Untitled Document

I'm doing more and more luminosity blending from RGB channels.

I do a lot of scenic photography where the red channel has some particularly nice effects on the blue sky when blended. It's often the channel I want to use for other reasons as well.

My newest camera is the best, and I can use the red channel here just fine. My legacy files are not as good.

When working on these older files I am looking at a lot of noise in the red channel which shows up in the new composite.

What can I do to reduce this problem?

Ron Kelly

___________________________________________________________________________

Ron writes

If you have a copy of Noise Ninja or Neat Image you can select the red channel and use them to reduce noise in the channel. If too much detail is reduced, you can use the fade command to reduce the noise reduction and increase the detail.

Regards

Raymond McKinley

___________________________________________________________________________

Ron, before attacking the noise/artifacts directly...approach them indirectly!

If in RAW, use the chroma/colour noise reduction, or do it in Photoshop if you prefer using colour blend fades/layers and common filters like despeckle, median, dustnscrathes, surface blur, smart blur, gaussian blur etc (I find combinations of multiple filters at minor settings to be better than a larger dose of only one). One can also use edge masks if haloing or other issues are noted.

It is amazing how much *individual* channel noise/grain can be cleaned using RGB/CMYK mode colour blend filtering or AB in LAB mode. Then one can attempt to reduce the single channel noise after the indirect approach.

Best,

Stephen Marsh.

___________________________________________________________________________

Intensifying skies often makes a big difference in scenic images, so the luminosity blend from the red has become quite popular--leading a lot of people into the same problem that you are finding. *No* digicam I've ever tested consistently produces a good red channel in skies without resorting to excessive blurring elsewhere.

The specific problem of blending with the red into a sky is now so common that I devoted a full page of PP5E (p188) of examples showing it: pieces of sky from one digicam aimed at professionals and another at prosumers, shown at 100% and 300%, composite color, then red channel, then red used as a luminosity blend. The red channel is visibly noisy when seen alone but not when the composite color is viewed. However, when the red is used for a luminosity blend the noise becomes offensive in the composite.

There are two problems: how to get the noise out of the red, and how to restrict the move so that it doesn't hammer other areas of the image (such as detailing in the clouds). Stephen correctly suggested going to LAB and blurring the AB but this may not be sufficient if the noise is severe.

Sometimes the noise is so bad that we need a couple of steps to suppress enough of it to get a suitable blending channel. That said, the tool of choice AFAIC is applying, on a duplicate layer, the Surface Blur filter to the red channel. It's got a Threshold setting that can be set to avoid blurring edges, such as where the sky hits the horizon. If you're successful, even if there's damage in the lower half of the image, you can easily lasso it out on a layer mask without worrying about a complicated selection.

If there's still an issue, you can take the layered file into LAB and use Blend If to exclude anything that isn't strongly B-negative on the bottom layer. That will protect the cloud detail, and chances are that it will knock out everything below the sky as well.

I learned the hard way in the advanced classes that if the red noise is *really* bad, you can get significantly better results by duplicating the RGB file, converting it to 1.0 gamma, Surface Blurring *that* red channel (fewer levels of difference between normal sky and noise; easier to target without damaging anything else), reconverting back to the normal working space and using the resulting red channel to replace the original, whereupon luminosity blending should work fine.

Dan Margulis

___________________________________________________________________________

Dan, Stephen, Raymond:

Thanks for your suggestions on dealing with red channel noise.

It's going to take me a while to digest and test your suggestions.

Much obliged,

Ron Kelly

___________________________________________________________________________

Agreed Dan.

This has become a personal gripe for me! <g>

I never see anybody recommending to *first* use indirect methods of cleaning problem channels, usually the blue in RGB.

If a single channel is noisy, they just say to directly blur the channel as it is weak and does not make the whole image suffer, usually the B or Y. This also destroys detail too. Indirect colour component filtering can reduce damage while not affecting detail in the same way that filtering in normal blend or direct to the single channel does. A subtle but incredible cleaning affect on single channels can thus be obtained with indirect colour filtering.

Then as a *secondary* step, as you suggest - better noise reduction methods that do not degrade edges can then be used.

It just bugs me that the general advice out there is to destroy the channel with a direct blur as the first step - when a simple colour blending mode or LAB blur or other filetring can actually generate averaged detail from where a user thought there was only JPEG blcoking. The first time you see noise removed and detail appear due to colour component averaging is amazing when compared to simply filtering the channel directly.

Perhaps it is becuase this method attacks the problem indirectly, that it is not so obvious to many as a first step.

Sincerely,

Stephen Marsh.

___________________________________________________________________________

Hi Stephen,

What exactly do you mean by "indirect colour filtering"?

All the best

Bart Walczak

___________________________________________________________________________

Hello Bart, let me start by going over the issue again.

One usually starts with an RGB file, or perhaps CMYK. Upon inspecting the various separate channels - it may be noted that there is objectionable grain, noise, JPEG damage, moiré or similar in the separate R, G or B channels (or CMYK). Most often in the B or Y, but it can also be an issue in the R or C channels too (it depends on the image colours and content).

