Clear blue skies and noise often go hand in hand, but if you reduce the luminance with a polarizer,does that also reduce or increase the noise? Maybe neither?? Thanks.

Higher exposure cures the sky (ETTR with raw). I often develop two versions from the single raw file: one for the sky, intensity reduced, and one for the rest, and combine them in Photoshop (dummy HDR). That way the sky gets "nice" blue without the noise.

If there are some white clouds, they prevent higher exposure (they would burn out). In such cases a magenta filter helps: it reduces the green intensity, thereby allowing for higher exposure.

The noise is there whatever you do. The question is how apparent it is. With higher signal to noise ratio (from more light hitting the photosites) the less the apparent noise. Blue sky is notorious because the blue channel is more noise-prone. So anything that reduces S/N would exaggerate the apparent noise.

Mark - are you saying that a polarized sky would probably have more noise because it's darker?

I'm shooting with a Sony a900 and Zeiss glass and the noise from a recent trip to the Badlands is driving me nuts. The sky was the clearest blue and I made heavy use of a polarizer. Even though I was shooting at ISO 320 the noise in the sky is terrible. It occured to me that maybe the polarizer and the deep blue sky weren't a good combo in this case. Hence the question.

The polarizer could well be the culprit as it does darken the sky. It's probably easily fixable by selecting and placing the troublesome skies onto a separate layer and treating it with Noiseware or something similar.

This is a myth. The raw channels behave totally equally regarding noise. (As long as the camera does not meddle with the raw data, but that's a different issue.)

The noise is caused by low pixel intensity, and in the sky always the red channel is the lowest.

Look at the attached captures; they show a segment of the sky from a 5D shot (of Mark MacCalmont). The blue is 4.72 EV under clipping (i.e. it is in the fifth stop of the DR), the noise is only 3.57%. The green is comparable; however, the red is 2 EV lower, with noise 7.17%, just twice as high as in the blue channel. (Note, that this is a very low noise level.)

OK, if it's a myth, it's all over the literature. Interesting to see otherwise. Anyhow, my basic point, regardless of which channels it lives in, less light reveals more noise.

I find a lot of contradictory opinions on the subject of noise. I recently took Seth Resnick's D65 course and he said that Luminosity noise shouldn't happen with a good lens - it should be chroma only. I have Zeiss so that comment puzzled me - thinking of my Badlands stuff. I asked Seth again and he repeated the same thing. Pixel Genius, the company that he is part-owner of supplied the sharpening algorithm for LR and because the default NR in LR is color only, I wouldn't be surprised if Pixel Genius didn't contribute that too.

I mentioned Seth's comment to Michael Reichman, who seemed surprised and said that it wasn't his experience.

Here are two guys,much brighter than me, at the forefront of digital camera technology and even they don't have the same opinion about noise! Seth further mentioned, although not in the context of noise, that he doesn't like to use filters, including a polarizer and he much prefers to reduce luminance in the sky using the develop module in LR to achieve the same effect. This got me thinking about the possible relationship between noise and a polarizer.

I believe that this is reasonnably easy to fix. The noise reduction packages like noise ninja shoudl help, if you need more control you have various options including moving to Lab space and selectively blurring the a/b channels. You will probably want to work on a separate copy of you image and then overlay in PS and mask to only impact the sky with the agressive noise reduction techniques.

I had seen these problems in the a900 images I worked with a couple of month ago also. I am afraid that this is the result of the 12 bits technology used in the camera. In such cases it is highly advisable to stick to base ISO which will reduce the problem.

Cheers,
Bernard

Yes, it is a wide-spread misunderstanding. I created a demo specifically for this purpose. The following captures show the red, green and blue channel of different patches of a color checker card shot. I chose the patches so, that the red, green and blue intensity is the same - and surprize, the noise too is the same.





However, this does not remain so: the WB application, color space conversion, saturation (and the demosaicing itself) shifts the color channels away from the original intensity. For example the WB coefficient can double or more the intensity of the red or blue compared to the green, and the noise too gets increased - but now people see the color much brighter, than it is in the raw data.

Do you mind posting the raw file? Btw, was the light not enough for ISO 160 or 200?

Plain nonsense.


This is all right, if the polar filter's purpose was only to make the sky darker and more blue; that can be achieved in post processing as well (and if you shoot raw with ETTR, then this usage of the polar filter is nonsensical anyway).

