Basic Guide to Color Calibration using a CMS (updated and enhanced) - Page 5

One quick question about remeasuring color decoding. I have been doing all of the adjustment using the 75% color windows, but when I am done adjusting, I measure using the 100% color windows and look at the CIE diagram. Is that right or should I be using the 75% windows for the final measures?

The reason I ask is because of this quote in the getgray calibration dvd documentation:

"Some devices color-decoders work well at 75% levels but are inaccurate at other levels."

In this faq written by Greg Rogers he has an optimization procedure for color accuracy when dealing with non-standard primaries which involves moving the white point:

"To optimize the color accuracy of a display that does not have the correct primaries you can use the following procedure. First draw a color gamut triangle using the measured primary color values as the vertices of the triangle. Then pick a proposed new color temperature near the D65 standard. Draw a line from each primary color through the proposed new color temperature and extend it until it intersects the primary triangle. The complementary colors should lie at the intersections with the color gamut triangle. Select the best new color temperature that simultaneously minimizes the color accuracy errors of the three complementary colors. Measure the complementary colors to verify the results."

I typically try to minimize the error in the secondaries with reference to the standards whereas this technique minimizes errors with reference to the gamut your primaries actually create. From a color science point of view, what is the advantage of this technique where all the saturation lines intersect at the (non-D65) white point but the deltaE with reference to the standard can be quite high?

OK, fair comment. I am just testing thoughts/ideas in relation to calibrating a source/display/screen in an industry that appears to be quite anarchic in its implementations. I have no idea of what primaries the pj uses with an RGB input. What should "normally" be expected?
I am using an upsampling SD dvd player (Bravo D1) that may or may not apply rec 709 when outputting YCbCr 720p and 1080i signals. I am even more confused as to the primaries it uses when it outputs an RGB 720p or 1080i signal through DVI. Then, I have an HD capable dlp pj (Optoma HD7100) that I assume applies rec 601 to SD signals and rec 709 to HD signals. Thanks God, the screen is passive although it has a gray tint to it and it is retroreflective (Optoma Graywolf II).
For calibration I have the Optix Monaco colorimeter, HCFR, S&V and THX calibrating dvds and HCFR software.
If you were called to professionally calibrate this chain when using a DVI 720p at 48Hz and "native" mode with no scaling done by the projector, what factors do you know apply to the chain, what would you do and what assumptions would you be making? Would you calibrate it to rec 601. 709, other? Would you try to bring RGBYCM to the CIE? Would you want to get RGBYCM values right as a percentage of the white Y value? What would you rather have, a flat gamma log line or maximize the dynamic range yet preserve as pure whites as possible?
When using DVI, the pj offers a CMS with RGBYCM lightness, saturation and tint, adjustable gamma values by 500k, adjustable colour temperature by 500K with x,y adjustment, RGBYCM gains and offsets, sharpness, contrast and brightness.
Kras, zoyd, Tom, others, please??

I read this as pick a non D65 white point that gets your secondaries closer to the REC709 target - since secondaries are just the primaries subtracted from white. This agrees with what I have said before - primaries you are often stuck with while you are white adaptable - but natural colors are mostly within the secondary gamut - so get it right first.

Look at the gamut backwards - start at the REC709 secondaries - draw a line over to measured primaries - where they nearly intersect is your compromise white point.

After rereading the procedure I agree, picking an optimum white point will move the secondaries closer to spec and reduce error.

Attached are my latest. I reduce color to get luminance (Y) to 21% for red and 9% for blue (Rec 601). Green stayed around 51%. When I adjust tint the yellow and cyan move towards green. I adjusted it as close to the targets for both. Is there any way to move magenta closer to the target? I am also curious if I should try to adjust my color temp based on Greg Rogers' method or the variations discussed here.

Any help would be appreciated.

Here is my grayscale.

That's a very interesting idea. But color pictures also have black and white and gray. As long as our gray scale is at least in the ballpark, that might be a very good compormise.

BTW, the Hitachi DIrector Series (and similar higher end models) plasma displays have full color management and be tuned up to be super performers.

Chuck

angryht

Indeed looks like making your white point a bit yellow (less blue or more red/green) might get the secondaries on track. You will find that only small error in white leads to a greater shift in secondaries perceptual error.

Thanks, Kras. Any thoughts as to what color temp would get a better result. Here's what I am thinking: if I project a line from my measured green to the target magenta it intersects the black body curve at about 6000K. So if I change the color temp to closer to 6000K, my magenta should get closer to the target, right? I guess I am a little uneasy about changing from the 6500K. Is this a common thing to do with DLP projectors, inherently undersaturated green values?

Don't think about this in terms of changing color temperature from 6500K to 6000K, and forget about the blackbody curve. There are lots of CIE x,y values that would produce a correlated color temperature of 6000K. You want to find the best x,y point to minimize the complementary color errors and use that x,y point. Draw lines (3) from the primaries through a potential x,y point and pick the x,y point that minimizes the errors at the complementary colors (where the lines intersect the primary color triangle. If you don't move too far from D65 you probably won't notice the difference in grayscale color temperature, compared to the improvement in the overall color accuracy. (Advanced tip: It is actually better to do this using the u',v' color space if you have a way of making those measurements.)

