or Connect
AVS › AVS Forum › Display Devices › Display Calibration › Colorimeter spreadsheet for SpyderTV
New Posts  All Forums:Forum Nav:

# Colorimeter spreadsheet for SpyderTV - Page 6

I'll answer my own question. I believe there's an error in the formulae calculating the gamma. I've corrected it in my version and the numbers now make sense.
in Cell P72, for example, make the formula look like:

=LOG(M56, 10)*LOG(B56, 10) instead of =LOG(M56, 10)*LOG((B56+0.099)/1.099, 10)

specifically, change (B56+0.099)/1.099 to B56, for cell P72. Change all formulas that look like this, in this way and you'll be good to go.

Best,
jeff

### AVS Top Picks

Quote:
Originally Posted by greeno

I'll answer my own question. I believe there's an error in the formulae calculating the gamma. I've corrected it in my version and the numbers now make sense.
in Cell P72, for example, make the formula look like:

=LOG(M56, 10)*LOG(B56, 10) instead of =LOG(M56, 10)*LOG((B56+0.099)/1.099, 10)

specifically, change (B56+0.099)/1.099 to B56, for cell P72. Change all formulas that look like this, in this way and you'll be good to go.

Best,
jeff

It looks better...BUT is the calculation still correct ?
Yes the calculation is correct. From the source used for the spreadsheet calculations, http://www.brucelindbloom.com/index.html?Equations.html, you can see that the calculations are correct.

The intent of those extra factors was to protect against taking the log of a small number, but the way they (creators of the spreadsheet) did it is incorrect. In fact, there is no need to protect against small numbers if you look at what the source values are. The proper way to implement protection for some occurance is to do it in such a way that for arguements that need no protection you get the same answer whether you protect or not. This is clearly not the case with what was in the spreadsheet. You can test that for yourself.

Best,
jeff
Also, I've emailed rader about the corrections and emailed him my corrected version. I'll leave it to him to post the updated version.

Best,
jeff
Quote:
Originally Posted by greeno

Also, I've emailed rader about the corrections and emailed him my corrected version. I'll leave it to him to post the updated version.

Best,
jeff

Would love to try it on the data I have got from previous measurements...PM?
When I plugged in my numbers my gamma went from 2.30 to 1.83 with the new sheet. That might explain why I thought my picture looked a little washed out at times.
I thought the latest Rader sheet was using the Rec709 gamma calc? Greeno - are you using a simple power function for calculation? I'm not very familiar with gamma but a little info might help me understand what the difference is (i.e what was the Rader sheet actually using?). Thanks

Gary
I always cal to 2.5 as well. Supposedly to compensate for "dark surround" conditions but I actually prefer it regardless. ( although this may be down to the simple fact I never look at any display in anything but a dark enviroment , work or home. Does wonders for the tan too).

I think he's saying the equation that maps the transfer function to the target gamma is incorrect on the spreadsheet ...if you modfiy it away from 2.2?

I gave up smoking equations years ago myself so its all greek to me.
Quote:
Originally Posted by Mr.D

I think he's saying the equation that maps the transfer function to the target gamma is incorrect on the spreadsheet ...if you modfiy it away from 2.2?

Ah, gotcha. I really should get my head around the calcs at some time though I think.

Gary
If you want a good (simple) explanation of display gamma, look at the link below.

Todd
Thanks for the link. Combined with Poyntons Gamma faq I might understand the calculations a bit better now

Gary
OK after reading that Sencore doc and the posts in this thread I am confused. Do I or do I not have to get a reading from 0 to 100 to accurately calculate gamma? After Reading that doc I would say not.
Yeah kind of , but as "gamma" itself is just a simplification of the "straight line curve" part of the transfer function its hardly accurate in its own terms. The transfer function itself is not definable as an easily expressed exponent so they grab the easiest bit of the curve to quantify and broadly define it as a gamma for neatness and ease of describing deviation away from the idealised transform.
I found a nice article here:
http://www.ebu.ch/en/technical/trev/...set-index.html and choose "257-Autumn 1993". they provide a differential way to calculate gamma that avoids most of the issues with constant offsets that are very difficult to determin. I 've used it on my data and it is pretty bullet-proof.

Best,
jeff
am i supposed to change just cell P72 or all cells P72-P81,R72-R81 & T72-T81?

