August 24, 2015 at 8:15AM, Edited August 24, 8:19AM

0

HMI vs Tungsten? Solutions for Blue Channel

I was recently on a studio shoot with a red cyc, and lit in tungsten.
I've dealt with a less extreme version of the issue before but this time it was being shot on a 6k Red Epic Dragon and the deliverables needed to be in 5k for a large LED billboard.

I've had a lot of noise and was discussing with the dp and dit of how we can deal with this issue. Since we were on the clock, we gave it a little more power and opened up a half stop and rolled.Since then i've done some research on sensors, and electricity.I think this information can be useful for deciding for lighting and camera scenarios in the future.

>Camera Color Temperature<
---As Camera's have native ISO's, they also have Native WB.
Red is rated at 5600K. As any increase in ISO is just a metadata change the WB is also a metadata change. Shooting in Tungsten automatically increases noise slightly.
---The recommendation i got was to still shoot at 5600K with a lens filter to tungsten just like the film days.
going from 5600k to 3200k by filter you do lose 2/3 stop.

>HMI vs Tungsten<
This was a problem not only in Digital but in Film as well.
Blue (cold) wavelengths are shorter and faster than Red (warm) wavelengths.
The sensitivities of the sensors are the same but the sensor needs a lot more BLUEish tones in a scene to render a clean neutral Blue Channel.
In a LRB Beyer sensor (every camera)
There is no Green channel.
The green is artificially generated by the cameras CPU by scanning and guessing what is missing from the RED and BLUE. Since there is not much blue, the camera is not able to generate a clean Green Channel correctly either. This results in some weird tints in skintones.

---By using Tungsten lighting. You're providing even more information to the RED channel and it further increases Noise in the other two channels.
---By using HMI's you are feeding the blue channel and also naturally adding Cyan to the red channel.
Since there is more information in the Blue channel, the camera can now do a better job at generating the Green Channel. This will reduce overall Noise.

>SET DESIGN FIX<
This is not Ideal, especially for the art director, but with a software like livegrade or resolve live it is very possible to even see it live on a set/client monitor. thru a router and some lut boxes. This will work with either tungsten or HMI or any lightsource besides normal fluorescents.

Instead of RED BG, you go BLUE BG. as long as none of the looks (outfits) are blue hue.
This accomplishes two issues
---Blue is not present in skintones so it will not affect. Any bleeding can be desaturated a lot easier by keying.
---Sensor is fed with enough Blue to feed the Green channel. As a result reducing noise.
---In live grade or Resolve Live you can shift the entire BLUE channel to RED.
this signal can be routed to Client monitors via LUT.

>EMERGENCY SET FIX<
Throw a blue filter on the lens. 1/2 CTB / 205 wil cut 2/3 stop but will significantly reduce noise.

>POST FIX<
this is isn't the easiest fix but in a channel based node software like nuke or flame, you can feed the clean red channel into the other two channels and match the matte with a curve adjustment to clean the noise. This is much more effective than using noise reduction software. It also isn't overly difficult but it will definitely take long render time.

Please feel free to correct, make suggestions and advise on this topic

9 Comments

I think you meant to write RGB "Bayer" filtration, which is used in most CMOS cameras, and it definitely has green filtered sensors. ( actually twice as many green filtered sensors as red and blue filtered sensors )

RGB "Bayer" Color and MicroLenses
http://goo.gl/7W6PTH

...I've never shot with a RED camera, but from your description it sounds like the camera has a native "white balance" of 5600 Kelvin like many camera sensors do, so you either light with 5600 Kelvin lights ( like HMI, LED, or Remote Phosphor lights ), filter your lights ( or lens ) for a 5600 color balance, set your camera balance to 3200 Kelvin, or shoot RAW and correct the color balance in post.

Normally when keying out the background in post, you pick a color that is not present in the subject you are shooting ( usually either blue screen or green screen ) and shoot with that. A red background keying color seems like an odd choice because it would effect skin-tones in a bad way.

As far as obtaining the best color with the cleanest image, I think the only way to be 100 percent certain of obtaining the best result is shoot some tests of your set-up using different techniques, and then see which one produces the best image quality after being processed in post. If you don't have time for this, then talk to somebody who has shot a lot of this type of work using the same camera you are using. ( RED offers many different models with different sensors, so you would want to try and find somebody who shoots with the identical RED camera )

August 24, 2015 at 11:18AM

1
Reply
Guy McLoughlin
Video Producer
30084

Thank you for your advice and thank you for correcting my incorrect sensor information.

We were not keying out red. A red cyc was the look.
One of the advices i received was that there was a lack of blue. So i was suggested one solution would be to make the background blue. Then use a hsl qualifier and shift the hue of the blue BG.