OK. Now I think I am getting it now. Here is the file in u'v' mode. Why is it better to do it in u'v' color space?

Doesn't this assume that the secondary is always aligned through the white point to the primary? If you look at angryht's CIE diagram the line connecting blue to yellow does not pass through his current white point.

But when I adjust the tint, cyan and yellow move. So I was going to start with magenta and then adjust tint. Does that make sense?

Make sure your tint is aligned first - greyscale is irrelevant for aligning tint when you use the blue filter method - because you have 'bluescale'. Since your color decoder appears to be hardset to 9300K white point - this means your magenta/cyan will be too blue - you then fix that by moving towards yellowish whites (color temp is irrelevant)

u'v' is better because it focuses more on magenta - the eye sees more of those hues than green. The u'v' chart shows a small error in white will lead to a larger error in magenta.

Is it possible the imperfection in secondary axis crossing is just sensor slightly off on primary readings?

Save a preset with off secondaries yet a perfect D65 result vs. perfect secondaries and whatever white that gets. Flip back and forth I think you will see the trick works quite well - in fact you will notice every time you switch that the white you switched to is wrong and the old one was right - because you adapted to it! No matter the direction of the switch! But the colors being off - you don't adapt.

So, should I use tint to adjust the cyan and yellow so it is lined up with the target values then adjust the gray scale so it lines the magenta up?

It seems that everytime I adjust the grayscale the magenta is always lined up with the colortemp of white. And since my green is shifted towards red so much, the line between green and magenta always puts magenta towards blue.

Is there a simple way (equation) to convert my xyY data to u'v'. I am using HCFR and there is no option for that.

I have a Directory Series RPLCD, and although it has hue and gain for primaries and secondaries, it doesn't seem to have anything for saturation. Is there something in the service menu?
Roy

angryht

See my spreadsheet at the end of the sticky charts thread...

NO - I am saying adjust the tint first they same way you would without the sensor - using AVIA and your blue filters. to get cyan/magenta balanced with blue optically filtered.

Then do the measurements and adjust greyscale for best secondaries.

which is why you need to make your white more red - it makes magenta more red. But you actually want to make white more yellow so that there is more green in cyan as well.

White point shifted blue-red axis shifts mostly Magenta - White point shifted blue-green axis shifts mostly Cyan - White point shifted red-green axis shifts mostly Yellow. They key is finding the interactive balance so Magenta, Cyan,Yellow hit their target.

Got it. Thanks.

Thanks you for the clarification. I will do that then adjust the grayscale based on the comments you provided above.

Once again thanks.

Roy, I was referring to Hitachi's Director Series plasma displays, which have saturation, hue, and intensity settings for primary and secondary colors, as well as color decoder alignments. By aligning the color decoder first (and, as I recall, it's very close already), then applying the full CMS, you can get an outstanding picture from these Hitachi plasmas.

Chuck

Actually, I need to amend my last statement.

I just looked over my notes to be sure, and it looks like I may be somewhat mistaken. Yeah, it seems to be an actual, true CMS, but no level controls for the colors. It allows gain and phase controls for primary and secondary colors, plus a color decoder alignment that allows levels for red and green only, so that can possibly get you in the ballpark. I worked on this about three months ago, so I didn't have the benefit of Tom Huffman's color decoding spreadsheet. But it really looked wonderful, once it was tweaked, and had a more than acceptable black level.

Chuck

Picking up on the SMPTE-C vs 709 primaries discussion a little ways back, I took some measurements of the HDnet test patterns here. It appears that the color bars are mastered for SMPTE-C primaries and encoded in such a way that they will be reproduced correctly using 709 primaries at the display end.

Well, I think I've got it much closer now. Attached are my before and after CIE plots. I adjusted the white point toward yellow by increasing the red and green gains. That got magenta right in line with the target. Then I adjusted tint slightly and the yellow and cyan fell right into place. Then I verified that the grayscale tracked consistently throughout the percent stimulus. Looks pretty good.

So here is what I did:
1. Adjusted my 'user menu' color control until I was as close as I could get to the the 75% white as described in the first post of this thread. Actually, the best I could do was to get red and blue close to the target 21% and 9% (SMPTE-C). My green is only about 50%, which I would guess is fairly common for DLP projectors that are trying to be bright more than they are trying to project acurate colors.
2. Set my 'user menu' tint per getgray and the blue filter.
3. Adjust white point slightly towards red and yellow to get magenta towards the target. I used the constant measures in HCFR.
4. Double check that the grayscale tracks consistently throughout the percent stimulus.
5. Check contrast and brightness one more time then call it a day.

I must say that skin tones look much more natural.