Question about noise in low DR photos

paddler4

I have found a pattern with noise that I don't understand. Maybe someone here understands the physics better and can help.

I understand the relationship between noise and ISO, and I also understand that you can reduce noise by exposing to the right.

However, I recently took some shots at ISO 250 and got a lot of noise. This is very unusual--while the 50D is noisy at high ISOs, shots at low ISOs sometimes have so little noise that I do no noise reduction at all. The shots were taken on a drizzly day with very little dynamic range. The histogram was toward the right, but it was very narrow.

I'll post one of the originals below, followed by a cleaned up one. I removed noise with noise ninja in processing the cleaned up one.

Is it generally the case that small DR in a shot will increase noise? If so, what's the physics of this?

Thanks

Dan

kwcrow

I don't understand either, why this would have so much noise
I know that the blue channel is typically the noisiest, could that have something to do with it, since blue dominates this image?

dbd

The sensor outputs in digital cameras are converted to a fixed set of digital levels seperated by steps. With a large dynamic range scene the steps are small enough to often be unnoticeable. In your low dynamic range scene, one of the first things you did was to expand the white and black points of the image (or a similar process). This magnifies the size of the steps. Now the step size may be visible.

If you have noise removal and sharpening setups that work for wide range scenes they may not work right for the stretched image. Thresholds that seperated noise (small) from image features (big) may now pass the noise steps unremoved and enhanced.

Perform the processing of your narrow dynamic range images at as high a resolution as possible. Re-evaluate the thresholds you use for their effect on the stretched image. Consider experimenting with a slight blurring after the stretching step. Your image was shot on a hazy day and cannot become extremely sharp without artifacts. Stretching the dynamic range produces even more artifacts.

Dale B. Dalrymple
http://dbdimages.com

pathfinder

Posterization can frequently be seen when converting computer graphics - like a neutral gradient fill - are converted to 8 bit. Adding just a touch of noise prior to downconversion helps prevent posterization. I suspect it might help here as well.

Posterization can be minimized by the judicious addition of noise in Photoshop.

I know this sounds heretical, but it does work. Just use a very small amount of noise - less than 1% will frequently work.

There is some evidence that shooting at other than even full stops of the base ISO ( 100 ) contributes to more noise also. That is 125, or 250 rather than 100, 200, 400. Ziggy has posted about this several times.

paddler4

Dale,

Many thanks.

What you write makes sense for the general case, but it does not explain this specific one, I think. The top image, which shows the noise most clearly, is not stretched. All I did to that image is read it into LR with default settings and export it to jpeg. The histogram takes up about 1/3 of the possible range, except for a very thin tail stretching left.

The bottom one has lots of editing, including stretching the dynamic range, increasing vibrance, a fairly complex contrast adjustment, and noise reduction.

So, if I understand your note correctly, there still is a puzzle here.

However, your note led me to go back to the image, and I have a hunch. Maybe I am barking up the wrong tree entirely. Maybe what looks like noise if you blow the image up enough (the one posted is of course not full size) is not digital noise at all, but rather the effects of water droplets in the air. I took it during a momentary lull in a light rain where I stood, but there was probably still light drizzle somewhere between me and the mountains.

Dan

dbd wrote:

The sensor outputs in digital cameras are converted to a fixed set of digital levels seperated by steps. With a large dynamic range scene the steps are small enough to often be unnoticeable. In your low dynamic range scene, one of the first things you did was to expand the white and black points of the image (or a similar process). This magnifies the size of the steps. Now the step size may be visible.

If you have noise removal and sharpening setups that work for wide range scenes they may not work right for the stretched image. Thresholds that seperated noise (small) from image features (big) may now pass the noise steps unremoved and enhanced.

Perform the processing of your narrow dynamic range images at as high a resolution as possible. Re-evaluate the thresholds you use for their effect on the stretched image. Consider experimenting with a slight blurring after the stretching step. Your image was shot on a hazy day and cannot become extremely sharp without artifacts. Stretching the dynamic range produces even more artifacts.

Dale B. Dalrymple
http://dbdimages.com

paddler4

Thanks. I had not read Ziggy's posts on this. I'll hunt them down and do some test shots.

pathfinder wrote:

Posterization can frequently be seen when converting computer graphics - like a neutral gradient fill - are converted to 8 bit. Adding just a touch of noise prior to downconversion helps prevent posterization. I suspect it might help here as well.

Posterization can be minimized by the judicious addition of noise in Photoshop.

I know this sounds heretical, but it does work. Just use a very small amount of noise - less than 1% will frequently work.

There is some evidence that shooting at other than even full stops of the base ISO ( 100 ) contributes to more noise also. That is 125, or 250 rather than 100, 200, 400. Ziggy has posted about this several times.

paddler4

Re fractional ISOs: thanks for the tip. I'm convinced. To save others the trouble of searching, here is one of the tests that Ziggy referenced: http://www.northlight-images.co.uk/article_pages/cameras/canon_1ds3_noise.html

The images toward the bottom are particularly persuasive. I just turned off fractional ISOs (1/3 stop) on my 50D. didn't find them useful anyway--they were just turned on by default.

pathfinder wrote:

Posterization can frequently be seen when converting computer graphics - like a neutral gradient fill - are converted to 8 bit. Adding just a touch of noise prior to downconversion helps prevent posterization. I suspect it might help here as well.

Posterization can be minimized by the judicious addition of noise in Photoshop.

I know this sounds heretical, but it does work. Just use a very small amount of noise - less than 1% will frequently work.

There is some evidence that shooting at other than even full stops of the base ISO ( 100 ) contributes to more noise also. That is 125, or 250 rather than 100, 200, 400. Ziggy has posted about this several times.

jfriend

paddler4 wrote:

I have found a pattern with noise that I don't understand. Maybe someone here understands the physics better and can help.

I understand the relationship between noise and ISO, and I also understand that you can reduce noise by exposing to the right.

However, I recently took some shots at ISO 250 and got a lot of noise. This is very unusual--while the 50D is noisy at high ISOs, shots at low ISOs sometimes have so little noise that I do no noise reduction at all. The shots were taken on a drizzly day with very little dynamic range. The histogram was toward the right, but it was very narrow.

I'll post one of the originals below, followed by a cleaned up one. I removed noise with noise ninja in processing the cleaned up one.

Is it generally the case that small DR in a shot will increase noise? If so, what's the physics of this?

Thanks

Dan

If the second image is what you're trying to achieve, then you underexposed the shot by almost 1-2 stops. Underexposing is the quickest way to get noise. Your ISO 250 shot, 1-2 stops underexposed is like a properly exposed shot at ISO 500-1000. When you boost the underexposed areas in post processing, you make the normal noise that is lurking in those darker areas a lot more visible.

As you mention yourself, if you want low noise shots, expose to the right. This is particularly powerful in low contrast shots where you have a lot more room to expose to the right.

This may have also been exacerbated by the fact that the shot is so blue and the blue channel is often the noisiest so you may be seeing noise even earlier than you would in a shot of more balanced colors.