PDA

View Full Version : Noise issues



Kirk Gittings
25-Sep-2006, 20:50
I am working on a problem commercial image (a pair of twilght shots, merged, slightly underexposed, scanned color negative) with a ton of shadow noise. I use Noise Ninja but frankly have never been entirely happy with the resilts as they are always softer than I would like at a setting that actually gets rid of the noise. Am I expecting too much? Am I doing something wrong? What do you do to remove noise.

Kirk Gittings
25-Sep-2006, 21:00
Not that it really matters, but here is the image.

Tim Lookingbill
25-Sep-2006, 22:17
Kirk,

I don't use Noise Ninja. I'm currently playing around with a technique using selective Dust and Scratch in PS on a negative containing quite a bit of grain noise in the shadows of foliage.

If the noise is multicolored (chromatic) I convert to Lab and blur the A&B channels which makes the noise monochromatic. After converting back to my RGB working space, I single out these flat areas of noise so I don't apply Dust and Scratch to detail using one of the RGB channels with the most contrast (in my case blue for the foliage). Then I exaggerate the contrast to get a B/W posterized look make selecting with the Magic Wand easier. I could do a Layer Mask and paint back in the areas I want Dust and Scratches applied, but I like the one click Magic Wand method.

I apply Dust and Scratches zoomed in tweaking the settings to get a fine dithered pattern that doesn't flatten shadow depth. You'll have to use your own judgement on this to arrive at your own settings. Depending on resolution mine being 4800spi I apply a 4 Radius and a Threshold of 12 and adjust to get the right texture without forming clumps.

I prefer this method instead of blurring which gives a kind of burnished clay feel to noise in film. Blurring works better smoothing banded noise in skies.

Hope this helps.

Gordon Moat
26-Sep-2006, 01:18
I think a better scan would save you a great deal of time. A better scan would also represent your work in the best possible way. Of course, if this in only for internet display, or for PhotoShop practice, then there are several approaches.

On a low to mid range scanner, if you can control the exposure of the scanner, you can get sometimes better results overexposing one scan, then underexposing another, which sounds like what you did. After that you combine the two scans. I am a little surprised you didn't scan once as a transparency, then invert in PhotoShop. The way to do that is to get the scanner Histogram for each colour channel at a broad and even point, basically capturing the maximum information. It can also be a slight advantage to scan too light, then bring in most of the dark or black areas by controlling K (Black) in CMYK mode as a last step before printing/publishing.

Green is the compliment of magenta, blue is the compliment of yellow and red is the compliment of cyan. However, all those three RGB channels contain a component of black used to form a K channel. Controlling that as a last element can avoid some of the issues of shadow problems, since you leave it out of the scanning stage and control it later. Sorry if I don't explain this well in words; these reply boxes are often too small to go into much detail, and I also prefer showing people rather than trying to explain it only with words. Hopefully some of this made some sort of sense.

One thing interesting about PhotoShop is there can be often three or more solutions to nearly any problem in imaging. Quite often those come about due to deficiencies of film scanning, more apparent and troublesome at low to mid range of scanning gear, though sometimes with high end gear not operated at optimum. It is really tough to tell in a small JPEG, since the compression artefacts are somewhat heavy.

Anyway, likely you will get a ton of answers and approaches. I come from a pre-press and publishing/commercial printing background, and prefer working in CMYK. I also tend to think in terms of how inks mix on paper, what the brighness or whiteness of the paper does for the final prints, and how all those various elements best work together . . . take the desired results and work backwards to figur out how to accomplish what you want. Seems most photographers today only want to work in RGB, so I doubt anyone will agree with the ways I work on images. Best of luck with this one.

Ciao!

Gordon Moat
A G Studio (http://www.allgstudio.com)

Walter Calahan
26-Sep-2006, 05:14
You don't need to buy add ons to fix noise. LAB color is the tool you need.

You want to keep the detail sharp but get rid of the color noise.

Convert the image to LAB color. Turn off the "L" channel (this is where your detail is).

Add a slight amount on Gaussian Blur to both the "A" and "B" channels. Just enough to blur the noise. Say 0.3 to 0.5 to start.

Then turn on the "L" channel and turn off the "A" and "B" channels. Do an unsharp mask to only the "L" channel to snap up the detail.

Turn on all the channel to see the results.

This is best done to a duplicate layer of the original image (command J). This way if your correction is too much you can reduce the affect by dialing back on the percentage of impact the duplicate layer has on the original layer, say change the slider to 50%.

