RAW - COLOR NOISE - Canon 50D

[fecanon]

I have a problem with color noise in the raw video (Canon 50D). Appears mainly in frames where there is little movement or no movement. It appears as blue and red over color dots. The strange thing is that the CROP MODE noise is not so strong. The problem persists in all versions of ML. I use mlv_dump to unpack MLV to DNG. DaVinci Resolve Little to edit.
I just tried UFRaw in Gimp and the result is the same. It even say that Crop mode does not suffer from the overexposed color dots noise almost never. 
I am of the opinion that the problem must be recalculated when one pixel counts of 9 pixels real. As I understand it? Or it could be 8 pixels skipped entirely? I'm not sure.
Does anyone know how it could possibly fix different settings ML?

[Levas]

Could be some hot pixels, (are these colors shown up on higher iso's ?) try this:

Taking the lens of the camera and place the body cap on it. Go into the Canon menu and choose sensor cleaning, clean manually, wait a few minutes and turn your camera off.

Put the lens back on and test if those colors are gone...

[dyfid]

Resolves demosaic is not always acceptable however in many cases I find it very similsr in output to Amaze. But it does fail on some ML raw shots. What demosaic option did you use in UFRaw?

I find DCB with a couple of iterations helps a lot in these instances.

[ayshih]

This "noise" can happen with fine bright points and specular highlights.  As with most cameras, the 50D skips rows in non-crop mode, which for example means that the blue pixels next to a red pixel in the image are actually from three rows away on the sensor.  As a consequence, a white feature on the size scale of only a few pixels can gain a color tint due to the poor color sampling of that feature.

It makes sense this noise is not present with motion because motion blur will smear out any small features, and thus mitigates the spaced-out color sampling from row skipping.  If these were hot pixels, motion shouldn't make a difference.

It also makes sense that this noise is minimized in crop mode because every sensor pixel is being read out.  You can still get color effects depending on the presence of very fine features and the effectiveness of the demosaic algorithm.

[fecanon]

RE: ayshih

I do not understand.

But that would mean that everyone has Color Noise Non-Crop Mode RAW

I've seen a lot of edit raw video and it does not seem to me that would be similar problem like me.

[Audionut]

Based on your samples, the noise doesn't look that strong.  I would assume 2 things, most people don't even see it, they smash it with noise reduction.

[dyfid]

You see it clear as day when it manifests itself, on specular highlights such as water surfaces:

https://www.dropbox.com/s/y6z7acar25s86co/000001.dng?dl=0

As mentioned earlier in the thread the DCB algorithm in something like UFRaw or Rawtherapee vastly reduces it to the point it's not visible, where as a demosaic algorithm like ahd fails to deal with it. Resolve's demosaic of CinemaDNG looks like some version of ahd.

GPU assistance and sampling noise from a number of neighboring frames both forward and backwards directions is needed for denoising to be done well. Where as running the problematic sequences through DCB in Rawtherapee solves it, for me anyway.