Color Matching a Canon C300 to an Arri Alexa
They said it couldn't be done. Maybe, but I got close. How close? You judge...
By Art Adams | January 26, 2013
According to my personal taste in color, the EOS Standard matrix has two fatal flaws. One is that it is way too saturated, but that's easy to adjust using the "gain" control under the matrix menu. (I used that control to desaturate the EOS Standard matrix for this spot.) The other is that red is much more saturated than any other color, such that any red object in the frame becomes a real distraction. Other than that, the EOS Standard matrix reproduces Chroma Du Monde colors fairly faithfully.
After placing the Alexa next to the C300 I white balanced both cameras on the chart and then matched their exposures by placing the chart's 18% gray background at 40 IRE on the waveform monitor. The Leader 5800 waveform/vectorscope has an overlay function, so I captured an image of the Alexa's vectorscope pattern and overlaid the C300's vectorscope pattern over it. I then tried to manipulate the C300's user matrix until its vectorscope pattern overlaid the Alexa's pattern. In theory this meant their colors matched.
This is really tricky because all the user matrix settings interact. Changing one matrix value will affect multiple colors. For example, manipulating the G-R value changes green's saturation (increasing or descreasing it) but also changes red's saturation and hue. If I change red's hue by a large amount then it will no longer appear red on a display, so then I have to use another control, maybe B-R, to pull it back--but that then changes blue's saturation.
It's quite a juggling act.
It took several tries and various approaches to get as close as I did. What worked in the end was to use the phase control, which simply spins all the colors in a circle around the vectorscope, to match the chart's flesh tone patches (between red and yellow). Then I pushed the other colors around to make them match as closely as possible. Here's what I ended up with:
The inner ring of colors is the C300; the outer ring is the Alexa.
Adam and I toggled back and forth between cameras while looking at real flesh tone, typically one of our hands placed in the center of the chart. The C300 really wanted to make flesh tones more red than the Alexa did, and portions of our hands that are normally redder than normal skin tone (like knuckles) blended beautifully into the surrounding skin on the Alexa but popped bright red on the C300. In the end I had to make two compromises to make flesh tone match:
- Red was so crazy saturated in the EOS Standard matrix that I couldn't desaturate it far enough on its own, using R-G and R-B, without seriously messing up green and blue, so I had to desaturate red by reducing saturation overall. Therefore this matrix desaturates every color except for red when compared to the Alexa.
- I couldn't get flesh tone on the C300 to line up with the Alexa's flesh tone and the chart's red target at the same time, so I compromised and made flesh tone a near perfect match while leaving the C300's red shifted slightly toward yellow, making it a little orange. This also happens in most of the C300's built-in matrices.
Here's the final "match." It's not perfect, but flesh tone is very close. I couldn't match the bright flesh tones and the darkest flesh tone patch at the same time as those are spread farther apart in saturation on the C300 than on the Alexa. Note: We matched cameras using the Rec 709 output on the Alexa but I recorded LogC, which I ran through Resolve's LogC-to-Rec 709 3D LUT to create the chart and vectorscope images in this article. As a result the two cameras matched better live, using Alexa's direct Rec 709 output, than do the colors translated through Resolve 9 Lite, which appear slightly different.
Alexa is inside, C300 is outside. Flesh tones match, but in order to do that I had to desaturate the C300's colors overall compared to the Alexa. Note the C300's red is a bit orange, which seems to be normal for this camera.
Because red isn't exactly lined up on the red vector magenta is pulled off its vector a bit, making it a little too red. As there's very little magenta in the world this doesn't bother me overly much.
Green matches in hue, as does yellow. Cyan matches in hue but as it's pulled in toward the center on the C300 it's a little more desaturated, which affects green and blue saturation to some extent. Cyan is a very difficult color for most cameras to saturate, and the Alexa does a vastly better job of it than most.
