Film-Tech Forum ARCHIVE: XYZ mastering

This is topic XYZ mastering in forum Digital Cinema Forum at Film-Tech Forum ARCHIVE.

To visit this topic, use this URL:
https://ft-forum.com/ft/cgi-bin/ubb/ultimatebb.cgi?ubb=get_topic;f=16;t=002774

Posted by Marco Giustini (Member # 4544) on 08-07-2016, 06:01 AM:

Hi all
I have a mastering question.
I would like to make some test charts with accurate colours. So far the RGB to XYZ conversion done by various software is never very accurate. My white is very close to the reference but cannot be used for a measurement. My green has a tad of red as well and so on.

I thought I could create my chart in XYZ directly but it has to be 12 bit per component and so far I haven't found a free/easy solution to do so.

I'd be happy to work in XYZ and type the colours manually on the dial (and colours will then look wrong on my RGB screen) but again the problem is the colour depth.

Can anybody help me?

Cheers

Posted by Carsten Kurz (Member # 5396) on 08-07-2016, 07:35 AM:

How would you practically use these testfiles? Single images uploaded to the projector, or to create J2K/DCP for ingest?

I don't know any 'native' XYZ image editors. I would assume imagemagick to be able to create files with precise color target, but I have no experience with it.

I guess any matrix type conversion from common creative software file formats and color spaces will never be accurate for the full input color space - you would need a custom LUT conversion to hit precise target color codes.

If only a few colors are needed, maybe a 'reverse' technique would be possible, creating a range of input values and then choosing those that create the exact target values.

- Carsten

Posted by Marco Giustini (Member # 4544) on 08-07-2016, 02:11 PM:

Hi Carsten

I'd like to make some framing chart CPL's.
Yes, I'll only need Red, Green, Blue, White and maybe a greyscale (10%).

What are you suggesting? I thought of opening a PNG/TIF file from the projector but those are already RGB. Maybe I could unwrap an existing RGB framing chart but that would contain jpeg2000 files.

Posted by Daniel Morez (Member # 5787) on 08-10-2016, 12:50 AM:

Hello Marco,

For making color charts, what has prevented you from creating and using less than 12bits per color channel images? With the exception of precision ramps and actual imagery gradients you don't really need 12bits per channel for a solid color. After all, 255cv will be the same as 4096cv in terms of how its scaled along the video graphics path (or projectors internal proc path if we are talking TI readable PNG files). For instance, in the case of DCI Ref White....

12bit 3794, 3960, 3890 = 8bit 237, 247, 243

It is certainly possible to convert and even manually adjust RGB levels in a PNG or any image file for that matter to meet DCI XYZ target equivalents. I think however with free or low budget tools I am aware of it is a more difficult approach. The above example of DCI Ref White is an easy once since the DCI spec gives us the explicit 12bit code values.

If you have not already done so I would suggest you try BM Resolve Lite (free). It has built in 3D transforms that if,(and I cant stress this enough) if its many project setup variables are set correctly, it will yield a proper RGB to X'Y'Z' converted output images/video.

Bravo to you for your efforts in perusing this.

-Daniel

Posted by Marco Giustini (Member # 4544) on 08-10-2016, 02:58 AM:

Hello Daniel,

Many thanks for your input, very appreciated.

I always enjoy making all sorts of test charts myself but I know that they cannot be used for colour calibration and/or shooting light level.
In the end I would only need Red, Green, blue and White to be correct and I would be happy to work in XYZ colour space (with my monitor displaying incorrect colours - knowing that they would be fine on a cinema screen) but I could not find a software that allows me to dial the 12bit colours manually.

I will definitely test BM resolve lite, thank you again!

Posted by Carsten Kurz (Member # 5396) on 08-10-2016, 04:39 AM:

The question is, how much precision do you actually need? If you know the exact conversion formula, you could of course calculate it backwards, no matter how inaccurate the conversion formula would be. For DCP-o-matic, the formula is documented. DCP-o-matic also has a basic XYZ vectorscope that can aid in monitoring the resulting X/Y/Z values. Although it would be nice if there was a cursor value display giving real numbers. Maybe Carl can add this.

I know that there is/was a J2K image analyzer tool in the OpenJPEG tools. Maybe it is possible to display color values from a J2K file directly with it. That would be after color conversion and gamma adjustment.

- Carsten

Posted by Marco Giustini (Member # 4544) on 08-10-2016, 01:53 PM:

DCP-o-Matic never ceases to amaze me! Thanks, I'll have a look.

