Luminance Noise - Can You Tell Me More?

Sigma 18-125mm f/3.5-5.6 DC for Canon

05/10/2005 11:51:13 AM · #1

Am working through the excellent Camera Raw book by Bruce Fraser and am currently reading about the Details tab including the Luminance Smoothing and Colour Noise Reduction.

This is one area in which Fraser doesn't really stop and explain what luminance noise/ colour noise are - just how to use the controls - and I'd appreciate any clarification...

PollyBean

Canon EOS-5D

Canon EF 24-105mm f/4.0L IS

05/10/2005 01:13:57 PM · #2

I read that the red channel shows up more noise than the green and blue channels.

I really know nothing about this. Maybe some of you do, and could expand on this?

Bear_Music

Samsung Galaxy S24 Ultra

Sony 24-70mm F4 Vario-Tessar T* FE OSS

05/10/2005 01:32:02 PM · #3

If you switch modes to lab color, you'll see the 4 channels, including luminance, in the channels tab in the layers pallette. You can work on each channel separately there.

Robt.

ElGordo

Sigma 18-200mm f/3.5-6.3 DC OS for Canon

05/10/2005 01:37:30 PM · #4

Originally posted by PollyBean:

I read that the red channel shows up more noise than the green and blue channels.

I really know nothing about this. Maybe some of you do, and could expand on this?

This seems to vary from one type camera to another. But most of the time, the cleanest channel is the green. My Nikons exhibit more noise in the blue channel while the Fuji is noiser in the red. To make a realy clean B&W I almost always use just the green channel.
I believe the green is cleaner because there are twice as many green pixels as red or blue in sensors that use Bayer RGB masks. Early on, Canon made some cameras with CYM+G sensors, the Pro 70 and Pro 90 IS. Both these cameras exhibited lower noise than my Bayer masked cameras.

Message edited by author 2005-05-10 13:41:14.

Canon EF 17-40mm f/4.0L USM

05/10/2005 01:45:21 PM · #5

In almost every camera: out of every 4 sensors, two are green, one is red, one is blue. Apparently this is because in the human eye there are more green cones than other colours so the Bayer array reflects this. However, I have read on some obscure site that this is controversial. Nonetheless, this is why they're made that way.

The Bayer sensor array:
RGBGRGBGRGB
GBGRGBGRGBG
RGBGRGBGRGB
GBGRGBGRGBG

For this reason, the green channel is usually has the least noise, as there are twice as many green sensors as either blue or red.

However, noise levels in the real world don't always reflect this. For example, my FZ10's cleanest channel was the red channel.

The blue is almost always the noisiest... apparently in evolutionary terms blue isn't a very useful colour for humans, it's just sky which also suffers from polarisation effects and the brain isn't too concerned with it. I believe polarisation may have something to do with why blue is so noisy in cameras, but I could be wrong.

Anyway, all this is off the top of my head so some of the details may be slightly off... but I'm sure someone here will correct me!

dahkota

Canon EOS-7D

05/10/2005 01:48:52 PM · #6

Originally posted by Kavey:

Does this help?

"CCD noise comes in two parts, luminance noise and chroma noise. Luminance noise makes an image look grainy on screen, but is usually not visible when printed. Chroma noise is visible as random red and blue pixels and is usually less obvious both on screen and printed. Removing luminance noise reduces the sharpness of the image and removing the chroma noise damages some of the correct color. So noise reduction is a balance between how much softness and color damage you are willing to accept vs. how much noise you want to remove. Unless you use an uncompressed mode with your camera, JPEG artifacts also get added into the mix."
from: //www.michaelalmond.com/Articles/noise.htm

d, who learns something new every day.

Canon RF 24-70mm f/2.8 L IS

05/10/2005 01:51:27 PM · #7

Ah, I see I didn't really answer the question!
Sensors work by registering a current when they receive a certain amount of photons (light) depending on the ISO.
They are also sensitive to heat as well as light though.
They are also slightly sensitive to neighbouring sensors.

