Mitigating Noise

[Singletracklovr]

Specifically for High ISO, Low-light, Over cropped.

I am trying to figure out the best PP flow to reduce noise in a photo where shooting conditions are not perfect.

I currently have the Standard version of Noiseware and use it on jpegs. I was thinking
it might be better to mitigate some/most of the noise when PP the RAW file since the noise is usually intensified during the RAW to jpeg conversion.

What are your PP steps to mitigate noise?

Thank you in advance for your time and expertise.

[GB1]

Good question! I'm interested in others responses here too .. I have not been super-impressed with the NR software I've seen so far, which either don't work or blur the image too much.

To be fair I have gotten OK results with Neat Image and Topaz a few times, just not anywhere close to consistently ..

[drg]

Noise Reduction either or both in Chroma and Luminescence is somewhat dependent on how much and what type of noise is present.

Generally I do all my major corrections first from cropping/orientation and White Balance/Exposure and color correction etc first.

Second,if I've got effects or treatments (non-sharpening which includes clarity and more than minimal other contrast) they are next. This includes repair and cloning or element removal. These can all add edge noise which sharpening will just multiply like crazy.

Then third I see how much more noise I've introduced or where in the photo it is and how to address it. Individual repairs like the problems from cloning is usually at this point so I'm 'in the flow' for using the pen pad and mouse to smooth out those edges and 'blobs'. Hopefully I've done it right in the previous steps and haven't cause too many added problems.

Magnifying the image to greater than 1:1 can be vital at this stage. Pixel peeping for a reason, i.e. to see noise versus resolution.

Sometime by adjusting the exposure it magnifies the noise generated by the amps when it pushes up the RAW signal data in the camera and PP exposure adjustment makes it worse.

I've used Noise Ninja with advanced masking with fair success by itself. Added to it has often been masking an area and just blurring out as in a background or already bokeh'd region of the image.

There are usually both individual/localized noise fixes using the blur/smudge tools and masking and then an overall NR algorithm to smooth everything just enough.

Then, and only then do your final sharpening. As I pre-sharpen I've usually got a good idea where the worst noise is going to occur. Smart Sharpening also aids in not adding to the noise when it applies to the image but that's another long (not really) entry.

If a clean image is important as in generating offset printing images via a raster process, sharpen before the final noise reduction. A little bit of experimenting and you will find a setting for a range of image sizes (expressed in pixels x pixels) that will always work for the RIP engine you are using that will produce good output.

There are some advanced techniques including difference images, dark frame subtraction, masking layers, etc. to really clean up an image. Using a progressive threshold or variable intensity in a mask is often the only way to hold details where desired and not blur the whole image. If the central photographic elements are sharp/focused/in good exposure range for your photo, then a little noise on them won't hurt too badly. It is that background or highlight scum that really jumps out in a photograph on a display,

When printing you can change the black and white point thresholds of data that is transferred to printing and cut out a bunch of noise and debris at the top and bottom of the data range without penalty.

I use Lightroom and Photoshop for %90 of what I do digitally and unless I've got many noisy images I clean them up by hand and then use a very mild Noise Ninja treatment. Film and a few special cases of imagery require far more work or different software that is almost unique to each process. I can do it in PS but sometimes it takes longer.

There's some other software that have advanced algorithms and more are appearing everyday. The next release of Lightroom (3) has tremendously better NR built in and being non-destructive you can build multiple variants for side-by-side comparison. Export to CS-PS for fine tuning including masks if really needed.

Lightzone software has NR and exposure adjustments that are different enough to warrant looking at it for very high ISO work too.

Picture Window Pro has very powerful NR tools, but the interface has a steep and strange learning curve. Great for film and regular/uniform but maskable noise reduction.

As a lot of noise in both color and brightness (Chroma and Luminescence) comes from bad exposures. At high ISO push that histogram way to the right, right up to the clip point. Even clip the highlights a bit (as long as you don't need them) and then use RAW adjustment to redo the best exposure for that image. Most cameras will tolerate more than you think even if the overall dynamic range of your sensor is not the widest. It's the range in which you capture the image that is important for post processing.

It's a start anyway.

[Singletracklovr]

Wow! dr g, thanks.
This is an awesome start. [insert bowing emoticon]
So much to adsorb and digest.

It looks like my first order of business is to learn how to identify the types of noise.

[OldClicker]

At high ISO push that histogram way to the right, right up to the clip point.

I understand the advantages of 'exposure to the right', but I've never understood HOW this is accomplished. If I have set the shutter speed and aperture to what I need for the shot, the only thing to 'push' with is ISO. At high ISOs, won't raising the ISO (at best) just cancel out any advantage? - TF

[Singletracklovr]

Terry is your question about bumping the camera's ISO setting after you determined the proper exposure or bumping the ISO in PP?