I'd like to use that chart to shoot colours - so they must be perfectly accurate, no tolerance.

Posted by Carsten Kurz (Member # 5396) on 08-10-2016, 05:53 PM:

Well, if you need 12Bit precision, you will hardly hit that target when creating 8Bit source images, because you can not create the precise inbetweens when the 8Bit is interpolated to 12Bit. Which image editor actually does allow to dial in colors with 16Bit accuracy? At least MY photoshop doesn't, even while being able to work on 16Bit images.

- Carsten

Posted by Marco Giustini (Member # 4544) on 08-11-2016, 02:49 AM:

Having looked at the Black Magic software (Thanks Daniel, a nice free piece of software!) I feel that finding what RGB values I need to achieve some specific XYZ ones should be easier.

I researched some time ago about working on 16bit software and the only free software I could find was something that had been dropped/discontinued and for the life of me I could not install it (Linux).

Carsten, are you saying that I won't be able to achieve the perfect colours (I'm only looking for RGBW) if I use the conversion formula? In the end what I am looking for is a white that reads like the proper FS_White chart when using a colour meter.

Could you point me to the documentation that mentions the formula being used in DCP-o-Matic?

Thanks

Posted by Carsten Kurz (Member # 5396) on 08-11-2016, 12:08 PM:

Marco - If you want to achieve precise XYZ coordinates with 12 Bit precision, you can not rely on 8Bit source images, because the spreading of the values when resampling 8 bit to 16 and back to 12 Bit will leave 'holes' in the quantizing range. If your target value falls into one of those holes, you will not be able to dial it in with an 8Bit color picker. I guess for most people, 8Bit accuracy to shoot colors would be sufficient anyway, but I do understand you want to dial in the exact numbers that later are read from the J2C files.

I heard that more recent versions of Photoshop actually offer a 16 or 32Bit color picker when working with 16 or 32 bit.

The conversion matrizes can be found in every DCP-o-matic metafile, and the formulas are given on DCP-o-matic website. You could also ask Carl directly.

http://dcpomatic.com/manual/colour.pdf

Carl already 'approved' my suggestion for an XYZ color indicator in DOM. Shouldn't be complicated to do, as they are already computed for the vectorscope, so it's basically a display issue.

It was also me who suggested adding the vectorscope initially, because it's the only way to find out how the various libraries, subsystems and pipelines that DCP-o-matic uses (e.g. FFMPEG) deal with issues like limited RGB range, 10/12 Bit in input files, etc.

Another option, of course is, to create your own 12Bit TIF or J2C files and bypass the color conversion in DCP-o-matic completely.

- Carsten

Posted by Marco Giustini (Member # 4544) on 08-11-2016, 01:12 PM:

Carsten,

Thanks for your time on this - I really appreciate that!

I'll try the '12bit TIFF bypassing conversion' route first. If I could find something that allows me to dial the exact colours, that would be just great.

By the way, are the exact coordinates on the DCI specs?

Failing that, I'll try and go via the formula route - I understand what you are saying and I agree. I guess it all depends on how inaccurate I am going to be. To be honest my white on my (inaccurate) framing chart is 255-255-255, I had no idea that to have DCI white my RGB was supposed to be different (thanks for that)!

I'll keep the forum posted with any progresses.

Thanks

Posted by Marco Giustini (Member # 4544) on 08-15-2016, 08:52 AM:

Hi,

I've tested my white CPL and it's spot on. I've taken several 8-readings measurements and then averaged all together. the difference is 0.00003 on x, 0.0003 on y and 0.033 on brightness.

I had a look at the formula behind DCP-o-matic and maybe because of my lack of math skills or because the formula is converting RGB coordinates into XYZ coordinates (not RGB values?) I was unable to find a way to convert other known X'Y'Z' values.

I've found the DCI X'Y'Z' values, I now need to either

1. find a software that allows me to dial a 12bit value into a colour swatch (I am trialling Photoshop: the eyedropper gives me 15 bits resolution but the colour swatch is still 0-255)
2. find the correspondent 8bit RGB - which may be simply impossible as mentioned.

Carsten, I see why you are saying that a better vectorscope would be the best solution: a solid pattern would be a line and it would be easy to find the XYZ value for that line. Once the corresponding RGB values are known by using full screen solid colours, then I can use them where I need to.
Still, I reckon I'd need a software that allows me to dial a colour with 12bit or higher resolution.

Any ideas?

Thanks for your inputs on this!