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...

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…


The giant iPhone sat in a field of bright green grass. The background consisted of the Golden Gate bridge, clouds and patches of blue sky. While the iPhone was being set up I played with the matrices to see if I liked any one over the others. I discovered that the Cinema1 and Cinema2 matrices were nicely desaturated, but they made the green grass a dull blue green and turned the already orange-red Golden Gate Bridge a yellow-orange. The blue sky wasn’t very saturated. The image looked okay, but I wasn’t particularly pleased.

During my tests at NAB I’d discovered that the EOS Standard matrix was the only one that put the colors on a Chroma du Monde chart either into their boxes or onto the right vectors. (Getting the colors into their boxes on a vectorscope is only part of the battle; what’s more important is to get them on the vectors, or lines that run from the center of the vectorscope through the boxes. Distance from the center of the vectorscope tells you how saturated the colors are, but the colors are only accurate in hue if they fall on their proper vectors.)

Normally I’d never use this matrix just based on the name. Do I really want the “EOS Standard” look? This implies that I’m using a still camera color matrix on a video project, and do I really want that? What is the Canon still look? The name put me off at first, but from a practical perspective this matrix seems to be the only one that offers real world color accuracy in the C300.

The problem is that that EOS Standard matrix is way too saturated for my taste. People with unsophisticated artistic vision tend to like bright, “poppy” colors, as do children. As we grow older and more sophisticated we tend to drift away from enjoying pure red and blue and favor more subtle, sophisticated colors that are less saturated. (The psychology of this is fascinating, and if you can find a copy of “The Wagner Color Response Report” by Carlton Wagner you can learn a lot about how environmental color affects different classes of people. It was written in the 1980s and is now out of print, so it may be hard to find.) As a cinematographer who’s been around a while I don’t like bright zingy colors: I prefer less saturated and pure colors that are a bit more complex.

Brightly saturated colors can also be a nuisance as they exaggerate color errors. If your white balance isn’t perfect you’re more likely to see obnoxious color shifts if your color saturation is too high; likewise you may see brightly saturated highlights skew in odd directions as they become overexposed. Undersaturated color is not only more tasteful but it solves a lot of color problems.

Canon gives us a “gain” control next to the matrix setting, and I can only assume that this is similar to the matrix “level” control in the Sony F3, EX1, EX3 and FS700. Negative gain numbers decrease saturation and positive numbers increase saturation. As best I can tell that control is increasing the effect of the matrix on the sensor’s raw color signals. If you set gain to its lowest negative number the image will be completely desaturated because you’re seeing the luminance values created by each color signal but you’re not dividing those values into distinct colors.

Color signals overlap each other, and unless they are subtracted from from each other the camera can’t make them pure enough to be distinct. At a fundamental level that’s what matrices do: subtract color signals from each other to make them into pure, separate colors. How that math is performed determines how those colors look: is green truly Rec 709 green or blueish green? is red truly Rec 709 red or orange red? Rec 709 red, blue and green are very specific colors with very specific values, but there’s nothing that says that camera manufacturers have to reproduce those colors accurately. In the case of many of Canon’s matrices they’ve decided that their camera looks better if they skew the colors a bit. That’s what gives Canon cameras their “look.” All manufacturers do this to some degree.

I didn’t like any of those other matrices for this project. EOS Standard gave me green grass, an orange red Golden Gate Bridge and a blue sky, which is exactly what I saw by eye and exactly what I wanted. It was interesting to toggle back and forth between the Cinema matrices and EOS Standard because the EOS Standard green grass was such a saturated green that it was a bit distracting, whereas the Cinema1 and Cinema2 grass was a desaturated blueish green that didn’t distract from the iPhone at all. As I liked the hue of the grass in EOS Standard but not its saturation, and as I felt the saturation of the EOS Standard matrix was too much overall, I decided to take the matrix gain down a bit and see if I could find a saturation level I liked.

I started out at -10, and the less saturated grass was less distracting but reds still popped like crazy. A person wearing a red hat in the distant background stood out like a neon light. With a little more C300 experience I could have selectively reduced red saturation in the user matrix (the one with settings like R-G, G-B, etc.) but I really need to do that looking at a color chart as tweaking one color can skew all the others. Instead I committed myself to overall desaturation, and took the matrix down to -20.

Reds were still too saturated, so I took the gain further to -30. At this point the grass color, bridge color and sky color were still accurate but were pleasingly desaturated, and flesh tones looked normally saturated. One of our actors had whiter skin than the rest and he looked a little pale on the monitor, but that’s how he looked in real life so that wasn’t a big concern to me. Everyone else’s skin tone looked quite pleasing, and a red and blue shirt worn by one of our actors no longer screamed “I’ve got red in me! Look! Ahhhhhhhhhhhhhhhhh RED!”.

I took the matrix gain down to -40 and the image looked unpleasantly desaturated, so I backed off to numbers like -35 and -32. I discovered that, for this particularly setup, anything below -30 was way too much. The lower the matrix gain number the faster the image lost saturation, so a one point change from -10 to -11 is no big deal but -30 to -31 is a tremendous change.

In the end I used Cinema2 gamma, because it did the nicest job of holding highlight detail in the clouds while adding a pleasing contrast to the rest of the image, and the EOS Standard color matrix with matrix gain set to -30, which gave me accurate but understated color that looked like what I saw by eye. I white balanced the camera in sunlight and came up at 6100K, which looked great. The project is currently in post and should be released next week, at which point I’ll write an article about it. Meanwhile, take a look through the matrix settings in the C300 and see what you like, and-most importantly-compare what you see by eye against what you see on the monitor. EOS Standard will probably give you the best color match, but that may not be appropriate for what you’re shooting. Pretty color and accurate color are not the same thing, and both have their uses. Audiences tend to like abstraction, which is why the iPhone app Hipstamatic and Lomography film stocks and cameras are so popular: people often like to see interpretations of reality, but not reality itself.

Alfred Hitchcock said “Drama is life with the dull bits cut out.” As cinematographers we strive to capture events in a way that interprets reality, but often isn’t reality, because reality isn’t always that interesting when viewed in a rectangle on a TV set. It’s reality from the perspective of an artist.

Cinematography is also storytelling, which is both an art and craft: it’s art because the process of relating a story is indelibly affected by the perspective and personality of the person telling it, and it’s a craft because that person must ensure that the story communicates the emotion and meaning that they want to impart.

Colors and contrast are like words: choose the right ones for the story you want to tell. There are no rules. If it works, it works.

Disclosure: I have worked as a paid consultant to DSC Labs. I have no connection to Hipstamatic and Lomography.

Art Adams | Director of Photography | 10/25/2012 | www.artadamsdp.com


Share:
Art Adams

Art Adams - Director of Photography

Cinematographer Art Adams shoots spots, visual effects, web/interactive/mobile and high-end marketing projects. His website is at http://www.artadamsdp.com.

Art has been published in HD Video Pro, American Cinematographer, Australian Cinematographer and Camera Operator Magazine He is a current member of the International Cinematographers Guild, a past member of the Society of Camera Operators (SOC), and an industry consultant and educator. He lives in the San Francisco Bay Area.