CurtPalme.com Home Theater Forum

[perisoft]

So, I've been using a weird method to color calibrate. I can't afford/justify a colorimeter, but I do have a Nikon D80 DSLR my job paid for. I've taken to using the DSLR to set whitepoint - I throw up a full white screen, set the D80 to 6700k whitepoint (it only has 6300 and 6700 - grr) and to show me the R/G/B curves, and take a picture. If the R G and B histograms peak at different points, I adjust the PJ gains until they match. Voila! Basically calibrated whitepoint!

I can set the black level the obvious way, and then in my HTPC adjust the gamma for each channel. I'd been doing it by setting the gamma so 4 or 5% is barely visible with each color, and that worked reasonably well - a HELL of a lot better than no gamma. But not perfect.

Now, a guy out there made a very detailed gamma adjustment program (I can't remember the name of it now... videoequalizer I think - it's linked here in a couple of places, but I forget where). It suddenly occurred to me that by putting up a gray ramp, and taking photos zoomed in of just the areas I want to calibrate, I can alter the gamma for each bit separately and build a curve that's correct all the way through.

So, I:

-Set whitepoint 'normally'
-Put up a gray ramp
-Color selected green
-Set a normal gamma curve for all three channels as a baseline, until 4% was just visible (with the green only)
-Re-set so I could see the full gray ramp

Now I've got a horrible green mess.

From there, I went iteratively, zooming in on 50%, then 30%, then 10%, then 20%, etc, and fighting tooth and nail with the frequent crashing and buggy user interface of this wonderful gamma program. The result? Happy PeriSoft!

My previous method got me this (which I posted in my triumphant black level thread - note that I overexposed it a bit to show the lower IREs...):



You can see problems down under 20IRE, and bluishness where there shouldn't be (of course, this is assuming your monitor is right. You can double-check in an image editor just by looking at the R/G/B values - they go all over).

And here's what I have now, after an hour and a half of work:



Not too shabby... If you stick that in photoshop and look at the low IREs from 3, 4, 5, 6, 7, 8, 9, 10% - the R, G, and B levels are within 2% of each other. And that's with PERFECT blacks!

(By the way, I never turned off manual focus and didn't bother to focus well, so all of these shots are blurry.. )

And the result with some screenshots. Pointless if your display isn't calibrated anyway, but what the hey... The cool part is the wall-e shots: One where wall-e is waking up, and it's dark in his truck, and the next as he's outside. I had to really lengthen the exposure on the inside shot vs. the outside shot, which you can see - the 'pause' in the upper left is a lot brighter in the first shot, but is the same in real life! But there's still a ton of detail in the low IREs of the inside shot, and it's not skewed. For the others I just tried to pick some color challenging scenes of people.

Anyway.. just thought I'd share another little triumph of projection...









[/img]

[Elaine Benes]

So is there some way someone who's not a photoshop/computer/camera expert can use this to achieve a similarly excellent gamma curve ?

Is it only possible from an HTPC ?

[perisoft]

[jask]

Great idea! I will try this out on the weekend.

[Zebu Fellenz]

I'll have to give this a try. Looks like a way I may be able to improve my greyscale on the cheap.

Erik

[Mark_A_W]

Can you post a picture of your VideoEqualizer curves?


I struggle with VideoEqualizer...the settings never stick..

[perisoft]

[perisoft]

Oh, an interesting comparison of a shot from before the new calibration and after... the before looked pretty good until I saw it next to the after. Again, this doesn't mean much without a good display to look at it on - but LCD and CRT monitors tend to have realtively good grayscale tracking (at least, good ones) so the comparison won't be too far off if the whitepoint is basically right.

Old first, then new.

[Mark_A_W]

Have you got Powerstrip active Perisoft?

I think Powerstrip may be overwriting VideoEqualizer - they fight to set the LUT.

I'll disable it and see.

[perisoft]

I am, but I specifically disabled powerstrip's gamma when I found it was screwing with the ATI control panel's all the time.

[antorsae]

Hey Perisoft,

I am working on my software to use a Canon DSLR to measure N points of gamma and later equalize 256 points using the (Nvidia) 10-bit gamma LUT.

I was working on this a few months ago but suffered a fatal HD crash - so I've had to write the software from scratch. I will record a video and post it so you can see it in action. I am playing around with this to see if I can get PERFECT color matching on my two marquees to do edge matching.

[perisoft]

[antorsae]

Perisoft,

The reason I am supporting Nvidia is because my primary goal is to match my two PJs color-wise across all 256 IRE values, and I am running horizontal span mode, so that windows sees only one monitor but there are really two. Nvidia offers an SDK so that I can equalize the gamma ramps for each PJ but it is still transparent to windows.

I can very easily convert the software so that if only one PJ is to be supported, I use graphic card-independent API, e.g. as described here http://developer.amd.com/gpu/radeon/RadeonSampleCode/GammaCorrection/Pages/default.aspx to upload the LUT to your card. This would work with either NVIDIA or ATI for single projector systems.

On a side note: I woke up early today and did a quick test and I have a first operational version of the software. Measuring 30 points takes about 2 hours of unattended operation (F/22 and Tv of 30 secs). What is great is that I can have my projectors completely off calibration wise, and have them calibrated AND matched running this software.