So, the obvious way to solve a single noise channel for most folk is to *directly* filter the offending channel, as this is where the problem shows itself upon first inspection. So the single channel is targeted and gaussian blur or dust and scratches, or smart blur, or surface blur or whatever is run. This does remove junk, but can also remove detail too. There is also the reduce noise command of CS2 that can work on single channels, which is another direct approach.

In the case of the B or Y channel, little harm is done to the final appearance of detail in the composite image as this colour is more forgiving than say the green for most images. That is still no reason to needlessly damage the channel, in my opinion.

In RGB and CMYK modes, both luminance and colour are contained in each separate channel. By performing the noise reduction filtering on the colour component in say LAB mode or in RGB/CMYK with colour blends/fades one can filter the colour component noise while preserving the important luminance information. This can have a subtle and not so subtle effect on individual channel noise, more so in the B then the R and usually little in the G (which is similar to the L in this respect so not much of an effect is obtained).

This indirect method is simple to demonstrate. One can take history snapshots of current image (one of each separate R G B channel), then convert to LAB. Then one can perform noise reduction in the AB channels then convert back to RGB and take more snapshots of each R G B channel and compare to the originals.

The same can be done in say CMYK without converting out of the existing colour mode (a good thing) - by duping the layer and filtering with the layer set to colour blend mode. Then one can target the individual channels in the lower layer and toggle the upper colour blend layers view on and off for each channel to see what the *indirect colour component* noise filtering has done to each separate channel (for good or bad, this is not perfect).

Compare this to simply *directly* targeting a single channel and reducing noise. One can reduce a lot of directl filtering work, by first using an indirect method to address the issue.

So, by reducing the noise and junk in the colour component of the image (hue and saturation garbage) one can gain significant benefits to the single channels of the image. One has to be careful not to desaturate small or bright colours with large colour noise reduction moves and to beware haloing produced from filtering as this is not perfect but my point is that it is often a better first approach than destructive noise reduction methods. The process also has the side benefit of generating the appearance of missing detail in a damaged channel as part of the colour averaging process (perhaps similar to the demosacing process on raw camera data).

I use the term "indirect" or "indirect colour component filtering" as it seems to best describe the process (an oblique tactic rather than meeting the issue head on).

If English is not your first language Bart, or if I am still being too cryptic in my explanation don't hesitate to ask for further clarification!

Regards,

Stephen Marsh.

___________________________________________________________________________

Hi,

So this is essentially either filtering the AB channel in LAB or duplicating an image and applying filter to a duplicated layer's channel and then using mask or blend-if to recover the detail?

I get it. I thought it was something more complicated, at least the name I was not familiar with implied something like this. ;)

Thanks for your answer!

All the best

Bart

___________________________________________________________________________

Stephen Marsh writes,

That's right. Skies being used for luminosity blends are a special case. I can't think of any other category where I recommend blurring anything in RGB as a first resort rather than a last one.

If we happen to be in LAB anyway slight blurs of the A and B are almost cost-free and very effective in reducing noise. A bigger problem than people blurring in the wrong place, though, is overblurring in the first place. I keep seeing demonstrations of how we can wipe away (at considerable cost to the rest of the image) grain or noise that's harmless *unless* some major stress is put on the image.

JPEG artifacting is an example. We don't see it, provided the image is used with little change at the intended size. As soon as something major is done that happens to enhance those artifacts, though, they become objectionable.

It's the same way with skies in digital captures. The red channel, taken alone, usually is pretty ugly. Most of the time it isn't ugly enough to make a difference in reproduction--otherwise camera manufacturers would be hearing screams of outrage. But if we're using that channel as the base of a luminosity or blend, like Ron and many others do, then the noise gets stamped into the *green* channel also and now the sky *is* visibly noisy. (Same thing happens if, as some do, we multiply the RGB into itself and use that for the sky).

Blurring in Color mode in LAB or RGB is a great tool in other contexts. In skies, though, it usually isn't sufficient to smooth the red channel enough for a blend, because the initial noise is so severe that there is a luminosity as well as a color issue. In that case, given that a blur directly to the red channel is needed eventually, it's better to do it *before* the Color-mode blur than after.

If we blur in Color mode first, it softens and spreads the noisy pixels. That makes it more difficult for the Surface Blur filter to operate and may result in a bumpy, rather than noisy, appearance after the blend. So the correct procedure IMHO is to first get rid of the noise in the red, then do the blend, and *then* the blur to the AB channels.

Dan Margulis

___________________________________________________________________________

I can appreciate this issue/concern Dan and have seen similar in the past but not with the specifics being mentioned here. Also, I have not been using this filter to a great extent yet (I use NeatImage if I have to deal with noise). I will add this to my "to do" research list. Cheers! (I think?)

Best,

Stephen Marsh.

___________________________________________________________________________

Yes, Bart - it can be as simple as a LAB or colour blend/fade in RGB/CMYK and it can get more complex! But that is it in a nutshell. Do you have a better term that would lead to less confusion, instead of using *direct* or *indirect*? Which to me seems to sum up the method.

Sincerely,

Stephen Marsh.