However, the polar filter's main effect is filtering reflections from non-metallic surfaces, typically from water and glass. This function can not be substituted by any other way.

Well, let's add Jeff Schewe to the collection of bright guys. The sharpening book he and Bruce Fraser authored (and just published) includes quite a bit about noise and about lenses. Noise is covered from pages 34 to 51 and the word *lens* isn't mentioned once. Lenses are covered from pages 52 to 63 and the word *noise* isn't mentioned once.

From page 33: "Sensor noise is present in every digital capture because it's an unavoidable phenominon of the way digital sensors capture light." From there, the whole discussion about the causes and types of noise relate to the sensor and sensor electronics.

Thanks Gabor

Bernard, one can do that, but not to *purposely* stir-up another ancient hornets' nest, I think we have techniques of selective noise reduction which obviate the need to move in and out of L*a*b* with its inconveniences (except at the very front end of a workflow) and data destruction issues for those working in 8-bit.

Just to add to Gabor's post, here are some data from Nikon D3. This is from base ISO showing clear blue sky using 14 bit lossless compression. The exposure was not to the right as much as one would like, as shown in this rawnalize shot. The green is more than a stop below clipping. The white balance coefficients are shown.

Click to view attachment

And here are noise data on the raw channels. Contrary to popular belief, noise is greatest in the highlights--i.e. it increases as the square root of luminance. The datum of interest is the signal:noise, which is highest in the blue channel.

Click to view attachment

Click to view attachment

Mark is correct in stating that the blue channel is usually underexposed and has a lower S:N. With the D3, the blue multiplier is about 1.36 for daylight. With tungsten, which has very little blue, the underexposure is even greater. However, in the case of bright blue sky the blue has a high luminance in the scene and there is no underexposure as Gabor notes. Even without ETTR, the noise in the sky is hardly visible, as shown in this 100% crop.

Click to view attachment

That is complete nonsense and I think that you likely misunderstood what Seth was saying and you should check your notes. Bad lenses might have chromatic aberration and color fringing, but that is not noise.

Well, as I said, standard noise reduction software typically do a good enough job with skies, especially cloudless skies.

I only mentioned this LAB option as just that, one option in the tool kit.

Cheers,
Bernard

bjanes - I don't have to check my notes - I have ears, which is why I asked the question twice. Maybe Seth misunderstood the question (I doubt it) but I definately did not misunderstand the answer. While I and others may disagree with the answer, Seth is a very sharp guy and I don't dismiss what he says lightly. The fact that LR default NR is chroma only means that someone at Adobe agrees with Seth.

Well, you've got your facts wrong. If you go to the Detail Panel of Lightroom 2.4 you will see that Noise Reduction has two sliders: one for luminance and one for colour. Seth is for sure a bright guy, but not all processes of human understanding and communication are perfect.

Geez guys, 'nuff already. I didn't say that LR had only one Noise Reduction adjustment - chroma. What I said was that the default adjustment of LR affects chroma only. The default for the luminance slider is at zero - maybe indicating that Adobe agree with Seth (?) even though my own experience is to the contrary.

This started off as a simple question - did anyone know if a polarizer induced an increase in luminance noise by reducing the luminance of the sky?

I am confident that luminance noise has nothing to do with the lens. Shot noise, the major component of noise in any image, is a property of light itself and is present even before the light enters the lens. If you want a scientific explanation of noise, see this post by a real scientist. Nowhere is the lens mentioned. The main contributors to digital image noise are shot noise, read noise, and pixel response nonuniformity. With long exposures dark current may contribute, but it is not a significant factor for most situations where exposure is considerably less than one second.

Roger Clark is another scientist who has analyzed noise. His method eliminated pixel nonuniformity, and shot noise and read noise account for the total measured noise without taking the lens into account. I would not place much confidence on hearsay evidence of an authority who was likely misquoted.

I have heard of a technique - forget the proper name - used by night shooters, where they do a black frame also, in order to tame noise. I presume it's "sandwiched" with the real night shot and takes the randomness out of the noise. Never been able to find out how it's done though, so I don't know how appropriate it may be.

It's called dark frame subtraction. All you have to do is to take the image of the dark scene and then another with the same exposure time and then subtract it from the original image. It can be done automatically on most higher end cameras, but obviously doubles the time needed to capture the image.