New Calibration Tool, Excel Download - Page 3

I'm running Vista Ultimate Edition on my Laptop which has Works v9.0 . Your Spreadsheet won't load . The error message says there are 290 faults and then it shuts down . Just a version to version incompatibility I'm sure .

Is there any way to "translate" to a newer version ? Previous Works platforms would at least attempt it . I've tried all the available Options in the pulldown list under "Open" but still get the errors . Any ideas ????

Scott................

Converted to excel 2007, dont know if this will help.

 

Gamut, Saturation, Brightness, dE Chart converted.zip 438.06640625k . file

That's an interesting technique which I have not yet tried. On my Samsung LCD, I had the impression that there was a gradual curve from 0 through 50% back to 100% saturation for some of the colors. But you may be right, that it may be a nearly straight line from 0% to 75% and all the twist happens at max saturation.

So, to handle inaccurate primaries, I take it you measure Y of 75% white, then use Accupel or angryht's spreadsheet to calculate the coordinates of the adjusted 75% secondaries based on the actual 75% primaries?

Thank you for this attempt . I get a little further and with less error messages but it still won't load properly . I think I'll try and find the Excel Disk that came with my Desktop Win/XP system and see if Vista will let me install it alongside Works 9.0 .

Thanks again ,

Scott.............

Just to be absolutely clear, you are saying that you use the 75% saturation to measure the primaries and secondaries, then you measure 100% stimulus white and use those numbers to determine the percent luminance of each color. Is that right?

angryht,

Yes, that is what you would do because I believe the 75% saturation patterns are 100% stimulus (i.e. brightness). I may be wrong but I don't believe there is a disc with 75% saturation and 75% stimulus color window patterns. I know all of the saturation patterns on the HCFR disc are 100% stimulus. It would be interesting though to make a set of color window patterns that covered the entire 2x2 matrix (25/50/75/100 saturation and 25/50/75/100 stimulus).

cheers,


--tom

Thanks, tom, I would like to check 75 percent stimulus of the 75 percent saturated. I did a couple checks with the 75 percent saturation and 75 percent stimulus and got some different results in the color calculator. Now I'm not sure what to target. Take a look.

 

color corrector compare.zip 11.2783203125k . file

I don't know of any disc with 75%/75% saturation/stimulus patterns on it.

So, what you did was measure the 75% saturated/100% stimulus patterns (off the AVSHD disc?) and then put the numbers into the color gamut calculator to get a custom color space given those measured primaries? What readings do you get with 100% saturated/100% stimulus patterns? I assume what you want to do is calibrate according to your measured 75% saturation points? I suspect there must be a logical reason why luminance (especially blue) luminance seems to be so different but I'd have to think it through a bit more.

cheers,


--tom

That's correct.


I'll have to check that. I usually stick with the 100% sat/75% stim (like the file above).

I guess I'm not sure. I have always calibrated based on 100% sat/75% stim but I'm curious about the 75% sat.
Yeah, I'll do another set of measurements just to make sure I didn't pull up the wrong test pattern. It does seem odd.

Thanks!

angryht,

After a little thought, my hunch is that using the custom color gamut calculator to determine a custom gamut given your measured 75% saturated primaries doesn't really make logical sense. The idea of the calculator is to define a custom gamut given the 3 primaries (100% saturated red/green/blue) but what you're measuring is not that. i.e. the pattern that you are using to measure red contains blue and green as well thus the assumption that it is "full" red is not true. So, while the calculator is returning correct results given your white point and primary x,y points, it's not actually what you're measuring with those patterns. Hopefully I'm making some kind of sense and someone will correct me if my logic is not right.

cheers,


--tom

I puzzled a little about this, too, but talked myself into the idea it does make sense.

As I understand what the calculator does, it returns the secondary colors that are the exact complements of the primaries, i.e., the exact reflections through the white point. So it should calculate the correct xy coordinates.

Your observation about the mixture should have some very small effect on the Y value, if you are calibrating to a gamma other than 2.20, as the AVSHD saturation patterns are created with flat Y values at this gamma, but the effect is very small. Given that one uses this procedure to provide the best calibration of an imperfect set, I would think those imperfections are much larger than the effect of a gamma near but not equal to 2.20.

If one wanted perfect precision, I think you're right. One would need to calculate the theoretical 100% saturation points from the 75% saturation measures, calculate the 100% saturated secondaries for those primaries, and then calculate the corresponding 75% saturation secondaries. But if my understanding is correct, the results would be identical at gamma 2.20.

angryht, you may be interested; I've added a second worksheet to your spreadsheet to calculate CIE 1994 deltaE values of the errors. I want to try comparing the measured primaries against the Rec709 colors using dE 94 (where Accupel/HCFR uses CIELuv), as well as dE 94 to compare the measured colors against the calculated color gamut.

To help me verify the dE 94 calculation, I tweaked your spreadsheet slightly so it would always use D65 white to calculate the new gamut. That way I could paste HCFR's measures of the actual white for comparison. I compared the calculation of the errors relative to Rec709 against one of Tom Huffman's spreadsheets, and it looks right.

Hopefully, I'll get a chance to use it soon to see if it leads me to any changes in my calibration.

Enjoy,
Bill

P.S. This technique of defining a new gamut may work best if you have only Color and Hue controls. If you have a CMS, better than creating a new gamut from the actual primaries is to tune the actual primary Y values to minimize dE relative to Rec709. See this later note.

 

color calculator.zip 21.1240234375k . file

Thanks, Bill. I tried to open it and I got an error. What version of excel did you use? I noticed that the extension is .xlsx, even when I changed it to .xls it gave me a bunch of errors.

My fault. I forgot it was in Excel 2007 format. I re-uploaded it as a .xls in the link above.

Thanks, Bill, you da man!

Dan, in using your spreadsheet last night to look at a set of results, I ran into a peculiar situation, where my Saturation-Luminance graphs were all low, and in fact the RGB luminance at 100% saturation of the three primaries were all low. What apparently was happening was there was a noticeable difference between the 100% White Y I found when I measured the grayscale, and the 100% luminances of the saturation measures when I made them several minutes later. Perhaps random variation, or time or temperature drift in the set or in the Eye One -- I don't know which.

I found the saturation graphs made more sense when I changed the Input Data worksheet to calculate the 100% white value (cell F2) as the sum of the red/green/blue 100% Y values (cells F6, F11, F16). This avoided the measurement variation I was seeing in white, and the resulting Saturation-Luminance graphs correlated very well with RGB balance errors and gamma variation in the grayscale. And it makes sense to me, as unless something hinky is going on, 100% white should be identical adding 100% red, 100% green and 100% blue in the RGB space.

As your initial impetus was, at least in part, that HCFR scaled each color independently based on its own 100% Y value, I'm thinking that perhaps they were avoiding a choice of whether to calculate the white Y value or of which white measure to use.

Thanks again for working up the spreadsheet,
Bill

I've had a chance to use the spreadsheet with the deltaE 94 measures in tweaking my set, and I found it awkward to use, with the color input data on one worksheet and the error measures on the second. So I've uploaded an improved version, where I copied the results of the error calculations back to the main page under the color values, and added a selector to choose among Rec601, Rec709 or PAL.

angryht, I don't know if you followed a recent discussion with Greg Rogers about how he intended the color calculator program to be used. In a discussion beginning about here, Greg was pretty insistent that our use of the calculator to define a new gamut at the actual primaries was incorrect and would not yield the most accurate colors. Interestingly, it became clear he was not recommending that one blindly calibrate to match the Rec709 coefficients. Rather, the correct solution to minimize error should be whatever combination of Y values minimizes delta E against the Rec709 points. Because of the way delta E is calculated, minimal error may not be at the point where the Y value at the actual primary equals the Y value in Rec709. I did an Excel experiment that supported Greg's recommendation, that using the Excel optimizer to find the Y values that minimize deltaE should reduce error over all the saturation points, compared to the strategies of defining a new gamut or matching the Rec709 coefficients.

So I have been experimenting with a new spreadsheet to try to make this optimization easy to do using the Excel Data Solver. One enters the current color measures from HCFR, then use Excel's Data Solver to give the optimum Y values for the six primary/secondary colors that minimize total deltaE. The spreadsheet has a couple of other advantages, for what-if analysis. There is a choice of error measures, CIELUV75, CIE94 or CIE2000. And the calculated values includes an elasticity of delta E with respect to Y, C and H, essentially how much dE would change at each point from a 1% change in Y, chroma, or hue.

Trying to make it easy to run Data Solver by using Excel macros was a little tricky, and it does introduce a minor headache that one must reduce one's security settings in Excel to run macros from an unsigned source. But for anyone bothered by this, it is still possible to enable the Data Solver add-in and run Solver manually without the macros.