Flesh tones match, which is the important bit, and for the most part all the other tones match in hue other than red and magenta. The rest of the C300 colors are desaturated, which is what I had to do to lessen it's tendency to exaggerate reds in flesh tone. After matching the colors as close as we could on the vectorscope Adam and I switched between cameras while looking at flesh tone on a monitor and reduced matrix gain (overall color saturation) in the C300 until flesh tone closely matched the Alexa's.
Most cameras render colors slightly differently under tungsten light than under daylight, and these cameras were no exception. Creating a C300 daylight matrix was a bit more complicated. I didn't want to use actual daylight as that changes, and it was an overcast day anyway. I didn't want to use HMIs or daylight Kino Flo tubes because those are "good enough" for most uses, but their color spectrums are not perfect. I opted instead to use filters to convert tungsten light to daylight.
The normal film filter used for this conversion, the 80A, read as too heavy on both cameras: the C300 white balanced at 7000K and the Alexa saw it as 6300K, according to their own internal Kelvin measurements. Internal CCT numbers generated by cameras are rarely correct, but they do indicate what the camera thinks it is seeing. If the camera thinks it's seeing 7000K then there's a good chance that there'll be a difference between my settings created using that white balance and a normal daylight balance of 5500K.
We used ambient daylight filtering into the shop through a window to see how the cameras responded to actual 5500K light and both cameras white balanced at more reasonable CCT numbers, so I backed off to an 80B filter. (The 80A converts 5500K light to 3200K; the 80B converts 5500K light to 3400K.) The only two 80B filters available in the shop that day were made by different manufacturers, and while the Schneider 80B worked very well on the C300 the Formatt 80B resulted in a warmer white balance on the Alexa. I added a 1/8 Schneider CTB filter to the Formatt 80B on the Alexa and brought the two cameras within 400 degrees Kelvin of each other. Based on a camera comparison that we did after building the matrix that appears to have been close enough.
Alexa is inside, C300 is outside. Once again, most colors match reasonably well except in saturation, which had to be lowered on the C300. The C300's flesh tone and red hues match the Alexa reasonably well. Red is a lot less orange.
The color image matched better live than through Resolve's LUT. If we compare these two images alone, though, flesh tone is accurate in hue but not necessarily in saturation. This time I was able to match both red and flesh tone, but while yellow matches on the vectorscope it's off a bit in the picture. Generally the hues of all the colors match better under daylight but the slightly reduced saturation in the C300 results in a lot of differences in saturation. Still, saturation is easy to adjust to taste in the field by manipulating the matrix gain control. (There's lots of saturation left in the C300. Both tungsten and daylight matrices were matched at -29 or so in gain; normal gain is 0, and -34 looks severely desaturated. The gain control is non-linear, and adjustments beyond -32 result in dramatic decreases in saturation.)
What's interesting is to look at the shape of the chart colors on the vectorscope under different CCT's:
This is the Chroma du Monde under 3200K (really 3000K) tungsten light. Tungsten results in a thinner pattern than does daylight. This is probably because there's very little blue in tungsten light so it's harder to expand the pattern--and increase saturation--on the blue-yellow axis.
Tungsten light is very hard on cameras as the blue channel typically requires +6db of gain just to make its signal strong enough to balance with the other color channels. This is the equivalent of pushing the blue emulsion layer in a film stock by one stop, and is the reason the blue channel is always the noisiest channel under 3200K light.
This is the pattern under 5500K daylight. The blue-yellow axis is wider, meaning that both yellow and blue saturation has increased. This results in fuller color saturation across the reproducible spectrum. Silicon sensors are naturally "daylight balanced" because silicon is least sensitive to short wavelengths of light, like blue, so they perform best when the light striking them has a LOT of blue in it to compensate for this insensitivity.
It's important to know this because some cameras hide these different responses to color temperature better than others.
You want settings? I've got settings, on the next and final page...
Get articles like this in your inbox: Sign Up