D800 vs 5DIII. This one will be tougher to guess.

Started Jun 12, 2012 | Discussions thread
Victor Engel
Forum ProPosts: 15,158Gear list
Re: Color space transformation
In reply to Press Correspondent, Jun 18, 2012

Press Correspondent wrote:

Victor Engel wrote:

Something is wrong with your logic. Let me describe a scenario that would fail.

As you know, humans perceive violet and purple the same way. There is a difference, though. Violet consists of light with wavelength shorter than blue. Purple consists of a combination of red and blue. Appropriately chosen shades of purple and violet are indistinguishable by human eyes.

But suppose camera 1, through the choice of color filters, is able to distinguish between those two colors. Perhaps the blue and red channels are rather apart in the spectrum. And suppose camera 2 sees both of those colors as the same. How do you propose to map the colors from camera 2 to camera 1’s colors? It's impossible, because one color would have to map to two colors.

I think in the general case you are correct. The recorded picture is a projection of the infinite-dimensional color space to a 3-plane of RGB.

This need not be visualized in terms of planes. Lines are sufficient and easier to visualize.

Naturally, if two different RGB 3-planes are at an angle,

Angle with respect to what?

the projections would be different leading to non unique colors in your example. However, there are two arguments against this. (1) Provided the original conditions,

what original conditions?

such cases of exact matches and non-uniqueness should be rare and represent a zero probability density.

They are not rare.

(2) Even more importantly, the orientation of the RGB 3-planes in the color space is defined by the white balance.

by the time you are in an RGB space, the damage has been done.

If the white balance is the same in both cases, the planes are parallel to each other.

I'm not sure what you are imagining as planes, but back in the infinite dimensional world, I imagine those planes are really non-flat surfaces, possibly even noncontinuous surfaces, due to the noncontinuous nature of the spectrum of the light source.

Therefore their componebtsa become linearly dependent. In other words, if one sensor (e.g. the human eye) does not distinguish two diffrent colors, then the other sensor, after a proper calibration, would not either.

I am not a mathematician, so the proper treatment of this matter would have taken me more time than I would want to spend on it. I will keep this in mind and let the marles roll - if any - to se if it becomes more clear over time. Meanwhile I would settle for these intuitive conclusions:

1. With proper care it should be possible to calibrate one camera to the colors of another camera with a great precision (for example, you probably could just calibrate your monitor to show a Nikon image with the Canon colors).

2. This calibration may not be 100% accurate, because the sensor response is not unique by color producing an uncertainty in some cases (your example). I would venture to say this effect should be negligible due to similar white balance and overall low probability of real life colors falling exactly into these values.

3. My initial hypothetical conditions of the contraction functions being non-trivial (linearly independent) and defined over the entire color space are not 100% realistic for real sensors. The imperfections of technology would probably have a much greater affect on the outcome of the calibration that any theoretical effects of non uniqueness of the color response.

If anyone's level of math out there is up to the task, it sure would be interesting to hear what specific conditions would lead to the best calibration scenario

Meanwhile, I will try following your and Mikael's guidance on camera calibration. Today I stopped by a local furniture store - where else! - and with no waiting lists bought a Nikon D800 they happened to have in stock, brand new, the USA version, made in Japan. Go figure who out there is an authorized Nikon dealer! I will also keep my 5D classic to see if I can reproduce its "magic" on the Sony sensor.

-- hide signature --

 Victor Engel's gear list:Victor Engel's gear list
Canon EOS 600D Canon EF 50mm f/1.8 II Canon EF 135mm f/2.0L USM Canon EF 300mm f/4.0L IS USM Canon EF 100-400mm f/4.5-5.6L IS USM +7 more
Reply   Reply with quote   Complain
Post (hide subjects)Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow