Beyond Desaturation: Converting Color to Black and White

Color grading is very rarely solely about color. Recently, this was especially apparent. Jon Waltz's Bang directed by John Merizalde (produced by Chris Black) weaves a tale in black and white, but the footage was not always monochrome.

The video was always designed to be shown in black and white, but DP Jake Ures elected to shoot in color as the excellent monochrome camera options (by Red, Arri, and others) proved to be too expensive for the video budget. This meant post converted black and white was the only option. To properly expose tones, the crew monitored in monochrome on set (and recorded in monochrome to the R3D raw files though this was later changed to REDcolor3 within the raw accessibility options of DaVinci). The edit was performed live on the raw files within Premiere by Chad Sarahina. When settling on the final look, we discussed several stocks and examined a wide swath of reference images before entering the grade.

The completed video:

Understanding Black and White Conversions

To better understand black and white conversion, it pays to understand the difference between luminosity and color altering methods. Luminosity works by converting each luma value to a particular shade of grade (0 being black and 100 being pure white). This is actually very similar to how black and white film captures light and is essentially the same function as reducing the saturation of an image. Unfortunately, this can introduce some interesting problems.

Above are four very different and vibrant colors which are easily distinguished. Observe what happens when you desaturate this image (saturation from 100 to 0):

Our diverse shades are now a single shade of the exact same grey! Seems too unreal to be true? Go ahead and download the previous image and try for yourself.

What's happening here is that all these colors have the same exact luma value (50 as expressed in their LAB color encoding) which means that as you desaturate, you throw out all the other information and base the shade solely off the color's luminosity. This is a serious problem for images that don't display much contrast or all shot under similar lighting (such as dim streetlights).

An alternative and superior form of post greyscale conversion is color altering. While more work, color altering gives you much more control and very closely emulates the effects of black and white filters (such as red or yellow).

Here is what that same image from above looks like if converted to black and white via a color altering method ("Black & White" conversion in Photoshop):

As you can see, the once identical tones are now easily distinguished adding an enormous amount of depth to the image.

Channel Separation in Practice

There are several types of color altering conversions, but for this project I elected to separate the image into three color channels (red, green, and blue or RGB), perform my initial corrections, and merge the images.

You can see a rough example of my node workflow below:

The "Levels" node sets basic levels for the shot before we get to the BW conversion. The meat of the color altering process occurs in the next few parallel nodes. I split the image into it's red (R), green (G), and blue (B) channels which gives me individual control over each component's luminosity (as well as saturation and hue). (Authors note: looking back, it's probably easier to simply use a splitter/combiner node alt+Y to separate your color channels. I did not do so here for reasons I've forgotten.) I merge these (into the inverted image in Parallel) and apply the proper desaturation to preview individual color channel adjustments. By manipulating each R, G, and B node, I can gain fine control over the black and white conversion occurring in node 6. Adjusting lift, gamma, and gain as well as some hue and saturation changes let me separate the colors and tones and produce a vibrant image with proper tone differentiation.

The "High" node applies gain and a small amount of gamma to the highlights via a luminosity qualifier limited to the highlights of the image (within the HSL tab).

The parallel "Tone Hi" and "Tone Lo" nodes apply a simple split toning effect to the image to return a little warmth to the highlights (the "Tone Hi" again achieved with an HSL qualifier for the luminosity) and some cool blue to the shadows ("Tone Lo" also with the HSL qualifier). Jake and I have found that this small amount of color helps to further separate the image without obviously adding color (when applied sparingly).

For the sake of simplicity, I've omitted various shot specific secondaries, power masks, etc.

This basic formula works extremely well in giving you control in separating tones when there is little information to begin with (as was the case in this fairly dim video). The same techniques apply equally well to brighter shots that use the full dynamic range of the camera:

The channel separation technique is simple (if a little time consuming), but gives you tremendous control over the tones of the final image.

A Brief Aside on Noise and Flicker

This piece was shot at tremendously high ISOs due to the small budget, high framerates, and various other limitations. This meant the footage, as it came to me, was extremely noisy. Additionally, the high framerates and unsynced practical street lighting meant flicker was an issue in many shots.

Noise was removed without too much trouble using Neat Video within After Effects. Note that I am denoising outside of DaVinci and after the color pass. I am using Resolve 10 which supports OFX and own the Neat Video OFX plugin. However, the OFX Neat Video for Resolve as it currently stands is extremely buggy (weird motion glitches and general crashes) and cleans noise significantly worse than the AE version. Typically, this is not enough of an issue to notice, but on a video of such extreme noise I had to turn to AE (which meant an additional render pass). Following the denoise, 35mm grain was added back to the image to smooth the digital look denoised footage inherits.

The battle against flicker was less successful. Some shots were deflickered using various tools (AE plugins as well as the furnace package within Nuke) and one or two shots made use of the offset frame technique to varying degrees of success. With that said, the attempts were largely made to lessen the effect and were largely unsuccessful in removing it anywhere close to completely. Deflicker tools have come a long way but are nowhere near ready to replace managing flicker on location (which is sometimes impossible to achieve).