown below. The histograms show 9 equally spaced samples (8 line segments) over the entire color space, using 729 of the 918 patches. They also show that about 20% of the readings are used to give some 17 segment samples (16 line segments), presumably close to the grays. There are also a few even more closely spaced samples near each end.

If you want to add neutral patches, remember the patches say if you ask for the xxx (gray) patch color, you will actually get the yyy color, which will probably not be gray. The printing software has to figure out which color to ask for that will give a neutral color printed, and it doesnt have that data to work with. If you want 21 patches of gray, and wanted to get additional data at (say) +/-5% on each of the colors to guarantee that you bracketed gray at all intensities, you would also need 8 more sets of data, for a total of 189 patches.

I have no idea of how the profile software actually works, but it is easy to imagine that the interpolation is based on equally spaced line segments. As such, you might want to look at the color values that they use and make your new color values fall at the midpoints of the existing values (as the histograms show).

Its easy to forget that you are talking about printers that use 8 inks, based solely on 3 color inputs. An internal RIP has to decide which of the 8 inks as well as the amounts of each to use for each set of RGB values. These RIPs are obviously optimized for OEM ink sets, and they may not work all that well if your inks arent very close to the OEM ink colors. Ponder the problem of the RIP in transitioning from K/DG/LG for perfectly neutral colors to C/M/Y/K/DG/LG and then to C/M/Y/K as they colors shift away from neutral. What happens if your C/M/Y colors are even slightly different from the colors used to develop the RIP? Its easy to imagine some weird swings in the printed colors with even small changes in the RGB input values and why lots of patches are required in the near-neutral area.

Similar, but less severe, problems exist for the transitions from PM to M and from PC to C.

Things are much simpler for printers that use C/M/Y/K ink sets, and getting a good profile for these printers should be much easier since the profile doesn't have to compensate for a sophisticated RIP that may not be perfectly matched to the ink set.