Not only the software maintains the correct ratio of R G B (i.e. color temp) across all IREs, but also it modulates the ramp to follow an arbitrary gamma value, e.g. 2.2, 2.5 - etc. My next improvement would be to add an offset at the low end and then "sink" the very very low end of the IRE scale (e.g. full black at 0,1,2 IRE and then climb to the dark -not pitch blak- offset following the regular exponential law thereafter).

Any suggestions in this regard (e.g. WHERE exactly in the IRE curve should full black start to climb and until what level) would be welcome.

Regarding the software - it's done in C++ with emphasis on object orientation. I did this on purpose precisely b/c I wanted to abstract the camera SDK from my software so it should be feasible to add more cameras. I am not familiar with the Nikon SDK but the minimal requirement is to be able to read RAW data without any processing on the camera. I am basically using the camera as a photon meter.

rgs - Andres

[antorsae]

OK, I think I found something. REC 709 does in fact include the TOE SLOPE which is what I intuitively referred to as "sink" and "climb". I will try this weekend.

See this: http://www.13thmonkey.org/~boris/gammacorrection/

[perisoft]

Interesting. If you're willing to share I'd love to take a look at what you have - I'm assuming it's a visual studio project?

One question - why does it take so long to do a run? And why F22? I was shooting defocused with the lens wide open, onto the patches in ntest. That's not optimal because the ambient and AC halo from the bright bits will lighten the dark bits, but the only other way was to custom make 10% windows for each IRE, which would have been a pain in the ass. Is that how you're doing it? Or with full screen color? Full screen would presumably be inaccurate since a full white field doesn't get displayed at the same brightness as a 10% white...

Which brings to mind a general question: What do you calibrate for? The same output will be different luminances depending on scene content - but what, exactly, is the effect? And is it the same for all projectors? EG, if you have half of the screen full white and a normal scene in the other half, presumably the whole other side will be somewhat darker than it would be if you had half of the screen black. Or do high IREs just clip off (crush), and anything greater than, say, 80, gets displayed as 80?

And presumably this will be different depending on whether you're talking about R, G or B? So wouldn't that alter your color balance dynamically depending on scene APL?

[antorsae]

Hi,

In the current version I manually set the lens parameters aperture and exposure. Last test I did was F/22 and Tv 30 secs. Right now (as we are speaking) I am having measurements taken at F/5.3 and Tv 8". The only important thing is to make sure dynamic range is maximized but that there is no overexposure nor an artifical limit of the camera's dynamic range (mine has a 12-bit ADC) so at a minimum I should set parameters so that IRE 100% is measured as high as possible but not over 65535 (RAW samples are 16-bit).

It takes a lot of time because it goes like this:

- First it shows checkerboard patterns and uses computer vision to detect them and locate the center of each (one for each projector)

- Then, for each IRE being measured, it shows a small circle for each color, for each projector. E.g. IRE 10% shows red circle, takes pic, green cirlce, takes pic, blue circle, takes pic, etc. and now for the other projector... now IRE 11%, etc... Obviously I've had to build a test pattern server (controlled via HTTP) so that the whole thing happens automatically. I run the pattern server on my HTPC and the controlling software on a laptop, but they can be run on the same machine if needed.

So if each picture takes 8 secs, then add 5 secs for transfer from camera to PC plus processing, that's 12 secs for measurement. If you want to measure 40 IRE steps, that's 40 x 12 x 3 x 2 = 2280 secs = 48 mins. Not bad (there are many optimizations one could do to measure multiple things in a single picture, but I prefer to stay safe. In addition, one could set different exposure times depending on where you are so in essence you would be measuring gamma in HDR, i.e. more than 12-bits; but so far I have not needed it).

Once I have the 6 transfer function curves I have a second command line program that asks for what the target gamma curve should be (I just added REC 709 gamma with toe slope, which is kind of what we call "gamma boost") and then creates a gamma-correction LUT for each VGA head compensating for everything that needs to be adjusted so that the end result is the target gamma curve we are after. Right now I have the option of either using the whitepoint of IRE 100% on either the left of right projector, or let the camera specify the whitepoint (note I still need to add ICC profile support to do this right).

[perisoft]

I see what you mean about dynamic range. Since my manual calibration was mainly a, "Hmm, I wonder what happens if I..." test, I just pointed the camera at a given patch of the right input, zoomed so it filled the frame, and set exposure to whatever the camera thought best. The main thing was to avoid any clipping at all, or else nothing would work...

Like I said, I'd love to test/work with you. One, it's fascinating stuff, and two, it's something that lets me do more than I ought to be able to with the equipment I have - which, as anyone who knows me from these forums will attest, is something I really get a charge out of.

[antorsae]

Here's the 6 uncalibrated transfer curves:



And here's the 6 guns again measured after equalization:



I still need to do some work...

[perisoft]

Not bad, though. Where do you think the error is coming from - do you think the camera's not seeing what's really there, or that the gamma adjustment isn't having quite the effect you're after?

[antorsae]

I think either the camera or the raw -> 16-bit linear is doing something funny as I think the camera is adding some compensation. The equalized curve looks almost perfect (let alone some rounding errors of the camera and LUT, which can be fixed) but what I want to find out is that leaving the manual settings untouched if the raw data is purely a linear measurement of the light in a single point.

Maybe a way to test it is to display two matched IRE windows separately and then overlap them and see if the raw data in the latter is just the sum of the two initial circles.

In other word: I want to verify that what I am reading is a linear factor of the light being present at a given pixel, with no further processing, averaging or metering.