[OldClicker]

Terry is your question about bumping the camera's ISO setting after you determined the proper exposure or bumping the ISO in PP?

About intentionally over-exposing in camera. - TF

[Singletracklovr]

I don't see any advantage to over exposing an image by increasing the ISO as long as the shot is metered correctly.

I am upping the ISO to keep a high SS and smaller aperture in extreme low light conditions. For the obvious reasons, freezing motion and maxing DOF.
The result is usually increased noise. So my original questions is how do I mitigate that noise in PP to improve the photo. Knowing it never be prefect.

here is an example:

shot 7:30PM, spot metered on nest.
ISO was set to 1EV above 3200 to keep a manual camera setting of f8 (need DOF in case mom flew in from behind) and SS of 1/250 to freeze the action.

[Medley]

I understand the advantages of 'exposure to the right', but I've never understood HOW this is accomplished. If I have set the shutter speed and aperture to what I need for the shot, the only thing to 'push' with is ISO. At high ISOs, won't raising the ISO (at best) just cancel out any advantage? - TF

Yes, it would. In a perfect world, exposing to the right is accomplished by capturing an image at normal exposure, checking the Histogram in-camera, then increasing the exposure by either lowering the shutter speed or increasing the aperture. Check the Histogram again, looking for clipping on the right.

Realistically however, you seldom get that kind of time to set up a shot. In manual exposure, you can expose for a +1/3 or +1/2 exposure compensation. Perhaps the best alternative is to shoot in a priority mode with +1/3 exp. compensation, and bracketing on. That gives you +0,+1/3,and +2/3 exposures in short order. One of them should fit the bill. When you get the time, look at the first few groups, as they should give you an idea of where you want to be shooting for the lighting conditions. Adjust settings as necessary.

Now on to noise. First of all, there are two kinds of noise, color and monochromatic. Most images with noise have both kinds to some extent. And one is significantly easier to deal with than the other.

Bob, if you zoom in and look closely at the background in the owl image, you will notice color variations. Specifically, you'll see some red pixels mixed in. That's color noise. Duplicate the background layer, and convert to Lab (Image> Mode> Lab Color). Now apply a Surface blur to both the 'a and 'b channels. Since all of the detail in Lab mode is contained in the 'L channel, you can blur the snot out of the 'a and 'b channels. Use Surface blur though, because you want to make sure you dont blend the blue background into the trees. The Threshold setting will aloow you to do this, while letting you add significant blur to the larger areas. Once done, convert back to RGB and check the background again. The color variation should have disappeared.

Monochromatic noise is much more problematic. I don't think you'll find a definitive pp process for ir, because it's a dynamic problem, and dealng with it depends very much on the individual image. Noise reduction by its very definition is an averaging of pixel contrast (the opposite of sharpening). Therefor, some sort of mask is always going to be necessary to get the best results. At the very least, you're going to want some sort of edge mask, to keep the edges of the image from being softened. The easiest way to accomplish this is to use the Find Edges filter. Duplicate the background, go to Filter> Stylize> Find Edges. The result needs to be desaturated, and probably adjusted with Levels or Curves. When you're done, switch to the Channels palette, and copy any one of the channels (they should all be the same) to a new alpha channel. Then go back and delete the layer from the Layers palette. Duplicate the background layer again, and apply your noise reduction. If I stck strictly to Photoshop, this is either a Surface blur or Noise> Median filter. Then switch to the layers palette and command-click the alpha channel. This loads the channel as a selection. Switch back to the Layers palette, and click the Layer Mask icon at the bottom of the palette. This applies the selection to a new layer mask.

There are a thousand variations on this, and none of them are really a total solution. If you want to be very exacting, you can use Surface Blur as the noise reduction technique, add a layer mask, fill it with black, then choose white as the foreground color and go to the brush tool. You now have a noise-reduction brush with variable softness and opacity (working on the layer mask, of course).

If you don't find the Find Edges filter to be good enough (and I don't....), you can duplicate the background layer, apply a High Pass filter, then a Threshold command. You'll find that varying the strength of the High Pass filter varies the results of the Threshold command. By doing this several times and layering the results (say with a "Darken' blend mode) you can build up a highly intricate mask.

- Joe U.

[drg]

The histogram 'is' the modern exposure tool for digital. Whether you use the fractional EV +/- or other (camera dependent) exposure bias controls or just change the shutter or aperture setting to increase or decrease your exposure doesn't matter!

The idea is to shift that histogram if need be at higher ISO(let's say for the moment 1600 or above) to get as much of the measured exposure information to the right as possible. This will reduce noise as the sensor 'buckets' will have more time to fill with light and require less amplification in-camera. Less amplification, less of that type of noise.

There's a pretty good discussion on the histogram from Sebastian Szyska here at Photography Review at this link:

http://www.photographyreview.com/histogramguidecrx.aspx

It should be a good start on clarifying details without me repeating all of them ad nauseum . . .