I would be interested in hearing if the spreadsheet is easy enough to use, and if anyone else finds this approach useful.

By the way, Dan, I have now copied the deltaE calculations into enough places that I may be ready to try to introduce a choice of delta E calculation into your spreadsheet that correctly draws the saturation graphs.

Enjoy,
Bill

 

color Y solver.zip 247.55078125k . file

Hey, Bill. Yes, I read the exchange you're talking about. There are still some things that don't add up in my little head. Have you read Tom Huffman's series of articles in Widescreen Review? In Part II he has the following quote:
So like everything else, I'm still a little confused.

Greg and Bill seemed to suggest that we were misreading Tom's advice, that it was meant to apply where there was not a full CMS that would let us move the primaries near the Rec709 targets, or perhaps that it was not meant to apply when one has a 6-axis CMS.

Were there only two possible strategies, matching the Rec709 coefficients or the coefficients for a gamut defined by the new primaries, I would prefer the latter recommended by Tom. It has a certain esthetic. In my analogy that the Y values represent a tent over the CIE space, attached at seven tent poles, it has the nice characteristic that the tension is the same in all six pieces of the tent. And when I simulated its performance compared to targeting the Rec709 coefficients, Excel seemed to say it did better too, consistent with some of us believing we saw better results on our sets.

That's when Greg corrected me and pointed out that there was a third choice, to adjust the Y values at each point to minimize the error compared to the Rec709 color. Which is not the same thing as matching the Rec709 coefficients. Almost by definition, this should yield the most accurate colors relative to Rec709. And if our concern is to get the best skin tones, wouldn't these be achieved with the most accurate colors in the red/yellow region?

The tricky part, and perhaps one of the motivations for Tom's advice, is that, without a computer solution, finding this optimal choice of Y values takes a lot of trial and error. If one is in a normal situation, where 100% white is the sum of 100% red/green/blue, and the coefficients all add to 1, there is a real tradeoff in adjusting Y to minimize error at one color, that it introduces error at the others. A third set of target coefficients can only be uncovered through a computer solution.

So far, I like the results I see on my Samsung LCD. Of course, as the eye adjusts to what it sees, it is difficult to detect small differences between the current calibration and the previous. And it still leaves open the possiblity that this computer solution is peculiar to sets like the Samsung where the colors nearest Rec709 are undersaturated.

Unless I'm missing something (not at all unusual, I know) the amount of trial and error isn't that significant.

Using Greg's Accupel calculator, selecting the Rec709 standard, inputting the xy values for RGB and xyY for white, all that is necessary is to input various Y values for a given color and then hit "calculate". dE is immediately shown, and the next step is to increase and/or decrease Y until the minimum dE is "found". Then adjust the TV to that Y target. Repeat this for every color. Maybe it should not work this way, but changing the Y value for any single color only changes the dE for that color. Maybe this is part of what Greg talks about with the 6 axis CMS where each adjustment is more independent. (?)

Now with our Samsungs' 6 axis CMS the xy coordinates of the secondaries can be brought very close to Rec709 standard, but the calculation to find the minimal dE for each secondary's Y would still need to be done.

It would be necessary to measure and make sure these changes don't effect the white point, but from having done quite a few calibrations I don't believe they will.

I have not done a calibration with this trick of searching out the lowest dE compared to Rec709 for all the Y values but I look forward to seeing how it looks.

I have been using the wrong way for several months and have been very happy with colors. (Wrong in that I have adjusted to minimal dEs for all 6 colors to the xyY values produced by the calculator using my native RGB xy and white xyY values, as opposed to minimal dEs compared to Rec709)

One can see the numeric difference between the two calibrations by using the two spreadsheets. You can take the earlier color calculator spreadsheet, or the Accupel calculator, enter your measured xyY values and see recommendation for creating a new gamut based on these primaries. Now cut-and-paste the actual measures into the new color Y solver spreadsheet, overlaying the primary Y values and the secondary xyY values with the recommendation from the color calculator spreadsheet. This will give you a calculation of the delta E relative to the Rec709 colors, as if you had done a perfect calibration using the technique you described.

Now using the color Y solver spreadsheet, press the button to calculate the optimized values. You can see the delta E measures, again relative to Rec709, from using these Y values.

As an example, using the earlier color calculator spreadsheet, I had RGBYCM coefficients and delta E(94) of:
Calculating Y values using a minimization strategy yields:
You can see that the recommended Y coefficients differ from the Rec709 coefficients:
where the optimizer traded off a little more error at green to reduce the error at red and blue.

