CAMERAS: Thoughts on the Canon C300's Color Science

There's pretty color and there's accurate color, and they are not the same thing. Read why...
Art Adams
By Art Adams 10.26.12

In theory I like the Canon C300. It's small, it's well designed, and it can make pretty pictures. The Canon "look," though, is very different from what I'm used to, and it's taken me a while to figure out how to bend the camera's look into something that I really like. After spending an hour tweaking profiles the other day I think I found something I like. I don't have pictures to show you (I did my tweaking on set for a project that is a little bit secret, at least until next week when it'll be released) but I can describe what I saw reasonably well and you can reproduce my results yourself on any C300.

In the past I've leaned toward using Cinema1 or Cinema2 gammas and one of the Norm color matrices, usually Norm1. "Gamma" is all about how the camera maps brightness values and is completely separate from color matrices, even though they share the same names--for example, there's a Norm1 color matrix and a Norm1 gamma, but they don't need to be used together. The color matrix controls, in a very fundamental way, how the raw colors captured by the sensor are combined into a "look," and this look is not only unique to each matrix but, to some extent, unique to each camera.

I won't go into my gamma choices in detail as that could be another article on its own (and probably will be at some point). Instead I'm going to focus on the C300's color matrices. I will say this about the C300's gamma presets: I like the Cinema1 and Cinema2 as they hold the most highlight detail possible without using Canon Log, and they don't look flat on a Rec 709 monitor the way Canon Log does.

Canon Log is designed to be flat as log encoding is only meant to store data as efficiently as possible for color correction, and is not meant to be viewed at all. Why Canon didn't incorporate a LUT on the monitor output to make Canon Log look "normal" while shooting is a mystery--if Arri and Sony can do it, surely Canon can figure this out as well. This is the one reason I don't shoot Canon Log: if the budget is low enough that we need to shoot with a C300, but there is some money for color correction, there's usually no money for an on-set LUT box to make Canon Log look normal so the clients don't panic. Everyone wants to see what they're getting these days, even if they are going to grade it and change the look later.

This seems to be a significant misstep on Canon's part. Their camera costs about as much as a RED ONE, and even RED does this.

On Tuesday I found myself shooting an 8' iPhone replica in front of the Golden Gate Bridge, and while the phone was being set up and the action blocked I went through the C300 menus trying to build a look that I genuinely liked. The Norm matrices are okay but not great, and the Cinema1 and Cinema2 matrices are nicely desaturated, but all these matrices have unique effects on colors, particularly red and green.

I had a chance to play with a C300 during NAB, when I gave waveform/vectorscope demos at the DSC Labs booth, so I had a lot of time to play with the camera in conjunction with a Chroma Du Monde chart and a vectorscope. The DSC Labs Chroma Du Monde displays the Rec 709 primary and secondary colors at 50% saturation, and with a vectorscope set to x2 gain I should be able to put those colors into their target boxes. I can be sure that any skewed colors I see are a result of the camera processing. What I saw was really interesting:

In most C300 matrices reds tend to skew towards green, so that most reds end up being a "fire engine" or orange red.

Green tends to be undersaturated. Also, most matrices skew green towards blue, so greens tend to be more blueish. (Adding blue to any bright color desaturates it, so this has the effect of muting bright greens.)

Blue tends to be accurate but undersaturated.

The lack of saturation in green isn't a surprise as nearly every camera made has this issue. I'm told that this related to signal noise, and that makes sense in a technical sense: in order to make green more saturated the matrix has to subtract the red and blue signals from the green signal, and one of those tends to be a bit more noisy depending on the lighting:

Blue is more noisy under tungsten light as the camera must add gain to the blue channel to compensate for the fact that tungsten light has very little blue in it.

The same happens to red in daylight) subtracting the noisy channel from the clean green channel results in noise.

At first this makes no sense: why would subtracting a noisy signal from a clean signal result in noise? As best I can tell it's because when you subtract a noisy signal from a clean signal you introduce noise into the clean signal, because you're basically leaving holes in the clean signal where the noise was in the noisy signal.

That's a little bit of a mind blower, and it's really cool if you're a geek like I am.

There's only one color you can be certain of in every color matrix: flesh tone. Canon takes no chances with flesh tone. My guess is that the reason most of their matrices skew red toward green is to make sure flesh tones-which contain both red and green-always look good.

Also, some of the prettiest blues have some green in them. A bright blue sky often looks prettier if it's a little cyan instead.

This is the difference between accurate color and pretty color: all of these matrices can look very pretty in certain circumstances, but not all are color accurate.

If flesh tone is the wrong hue we will always, always, always notice because we have a built-in instinctual memory for what flesh tone should look like. Grass and sky, however, can change in color and we won't really notice without comparing those things to their images. Those colors aren't imprinted on our brains the way flesh tone is, and Canon knows this.

Canon seems to be doing a lot of secondary color correction in their matrices. They aren't just placing red, blue and green in different positions on the vectorscope and letting the other colors fall where they may; they are tweaking both primary and secondary colors to put them in very specific places. If you want to see an example of this look at a Chroma du Monde chart on a C300 using the Canon Log matrix: the pattern is unlike anything I've seen. It looks as if chunks have been bitten out of the traditional six-sided Rec 709 vectorscope pattern. It's a very unique shape.

There's clearly a lot of code behind their color science.

What does this mean to you? No idea. I can tell you what it meant to me on my latest shoot, though. Turn the page...

Prev Next »