Thanks!
Quote:
Originally Posted by greeno

I'll answer my own question. I believe there's an error in the formulae calculating the gamma. I've corrected it in my version and the numbers now make sense.
in Cell P72, for example, make the formula look like:

=LOG(M56, 10)*LOG(B56, 10) instead of =LOG(M56, 10)*LOG((B56+0.099)/1.099, 10)

specifically, change (B56+0.099)/1.099 to B56, for cell P72. Change all formulas that look like this, in this way and you'll be good to go.

Best,
jeff
Actually you'll notice that I deleted my posting of my corrected version of the spreadsheet. After more analysis, including using the differential method of calculating gamma I linked above, I am not happy with my corrections either. I think there's an issue not accounting for a finite Y at 0%. The differential way above is more robust and does not care about constant offsets, since it is based on derivatives.

so at rader's suggestion, I agree that the spreadsheet should stand as it originally was. It can easily be modified to perform off-line analysis (add it in yourself) of the data to estimate gamma. I'm now doing that using the differential method.

Sorry for the confusion,
jeff
I have a Spectrascan 650 spectroradiometer (I use it for my business) and plan to use it to try to calibrate my Ruby. Can I use the reflectance mode off of my Stewart Firehawk? I would like to do this to eliminate any screen influence but am concerned about any unintended influences (low light level readings, etc.).

Thanks.
Quote:
Originally Posted by greeno

Actually you'll notice that I deleted my posting of my corrected version of the spreadsheet. After more analysis, including using the differential method of calculating gamma I linked above, I am not happy with my corrections either. I think there's an issue not accounting for a finite Y at 0%. The differential way above is more robust and does not care about constant offsets, since it is based on derivatives.

so at rader's suggestion, I agree that the spreadsheet should stand as it originally was. It can easily be modified to perform off-line analysis (add it in yourself) of the data to estimate gamma. I'm now doing that using the differential method.

Sorry for the confusion,
jeff

Actually, I liked your corrections (going back to Lindbloom's equations), but did notice that the "best fit" curve fitting of gamma in the spreadsheet was influenced strongly by the numbers at low % stimulation (because the values of the logs are so large for small numbers). I ended up with a gamma curve that appeared to be right on except for small deviations at 10% & 20% stimulation, but the calculated gamma was 2.10 instead of 2.2. When I changed the spreadsheet formula to calculate gamma from 20% to 100%, the result was gamma = 2.17.

I'll have to look at the differential method to see what that gives.

Thanks for the interesting information -- made me stay up half the night last Friday looking at formulae and reference material!
Thanks.
to summarize the differential method:
1) assume Y = P^gamma
2) differentiate dY/dP = gamma P^(gamma-1)
3) take log, log(dY/dP) = log(gamma) + (gamma-1) log(P)
4) fit a line to this and the slope is gamma-1. ignore the log(gamma) constant as it's not really log(gamma), it absorbs another offset or two that we don't care about.

When I use this method with my data, I get a gamma that's right on target, and I'm using all the data. note though that when using forward difference for example, if you start with N points of data, you'll have N-1 points of derivative data.

Best,
jeff
"to summarize the differential method:
1) assume Y = P^gamma
2) differentiate dY/dP = gamma P^(gamma-1)
3) take log, log(dY/dP) = log(gamma) + (gamma-1) log(P)
4) fit a line to this and the slope is gamma-1. ignore the log(gamma) constant as it's not really log(gamma), it absorbs another offset or two that we don't care about."

can this be done in a spreadsheet? I've never tried to do differentials in Excel... ?????
there's an easy way. excel has a function called "linest(yvalues,xvalues)" that returns the slope of a set of y,x values, i.e. it assumes it's a line. differentiating data, use the forward difference formula: (y(i+1)-y(i))/(x(i+1)-x(i)), with x(i+1)-x(i) != 0

use x as %stim (a number between 0 and 1) and y as Y normalized by Y at 100%stim.

differentiate as above. take the log and feed the values into linest(,).
Gulp....

Thanks !!
Quote:
Originally Posted by greeno

there's an easy way. excel has a function .....

Ever since my first graduate p-chem class, I have always dreaded the phrase:

There is an easy way.....