To some extent, this seems like a discussion of the number of angels on the head of a pin, as the delta E numbers in all three cases are acceptable. Still, aiming for the best target should give the best results. As the Y values do not match those calculated for a new gamut based on the primaries alone, using my tent analogy, they stretch the tent tighter over the color xy domain a little more in some places than in others, i.e., a better fit on the Rec709 primaries and secondaries may give a slightly worse fit on some of the less saturated colors. So only by trying it when calibrating a set can one decide if it looks better, the same, or a little off.

When you do get a chance to try it out on your set, let me know the results. So far, I have a data sample of one, mine. But I cannot fault Greg's logic here.

I played with some old data I have from some calibrations using your spreadsheet and Greg's Accupel.

One difference I note that concerns (OK, confuses) me is that with the Accupel calculator changing (for example) the measured Red Y value results on change in dE only for Red. With yours it changes the dEs of everything.

I'd be interested in hearing what the justification for the difference is.

BTW ... what exactly IS your "minimization strategy"? Is it the minimum sum of all dEs? Minimum "highest" dE?

I am making the assumption, like the coefficients of Rec709 or those calculated by Accupel, that the sum of the red/green/blue coefficients should add to white in a typical set. If you raise the red Y, I believe on my set the white Y changes too. And when the white Y increases, the green/blue ratios go down. If they are not really independent, as ratios of white, this is where a computer minimization strategy is ideal.

In case my assumption about the interaction of RGB Y and white Y does not hold, there are two other "coerce white Y" options built in: "Yw uncoerced", where one can say that white Y is independent of RGB, and the second, that white white can be "less than or equal to" the sum of RGB, saying you can push the primaries higher while keeping white Y where it is with the grayscale gains.

This is one of the interesting effects of the model that seems to match my earlier experience calibrating one color at a time. I would change Y to minimize the errors on one color, then remeasure all the colors and find that the others had gotten worse. By constraining white Y to be the sum of RGB, I find the calculated recommendation is one that I can achieve on the set, within the limits of the granularity of the controls and measurement variation of the EyeOne.

Excel Data Solver can optimize one target value. I chose to minimize the sum of all dEs across the primary/secondary colors. One could consider playing with alternatives. The maximum, but then you would have to include the others too, to indicate that you don't want to add too much error elsewhere just to pull the maximum down a little. Or the sum of errors across the calculated saturation points, and I have included a "saturation points" option under "Optimization target" to use these. But the accuracy of the model on the saturation points is dependent on assumptions about how those are modeled and whether the model matches the logic inside the set. Consider just one color, e.g., blue. The prediction is going to be different for the partially saturated 0%-75% blue points depending on whether one calculates using the blue-white-magenta triangle or the blue-white-cyan triangle. I trust the model to give a feeling for the overall effect at the saturation points, but I'm not sure about using them as a target.

Angryht,

Since your projector doesn't have any CMS controls (if I am remembering correctly) then adjusting your Color Decoder controls (Color/Tint) to best match your native primaries makes sense - and this is what Tom H. is describing in his article. If your primaries are way off from the standard and if making the Color Decoding neutral with respect to the primaries leaves flesh tones for example looking "weird" then you may have to tweak Tint to try and compensate.

cheers,


--tom

I did some more experimenting on my Samsung LCD, much more radical this time, and it appears one can force the color Y values to be any value irrespective of the white Y. So the choice I described above, "Yw uncoerced", better describes this set. And that's fine, that gives the optimizer a little more independence to find the best Y value for each primary, as well as each secondary, to make it appear as close as possible to the Rec709 color.

Thanks, Tom. You're correct (at least I think you are), my projector is a Sanyo PLV-Z60. It has sort of a CMS, the manual says it does, but it's one of those where you put up a color then adjust it. It's like the Sony CMS, I think??? I never really got very far in trying to adjust it. My thread is here: http://www.avsforum.com/avs-vb/showt...4#post15556544 I didn't get much feedback about it, so, I kind of gave up. Any other feedback would be appreciated.

So, for me, I just plugged my gamut into the color calculator and targeted the new Y values. For me it gave me better targets for the secondaries.

I think this is more in line with Greg's recommendation on the same subject. I asked him about the "sum of Ys = white Y (1)" concept. Here's the link.

http://www.avsforum.com/avs-vb/showt...5#post16404305

Look at posts 130 and 131. I hope I didn't misrepresent or misunderstand your spreadsheet in formulating the question.