CAMERAS: Rough Guide to Color Grading with the new DSC Labs OneShot
The best way to reproduce color accurately is to have a color reference in the shot with known colors on it. This chart does just that.
By Art Adams | September 18, 2012Now that we've made the neutral tones fairly neutral it's time to look at our primary and secondary colors. The vectorscope shows us where they should fall, so if we put their respective dots in the boxes we should see accurate color reproduction.
There are two things to note here:
Some cameras have a hard time saturating colors enough to go into their boxes. That's not always a bad thing as that kind of detail gives cameras their "look." It's more important to ensure the colors fall on the correct "vectors", which are lines drawn from the center of the vectorscope through each color box. If the colors fall on the proper vectors they are at least accurate, even if they aren't fully saturated.
For example, if the green chip falls on the green vector then the color is accurate, even if it isn't saturated enough to reach its box.
Cyan and green always seem to be the hardest colors to put in the boxes. I don't know why this is, although I have a couple of theories that I won't bother you with right now. What I will show you in a bit is why it may not be a good idea to fully saturate those colors to the Rec 709 spec.
For this part of the grade I'm going to go into the secondaries room and activate two secondary corrections. In the old days of film grading the primary grade ("primaries") controlled the coarse adjustments of the telecine: highlights, shadows, mid tones and their hues. Primary controls are the broad strokes. "Secondaries" are the fine strokes that take place not in the telecine but in the color grading software itself: this is where we add vignettes, selectively desaturate colors, etc. on top of the primary grade.
Primary controls manipulate the image overall, while secondaries manipulate part of the image.
We're going to start out looking only at the hue adjustment, as this is where we can put the colors on their vectors. Selecting a color on this line and pushing it up or down rotates that color around the center of the vectorscope without affecting the others overly much. If a color is landing off to one side of its vector this adjustment will move it around until it lands in the right place.
This shows the corrections I had to make to put each color onto its vector:
As you can see I really had to tweak every color. Green and red seem closest to their vectors. Blue and cyan needed some help, as did flesh tone. Now all the colors are technically accurate: the colors on the chart are being accurately reproduced in the video signal.
This is what the vectorscope looks like after the hue curve adjustment. (I had to zoom out from 2x for this view as several of the colors were way too saturated and were shooting off the scope!) All the colors are now on their lines, or vectors, that run from the center of the scope through their color target box. The flesh tones are falling along the I-line of the scope, between the red and yellow boxes, which is exactly where they should be. (The darkest flesh tone should fall directly on the I-line, while the others can fall a little to the side.)
Next I use an additional secondary control to adjust saturation. By saturating or desaturating a color I can move it toward the center of the vectorscope (less saturated) or away from the center (more saturated). In this case I want to put each of the dots into their boxes by putting a point on the adjustment line for each color that is improperly saturated and push it up or down until it hits its box. All of these adjustments interact so I may have to do this a couple of times per point.
Here's what the saturation curve looks like:
As you can see I had to reduce saturation overall except for a slight adjustment upward in magenta and a massive adjustment upward to cyan. (Remember I mentioned that cyan was always strangely undersaturated? Now you can see what I mean.)
Here's what the vectorscope looks like, now back to 2x magnification:
Every color is near perfect. Yellow is a little short, but I'm going to say that's close enough for this demo.
There's one very important thing to note here: see how green, cyan and magenta are elongated? That's due to noise from pushing the color matrix too far. In this article I talk about how matrices affect the image, and talk a bit about how to use the user matrix to manipulate colors and create "looks." The user matrix is a table that usually looks like this:
Each of those pairs affects the image in a fundamental way. Without getting into the math (because I don't understand matrices mathematically anyway):
When affecting a color pair, like R-G, red will change in saturation and green will change in saturation AND hue. That means green will move not just toward or away from the center of the vectorscope, but toward or away from red--making it a touch yellow or cyan, depending on which way your adjustments push it.
The problem is that excessive matrix manipulation can create noise, and color noise looks like a smear or a streak. The hue and saturation controls in Color are matrix adjustments, and I've pushed cyan so hard to make it "properly" saturated that we're adding a lot of noise to the green channel. The streaks fall roughly in the direction that each color will travel if I affect them in the matrix: green smears in the direction that it would move if I adjusted R-G, as does cyan because it is composed of equal parts green and blue. I don't know why magenta is smeared, but it's interesting to notice that its tail extends opposite that of green. (I played around with the saturation controls and couldn't find a setting where magenta didn't smear!)
If I back off on the cyan saturation I end up with this:
The green channel is a lot cleaner now. If you tried to manipulate this in camera you probably wouldn't have access to a matrix function that affected just cyan (some cameras do offer this control, but most don't) so you'd probably have to back off on the green saturation (G-R and G-B) to make green less noisy. This is normal, and it's how most cameras function out of the box. If you put green into the box using the in-camera matrix you'll probably end up with a smeared dot on the vectorscope and a lot of extra noise.
It is possible to get perfect Rec 709 colors out of a camera, but there may be a price. That's why I said earlier that getting the colors on their vectors--making them accurate--is often more important than making them perfectly saturated, because perfect saturation is sometimes too much for a camera to reproduce and can result in a lot of noise.
Just for fun I'm going to remove the correction I just made to cyan and put all the colors back in their boxes. Here's what the image looks like:
Last, but not least, I need to set contrast by making white and black look right to my eye. As you can see in the waveform image above, black is fairly black but white is a little dim at 70 IRE and I want a little more contrast than that.
I'll make white brighter in the "Primary Out" room, because I can control the gain and the blacks without changing their color.
If I change brightness in Color's "Primary In" room, where I've added all my overall color corrections, increasing brightness will also add warmth because I've added red to the highlight control to eliminate the blue cast. For example, if I use the highlight slider to make the highlights brighter I also add more red, because I've added a lot of red to the highlights control and that is amplified. Color can't separate brightness changes from color changes within the same room.
Changes made in the "Primary Out" room sit on top of everything I've done so far, as it's the last step in the grading chain, so boosting brightness here doesn't add any color to the highlights.
By raising the highlights slider a little in the "Primary Out" room I increased the white chip to 85 IRE as that's where it looks right to my eye.
This kind of luma correction is easy to do in Blackmagic Resolve simply by increasing the brightness of the highlights, as Resolve uses a YRGB model that separates luma (brightness, or "Y") from the red, green and blue channels. Changing Y [luma] in the highlights doesn't affect the color at all, in any mode.
Here's what we started with:
Here's the final grade:
Some of what I've done might horrify a real colorist. I'll admit that I'm an amateur at this stuff and I only do it so that I can learn what colorists can and can't do. It also helps me communicate my wishes to a colorist using terminology they understand. Amateur or not, though, understanding how all of this works helps me troubleshoot camera setups and know when a camera error can be repaired using a post fix or if a reshoot is required.
What I've shown is that if I put a chart in the shot with known values on it, and then grade for those values, I can easily neutralize an image that contains a very strong color cast. If this was a log or raw image I could very quickly create a color accurate and neutral rendition of what the camera saw on the set. I can then use this reference as the basis for further color corrections to create a specific look, or leave it as is so that the DP's intended look comes through exactly as she/he shot it.
This chart really isn't a guide for final color correction as that's more of an artistic process than a technical one. This chart is about dailies: getting the image right, or right enough, that all concerned can get a sense of what looks a DP is creating on set. This not only helps the DP maintain their chosen look throughout the post process but also reassures the director and producers that they are getting what they want and expect.
The FrontBox OneShot(tm) is available now from DSC Labs for US$287. Quantity discounts are available. Order by phone at (866) DSC LABS or +1 905 673 0929, or by email.
There is a smaller "Pocket" version that is available only through SMPTE for US$94.50. Please contact them here.
Note: I have worked as a paid consultant for DSC Labs, and I designed this chart.
Art Adams | Director of Photography | 09/18/2012 | www.artadamsdp.com
Get articles like this in your inbox: Sign Up