It's these other two factors that cause 'noise', which is more obvious in low-light conditions as you can see individual pixels of colour show up against your black photo. High ISOs are noisier as you are forcing the sensors to become more sensitive to heat and their neighbours' current.
Again, please correct me if I'm wrong.

kirbic

Canon EOS R5

05/10/2005 01:56:30 PM · #8

Luminance noise is noise that is present in all three color channels, and so it shows up only as differneces in brightness, not "flecks of color." Color noise is noise that differs beween channels, and therefore show up as flecks or spots of color.
In color digital imaging, IMO it diesn't make a lot of sense to talk about luminance noise, but makes more sense to tald about fixed-pattern and random noise in eac of the three color channels. The random component of the noise is not correlated between channels, and is therefore almost purely "color noise." The fixed pattern noise is due to both per-pixel senstivity differences and to read-out issues, and can be correlated between color channels, therefore is not purely color noise nor luminance noise.
The reason I say I would rather make the distinction between random and fixed-pattern noise is that we must deal with them in much different ways, assuming we want to minimize the noise to start with, not try to eradicate it later.
In order to minimize fixed-pattern noise, dark-frame subtraction is used. This uses a frame taken with the same exposure time and ISO, but with no light hitting the sensor. This frame is subtracted from the image, and because only the noise is present in the "dark frame" that's all that disappears from the image.
To limit random noise, multiple exposures of a scene are averaged. The random noise changes, but the image data does not, and therefore the noise is averaged out, while the real data is enhanced.
Both of the above techniques are powerful ways to improve the noise qualities of images. The second has the limitation of requiring a static scene and a tripod-mounted camera. Also note that the second technique is never legal for DPC challenge submissions, while the first is legal if done in-camera. Many fixed-lens cams and some newer DSLRs are capable of on-camera dark-frame subtraction.

ElGordo

Sigma 18-200mm f/3.5-6.3 DC OS for Canon

05/10/2005 01:59:32 PM · #9

I think you got it nailed pretty well Bobsterlobster. The only thing I would add is that at higher ISO settings the amplifiers that condition the pixel data also contribute to the noise mix.
Manufacturers continue to develop lower noise sensors and current DSLR cameras are magnificent in that regard. I think we shall soon have available cameras that are completely noise free.

Canon RF 24-70mm f/2.8 L IS

05/10/2005 02:00:09 PM · #10

Originally posted by kirbic:

I read somewhere an article showing through rigorous testing that averaging multiple exposures to reduce random noise doesn't work. It might work for astrophotography, and the theory sounds good, but it actually doesn't work for DSLRs. Anyone know the article I'm talking about?

kirbic

Canon EOS R5

05/10/2005 02:21:05 PM · #11

Originally posted by BobsterLobster:

It most certainly does work! I have used it for astrophotography, but it can be used just as well for terrestrial work. Noise is reduced as the square root of the number of exposures, so it takes four exposures to reduce the noise to 1/2 it's single-exposure value.

05/10/2005 02:33:06 PM · #12

Darn, wish I could find the article that showed the results of taking photos of the exact same scene with and without stacking.

Kavey

Sigma 18-125mm f/3.5-5.6 DC for Canon

05/10/2005 02:39:49 PM · #13

Between you all, I'm getting an idea! :o)

Bear_music - I'm not having a problem figuring out how to see it or even to adjust it but thanks anyway - I just needed a better idea of what exactly I'm seeing and adjusting (ie how it's caused, what the fixing sliders actually DO to fix it) etc.

Bobster, Fraser's book says same re the arrays - twice as many green than red and blue because human eyes are most sensitive to green. Was aware of the way the photon sensors worked but hadn't realised that they were slightly sensitive to each other - so some noise is the effect of an unwanted bleed from one to another?

dahkota, that quote definitely helps - the bit about luminance noise being more of an issue on screen but colour noise being more of an issue in print for example is something I've not read before.