[drg]

Yes, there are two types of noise that we generally have to deal with as I mentioned before and it helps with technical references to know that they are usually called Chroma or Chromatic noise and Luminescence.

Some terminology refers to quantifiable data we can repair or modify, and some terms such as Brightness (I still like that term) refer to a perceived component. That is just one example.

The 'L' noise is a component of the sensor design prior to color conversion in Bayer type sensors. There are some other possible components of it including amplification of the signal in High ISO conversions. Luminance noise is very fixable,though not perfect. We know why and where it occurs and we can measure it. This noise is being handled more and more frequently in the camera with techniques such as automatic dark frame subtraction. This has been one of the better PP methods for dealing with it, though there is the levels compression blend filter in PP.
1. You remove color from the image.
2. You invert the image. (This makes the noise more apparent in the next step)
3. You set the levels to the left or from 0 to about 20 usually. The now remaining black spots are noise. Set both Input and Output to these setting.
4. Apply a gaussian blur to the image. (Other blurs will work too)
5. Invert the image again. Will return to mostly black. A few white speckles perhaps.
6. This noise floor can now be subtracted from the original image via a layer.
This is a simplified variant of what is done for certain types of printing within the printer software to get good black without speckles and noise when printed. I did this off the top of my head and if I left out a step, sorry I'll fix it later.

The 'C' noise is the most obvious with the speckling that Joe mentions. However it has to be dealt with and often shows up in as part of color shifts from anomalies associated with linear amplification of a non-linear signal. Thus we do White Balance and Color Correction to eliminate as much of this as we can first.

Then we can start looking for ways to correct edge effects and other sensor errors or prismatic effects (mostly older DSLR's or legacy film lenses at this time) from scattered light.

I believe that the current or newer thinking and information on this topic clearly demonstrates that with higher or expanded ISO range digital photography exposure control plays a large role in producing better images. Using RAW data that has the most information possible recorded, shifted to the right, provides the best opportunity to select your exposure after the fact in post processing. It aint' perfect, but it works. Highlight recovery works better both in practice and theory than trying to recover dark and shadow detail. If the darker portions of an image are left at or below the clip level, or pure black, the noise hides. In the dark or left portions of the histogram report is where most of the worst noise problems are and poor or incomplete exposure leaves little room for fixes.

This combined with image processing software (both in camera and external) that has taken huge strides forward recently has given the option to shoot at previously unheard of ISO equivalence levels. The sensors have not 'really' progressed that far, its what we know about why they make noise and how to negate it or at least move it somewhere that we don't care!

[OldClicker]

Yes, it would. In a perfect world, exposing to the right is accomplished by capturing an image at normal exposure, checking the Histogram in-camera, then increasing the exposure by either lowering the shutter speed or increasing the aperture. Check the Histogram again, looking for clipping on the right.

- Joe U.

But if you are at high ISO, Wouldn't it be better to lower the ISO rather than either lowering the shutter speed or increasing the aperture? - TF

[drg]

But if you are at high ISO, Wouldn't it be better to lower the ISO rather than either lowering the shutter speed or increasing the aperture? - TF

Of course. The High ISO thing is helpful to be able to use a faster shutter (action photography), smaller aperture to increase DOF, or where lighting is poor and either flash isn't allowed or there is just not enough flash available. Shooting a stationary object from a tripod at the best ISO, shutter and aperture just isn't always feasible.

Now if somebody will build a camera with a native ISO of around 3200 we can all shoot by candlelight with little or no noise!

[Medley]

But if you are at high ISO, Wouldn't it be better to lower the ISO rather than either lowering the shutter speed or increasing the aperture? - TF

It IS definately possible, though I'm not quite as sure as drg is. I think it would come down to the individual noise-handling characteristics of the camera. The newer models that handle noise better at high ISO may actualy benefit more from shooting to the right, but the older models would likely benefit more from lowering the ISO.

If nothing else, it would make for an interesting experiment.

- Joe U.

[OldClicker]

Of course. The High ISO thing is helpful to be able to use a faster shutter (action photography), smaller aperture to increase DOF, or where lighting is poor and either flash isn't allowed or there is just not enough flash available. Shooting a stationary object from a tripod at the best ISO, shutter and aperture just isn't always feasible.

Now if somebody will build a camera with a native ISO of around 3200 we can all shoot by candlelight with little or no noise!

Sorry, but I'm still not getting it. Your statement was, "At high ISO push that histogram way to the right, right up to the clip point.", but at high ISO there is nothing to 'push' with??? - TF

[Singletracklovr]

Sorry, but I'm still not getting it. Your statement was, "At high ISO push that histogram way to the right, right up to the clip point.", but at high ISO there is nothing to 'push' with??? - TF

If you set the camera in manual exposure you can move the histogram up by increasing ISO.