Noise in the red channel

Started by eightcore, April 22, 2015, 11:09:53 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

eightcore

Hi,

I've been recording RAW video with my 50D since December 2013. It took me 1.5 years to find out that the image noise level in the red channel is extremely high.
Take a look at this screenshot.

The t-shirt is plain red, no structure at all. But in my footage it's covered by a ton of low frequency noise.
For testing purposes, I separated chrominance from luminance in Resolve and took a look at the noise levels of both of them. Result: The luminance channel is almost noise-free, so are the blues and greens, but the reds look like this.


  • I'm using BMD Film to VisionLOG LUT.
  • I have highlight tone priority enabled and configured a ML digital ISO gain of +1.0 EV. In my tests, this has led to the best dynamic range. Dark areas are quite noisy, which is why I denoise my footage with Neat Video (just a tiny bit, there is no visible loss of detail)
  • I can't get rid of this noise with Neat Video, it's frequency is too low and the irregularity too high.
  • It doesn't matter wheter I denoise before the LUT is getting applied or afterwards.
  • I tested a second 50D on this behaviour. Same result.
  • When using the Linear to BMDFilm 4K LUT, the issue is almost invisible, but the more I try to grade it towards the gorgeous colors of the VisionLOG lut, the more appearant the noise gets.

Is this a known issue? Do you have any advice in terms of settings or post processing? Please keep in mind that I would like to get the best dynamic range and if the solution would noticeable reduce the DR, I ain't even averse to switch to another camera.

Best regards,

eightcore

Levas

The picture looks very cool (not in a good way, but BLUE color way  :P)

Do you load the raw footage in resolve, Try cranking up the white balance to 6000Kelvin or something ?

Edit:
How do the video scopes look in resolve, can you send a screen shot of the waveform monitor in resolve ?


eightcore

Hey, thanks for your answer!

So, here's a warmer version. And that's the waveform (damn, I revealed his face :-))
Of course the overall image got even noisier as it has more red in it. Here's another obvious example



EDIT:

A few moments ago, I tried to grade it without any LUT applied (just normal BMDFilm colors), and it looks much better! At these noise levels, Neat Video will work very well. Red is obviously way less saturated than with the LUT getting applied.
I did it via TeamViewer though, I'll check the exact result tomorrow. Thanks so far! Stay tuned.

Levas

Hi again,

That's still a lot of noise indeed  ;)
But I see where it goes wrong, I didn't read your camera settings good enough...

Here are some hints when shooting in raw format 8)
NEVER use highlight tone priority, unless you need noisy images ;D
And NEVER use DIGITAL gain in camera.

Highlight tone priority lets your camera underexpose by 1 stop(to save highlights) and then add digital gain so the picture doesn't look underexposed.
Digital gain adds noise.
So you use 2 settings that use digital gain in camera, turn them off!
No highlight tone priority and no magic lantern digital gain.

How to expose:
Set camera to manual mode, set shutter speed to 1/50 (if you're using 24/25 fps)
Use "raw histogram" option in magic lantern menu, so you see the raw histogram in live view.
Now use diafragma and ISO settings (full stops, 100-200-400-800 etc..not the in-between iso's of 125, 160 etc...) to adjust the histogram so it touches the right side of the histogram.

With these instructions your red channel should look much better  :D

Kharak

What tool are you using for converting MLV to DNG/CDNG?

Did you try raw2cdng ?
once you go raw you never go back

eightcore

I use "Magic Lantern RAW Video Converter 1.9.1" (I'm on Windows). It doesn't export cdng but simple dng files. I've never tried raw2cdng, but what should be different? I think every Windows tool uses mlv_dump.exe, or am I wrong?

What I noticed this evening: Regardless of whether HTP is on or off or which ML digital gain value is set, the footage looks exactly the same in Resolve. Same DR, same noise levels, a 100 percent match. I still can't believe it, but I've tested it six or seven times.

HTP & ML digital gain off (ISO 100)
HTP on, ML digital gain off (ISO 200)
HTP on, ML digital gain +1.0 EV (ISO 400)
HTP on, ML digital gain +2.0 EV (ISO 800)
HTP on, ML digital gain -0.3 EV (ISO 160)

During recording, the differences were clearly visible.


Edit: Apparently it's a known fact that HTP doesn't affect raw video. So that can't be the reason for my issue.

Levas

Hmm, apparantly HTP and digital gain only affects standard .mov recording...  ???
Weird, I would expect the ML digital gain to have effect on the raw footage.

Then I don't know, maybe this noise is normal for 50D.
In the first example of your first post, there is clearly not much light on the red shirt, the person is standing in the shadow with the sun/light behind him. So this result is actually not that bad.
And think about it, the color red in the shirt is only recorded by the red pixels in your camera, there's no green and blue in it, so actually the shirt is recorded by only 25% of the available pixels on the sensor.

So for the future, never use basic red or basic blue clothing and search for a better spot where the person is better lit by the available light.

Audionut

The red channel is also not as sensitive.  It's generally a stop (or more) under green.

So a red shirt (in the shadows), with a light source that is blue/green, recorded by 1/4 of the available pixels, with pixels that are less sensitive.