Kirbic, had come across the technique to subtract fixed pattern noise before but hadn't really thought much about the random pattern noise. That said, where the non-fixed pattern noise is due to sensitivity bleed form a neighbouring sensor (for want of better way of expressing it) is that bleed down to how bright the light that the neighbouring sensor in question is trying to record and if so, would multiple shoots have any impact, i.e. wouldn't that pixel continue to cause bleed/ noise in the information recorded by it's neighbour? I'm probably not getting my head around it quite right but it doesn't quite make sense to me why multiple exposures would lead to different noise patterns? Is it because there are also other reasons for noise patterns? Just wondering out loud...

kirbic

Canon EOS R5

Canon RF 24-70mm f/2.8 L IS

05/10/2005 02:51:17 PM · #14

Originally posted by Kavey:

...Kirbic, had come across the technique to subtract fixed pattern noise before but hadn't really thought much about the random pattern noise. That said, where the non-fixed pattern noise is due to sensitivity bleed form a neighbouring sensor (for want of better way of expressing it) is that bleed down to how bright the light that the neighbouring sensor in question is trying to record and if so, would multiple shoots have any impact, i.e. wouldn't that pixel continue to cause bleed/ noise in the information recorded by it's neighbour? I'm probably not getting my head around it quite right but it doesn't quite make sense to me why multiple exposures would lead to different noise patterns? Is it because there are also other reasons for noise patterns? Just wondering out loud...

At very high ISO values, a major component of the radom noise is the quantum nature of light! Light consists of individual photons, and as fewer and fewer photons are gathered per cell, the statistics of random processes guarantees that random variation in the number of arriving photons will increase.
Another source of random noise is thermal noise in the electronics. This source of random noise can be reduced greatly by cooling the sensor, and is the reason that CCD sensors for astrophotography are actively cooled.
Bleed-over from adjacent cells wouldn't be considered noise, but would manifest itself as "blooming" or similar image defect.

Kavey

Sigma 18-125mm f/3.5-5.6 DC for Canon

05/10/2005 02:58:39 PM · #15

That helps, thanks.

Actually I guess that's why the shadow areas are always the most noisy... more prone to being affected by random photons hitting the sensor...

05/10/2005 03:00:51 PM · #16

Originally posted by kirbic:

Originally posted by Kavey:
...Kirbic, had come across the technique to subtract fixed pattern noise before but hadn't really thought much about the random pattern noise. That said, where the non-fixed pattern noise is due to sensitivity bleed form a neighbouring sensor (for want of better way of expressing it) is that bleed down to how bright the light that the neighbouring sensor in question is trying to record and if so, would multiple shoots have any impact, i.e. wouldn't that pixel continue to cause bleed/ noise in the information recorded by it's neighbour? I'm probably not getting my head around it quite right but it doesn't quite make sense to me why multiple exposures would lead to different noise patterns? Is it because there are also other reasons for noise patterns? Just wondering out loud...

At very high ISO values, a major component of the radom noise is the quantum nature of light! Light consists of individual photons, and as fewer and fewer photons are gathered per cell, the statistics of random processes guarantees that random variation in the number of arriving photons will increase.
Another source of random noise is thermal noise in the electronics. This source of random noise can be reduced greatly by cooling the sensor, and is the reason that CCD sensors for astrophotography are actively cooled.
Bleed-over from adjacent cells wouldn't be considered noise, but would manifest itself as "blooming" or similar image defect.

I would have thought that the quantum nature of light has very little effect on sensors... the random nature of quantum physics applies to individual events which force a single photon/particle to chose a statistical event. The sensors in a camera are not forcing one photon to decide whether it is a wave or a particle, and the effects of heat would have millions of times more effect than this quantum issue. Are sensors really that sensitive that they can detect one photon? I would have thought that it is hundreds/thousands that it's looking for at the highest ISO, in which case we're not on the quantum level.

Canon EF 17-40mm f/4.0L USM

05/10/2005 03:08:26 PM · #17

Well, I stand corrected...

//www.radimg.com/doccd.htm

I had no idea they were that sensitive!

dahkota

Canon EOS-7D

05/10/2005 03:09:35 PM · #18

Kavey,

To be honest, I really didn't know what it did either. Using "Raw Shooter," I would slide the Luminance Smoother filter back forth, looking for a change, usually going, "Hunh." And not much else. Thanks to your question I now know why. And to think, I was making my already unsharp images even unsharper! I really need to learn digital processing in all forms. Playing around works only so long... :)

d

05/10/2005 03:14:47 PM · #19

Wow... a fantastic read if you want to give your brain a workout and understand the quantum effects of noise on a CCD!

//www.fbw.hs-bremen.de/~olbers/sternw/sonne/doc/noise.htm