This was usually followed by a weeks worth of frustration trying to complete the single assigned problem.

niesman
IMHO, there's a fundamental flaw in normalization of Y the way it's done in the spreadsheet. Try performing same normalization to a perfect 2.22 Gamma Y curve(assuming black level measurement is taken at 7.5IRE) and the resulting curve gamma will be much lower!

The correct way (IMHO again ) would be to find what is the Bias(Bi) and the Gain (Ga) of the measured Y curve are and normalize as follows:

Ynorm = (Y-Bi)/Ga.

The Y that we measure is (S=stimulus > 0.081):

Y = Ga*(S+0.099/1.099)^gamma + Bi

Black level S=0.075 (assuming it's calibrated correctly) and according to REC709 (linearity below 8.1%):

Yblk = Ga*(0.075/4.5) + Bi
Ypeak = Ga*1 + Bi

Solving this set of equations is easy. You can have a look in the attached spreadsheet - comments are welcome!

BTW, Can anyone tell me why do I get this greenish look to all the movies once I'm closing in on D65 (even with the original gamma calculation)?

Â

Ok, I tuned my setting using the SpyderTV walkthrough, then I did the gamma and color readings with DVE. My gamma appears to be way off, how do I correct this? The colors look ok, seems that I have a bit of green push.

Also, when attempting to calibrate grayscale, do I turn the color down to zero on my set?

Â

Quote:
Originally Posted by eugbuber

IMHO, there's a fundamental flaw in normalization of Y the way it's done in the spreadsheet. Try performing same normalization to a perfect 2.22 Gamma Y curve(assuming black level measurement is taken at 7.5IRE) and the resulting curve gamma will be much lower!

The correct way (IMHO again ) would be to find what is the Bias(Bi) and the Gain (Ga) of the measured Y curve are and normalize as follows:

Ynorm = (Y-Bi)/Ga.

The Y that we measure is (S=stimulus > 0.081):

Y = Ga*(S+0.099/1.099)^gamma + Bi

Black level S=0.075 (assuming it's calibrated correctly) and according to REC709 (linearity below 8.1%):

Yblk = Ga*(0.075/4.5) + Bi
Ypeak = Ga*1 + Bi

Solving this set of equations is easy. You can have a look in the attached spreadsheet - comments are welcome!

BTW, Can anyone tell me why do I get this greenish look to all the movies once I'm closing in on D65 (even with the original gamma calculation)?

This is one reason why the differential method I linked above is very useful. Normalization and zero values (well sort of) have no effect since you're fitting to a derivative.

The greenish tint issue could be a sensor calibration issue. I had a sensor whose calibration had drifted and it produced "calibrated" results that were green tinted. Gamma has nothing to do with D65, although it affects grayscale tracking across %stim.

Best,
jeff
Greeno, I've read the method - what I'm suggesting is just a simplification of the same approach. Did you manage to implement the whole algorithm they describe in the article in Excel?

As for the "green push" - how did you overcome the sensor calibration issue? (I know it's got nothing to do with gamma).

I'm going crazy here It's been seventh night in a row I'm trying different RGB cut/drive values, but always end up with a green tint

Quote:
Originally Posted by greeno

This is one reason why the differential method I linked above is very useful. Normalization and zero values (well sort of) have no effect since you're fitting to a derivative.

The greenish tint issue could be a sensor calibration issue. I had a sensor whose calibration had drifted and it produced "calibrated" results that were green tinted. Gamma has nothing to do with D65, although it affects grayscale tracking across %stim.

Best,
jeff
So in theory if I can adjust the gain and bias for each color to achieve the perfect value for red, blue, and green then my grayscale values should be 100% correct right?
My "calibrated" settings that produced a green tint in blond hair and some explosion firelballs came about when using an old dtp-92 - not the current dtp-94. The way I got rid of the green was to use a calibrated meter, the dtp-94.

There was a clear green tint when I measured grayscale using the dtp-94 AFTER calibrating the set using the dtp-92. Re-calibrating with the dtp-94 solved the problem.

I did implement the main part of the differential method. Who cares about the intercepts? What we're after is the (gamma-1). that is what I've implemented.

Best,
jeff
Once you have a SpyderTV sensor and this spreadsheet and you get the xyY values from SpyderTV and input them to compare to the spreadsheet, how do you know what settings to tweak to bring them back into focus?
New Posts  All Forums:Forum Nav:
Return Home
Back to Forum: Display Calibration