We had 2 bodies on set that was shot on before and after and had no issues at all.
The issue rose when shooting on red bg, with tungsten. I was not on the preshoot but the gaffer also told me he did mention he had noise issues in the past when shooting redbg, with tungsten, on a red (ironic) but he was insisted on the lighting to match a previous shoot.
Going back to what i remembered from film, always having a tougher time when there were lack of blues due to it's shorter wavelengths, i still do suspect that the issue and answer was somewhere in the blue.

at the end we corrected it in post.

They do have another scene to take with the similar set up so we are going to try a few things during the prelight. unfortunately I am out of town and will no be part of that shoot.
But will follow up this post on what happens.

Thank You Guy

August 24, 2015 at 9:27PM

0
Reply
Kazu Okuda
Filmmaker
1516

>>>We were not keying out red. A red cyc was the look.

Thanks for the clarification.

>>>One of the advices i received was that there was a lack of blue.

So when they say that there was a "lack of blue", what are they referring to ?

That the blue colors in the video image were weak ?

Just like when lighting a large blue screen or green screen background, you have to make sure that the background color is not contaminating the color of your subject that is positioned on the background. ( ideally you want the background color and the foreground subject lit separately )

You see this all the time with outdoor scenes shot on golf courses or near large green leaf forests, where anything that is not directly lit picks up a green cast from the grass or the leaves. So you have to make sure that your subject is lit by a neutral light, and that the shadows are also being filled by a neutral light, otherwise the shadows will pick up the color of the background which can ruin the shot. ( i.e. The shot of the golf-pro standing on the golf green has his face lit properly, but anything in shadow is picking up the green color from the grass and it looks like crap )

August 24, 2015 at 9:41PM, Edited August 24, 9:42PM

3
Reply
Guy McLoughlin
Video Producer
30084

From lack of blue, i meant not enough information was fed to the blue channel.

Upon reading the article and a few more on bayer sensors, it makes more sense to me. I think my idea was getting there.
So my understanding is, tho there is sensors for all colors, the cpu still determines the dominant color for the pixel matrix and builds/fills the empty sectors by interpreting adjacent, nearest, next, closest field to generate the pixel.

Our set was Red Cyc, Red tones in skin, redish (tungsten) lights, on both BG and Subject, and all the looks were monochromatic (black dresses). There really was not much green nor blue for any of the sensors. On top of that there was a lot of gradations. (i know it sounds horrible)
so the camera pretty much had a hard time generating clean gradations in the pixels. Most of the noise was present definitely in the bg and shadows as you said.

we'll try the set up in hmi as a start for testing and also see if a blue filter works.

the next one will be in two weeks so i will follow up with results.
If you have any further suggestions please let me know

Thank You

August 24, 2015 at 10:06PM, Edited August 24, 10:06PM

0
Reply
Kazu Okuda
Filmmaker
1516

A full CTB is more like 2 stops of light loss, not 2/3rds of a stop.

The color temperature of "blood red" is approx 1000K. The color temperature of a blue cyc is beyond 12000K. If you illuminate a blue cyc with tungsten, you are going to get a much darker blue than if you illuminate with HMI. Similarly, if you illuminate a red cyc with tungsten you are going to get a much brighter red than if you illuminate with HMI.

Yes, blue channel starvation is a problem with underexposed footage in RED cameras. But if you light up a red cyc with Tungsten, you are getting red one red brightness from your cyc, which is going to overexpose, or cause you to underexpose your subjects, which is going to lead to blue channel starvation.

August 26, 2015 at 7:57AM

0
Reply

thanks for your input. sorry i meant 1/2 ctb. but yes then what would be the best way to address this in the future?

August 29, 2015 at 12:54AM

1
Reply
Kazu Okuda
Filmmaker
1516

It all comes down to knowing how the various elements in the scene compare and contrast with one another, and whether your imaging platform has the necessary range to properly handle everything you are throwing at it.

I think the winning answer would be to use an HMI light for a red cyc and rely on the colorist to make its red look how you want it to look. Or start with a full CTB filter over tungsten and then perhaps change down to 3/4 or 1/2 CTB if the R and B histograms of the DRAGON tell you that there's enough density in both channels to get an acceptable image.

But I also think that HMI would be a winner when illuminating a blue cyc, partly because the tungsten spectrum is not going to illuminate blue very well, and partly because I think (but don't know for sure) that when you are shooting highly saturated colors, it's really better to shoot at the native WB of the sensor. Shifting WB from 3200 to 5600 means really stretching out the blues, and that by itself may make your blue cyc look blockier than it should.

September 4, 2015 at 2:00PM

1
Reply

I wrote a typo above. I meant to say "red on red", not "red one red".

August 26, 2015 at 11:25AM

0
Reply

I got to be on the reshoot. Unfortunately the dp and director didn't want to change the set up.
As a test i shot with a stand in model and I threw a 1/2 ctb in front of the lens. It balanced the shadows (on the parade scope). Checked the channels individually and the noise level significantly decreased.

obviously it wasn't the ideal solution but i know next time that it is an option.

September 18, 2015 at 9:09AM

2
Reply
Kazu Okuda
Filmmaker
1516

Your Comment