You end up with a sharp image with no noise.

Easy. Any questions, please ask.

http://www.walterpcalahan.com

David Luttmann
26-Sep-2006, 07:09
Kirk,

I use the same method as Walter, but I combine it with a light "dusting" from the Neat Image software package. I find the control far better than Noise Ninja as it seems to maintain fine detail better while removing more Chroma noise.

If you try it....get the 16 bit Pro version.

Best of luck.

Bruce Watson
26-Sep-2006, 07:43
I am working on a problem commercial image (a pair of twilght shots, merged, slightly underexposed, scanned color negative) with a ton of shadow noise.

Do you mean noise in the shadow areas of the final print? That is, from the very thinnest areas of the negative? If so, that would implicate your scanner and/or software. You shouldn't be getting any noise at all from the easiest-to-scan parts of the negative from any scanner and/or software. They should all do the easy stuff really well.

If you are instead getting noise from the densest part of the negative you may simply be running out of the scanner's capabilities. In that case, a drum scan might save you considerable time, effort, and aggravation. Of course, YMMV.

Tim Lookingbill
26-Sep-2006, 07:55
Gordon,

You said..."scan once as a transparency, then invert in PhotoShop. The way to do that is to get the scanner Histogram for each colour channel at a broad and even point, basically capturing the maximum information."

I've been trying out this technique with negatives on my Epson 4870 flatbed and I can't seem to get each channel to spread out on the output histogram. Or are you talking about the input histogram?

I've been using the color management feature to get an accurate depiction of the orange mask with the added advantage of having the scanner write the data in a wide Target space like ProPhotoRGB. But with some scenes and film brands-Kodak UC 400-I end up with some shots having weird hue and saturation hot spots inverting and processing in PS using the NegPos technique.

Each channel histogram before setting B/W/neutrals never spans the entire 0-255 mapped graph once captured and viewed in PS. They're always narrow centered clumps within the graph. After applying these settings is when I get normal fully spanned histograms. Am I doing something wrong, because I've used almost every techique I've come across on the web including NegPos and I never can get fully spanned histograms on capture.

As for eliminating noise through exposure, I get film grain noise in shadows no matter how much I blast Kodak UC 400 negatives with adjusted scanner exposure. My understanding is, if it's in the film, no amount of scanner light can bleach it out. I believe it's the angle of slope chosen for the gamma type tonal correction that tends to exaggerate film grain noise much like what some have experienced processing RAW digicam files.

Only my observations. I'm new to scanning negatives and still trying to figure out a way for getting consistant results. So far it's been hit or miss and I can't seem to understand what influences such results.

Gordon Moat
26-Sep-2006, 10:49
Good morning Tim,


Gordon,

You said..."scan once as a transparency, then invert in PhotoShop. The way to do that is to get the scanner Histogram for each colour channel at a broad and even point, basically capturing the maximum information."

I've been trying out this technique with negatives on my Epson 4870 flatbed and I can't seem to get each channel to spread out on the output histogram. Or are you talking about the input histogram?

I've been using the color management feature to get an accurate depiction of the orange mask with the added advantage of having the scanner write the data in a wide Target space like ProPhotoRGB. But with some scenes and film brands-Kodak UC 400-I end up with some shots having weird hue and saturation hot spots inverting and processing in PS using the NegPos technique.

Okay, sounds like you are using SilverFast? If not, then I am less familiar with the Epson software. Regardless, there is an input adjustment for capture. When this works nicht so gut is negatives that have a bias towards one or two ranges of hues. Without seeing your image, it is tough for me to give you specific recommendations, though I can give a few pointers. Consider that not every image will run the same adjustments.

So let us assume that the green channel shows near a full input histogram, while the red channel is fairly close to that, and the blue channel is only near half a full histogram. If you adjusted the blue channel for a full histogram, then you will get a colour shift in your image. You can try to adjust that in PhotoShop after inverting, though sometimes you might be giving yourself too much work. In that case, rather than maximizing per channel capture area, simply adjust the widest histogram channel (in our example that is green), and then only adjust the other channels a similar amount; doing this will maintain colour balance in your image.


Each channel histogram before setting B/W/neutrals never spans the entire 0-255 mapped graph once captured and viewed in PS. They're always narrow centered clumps within the graph. After applying these settings is when I get normal fully spanned histograms. Am I doing something wrong, because I've used almost every techique I've come across on the web including NegPos and I never can get fully spanned histograms on capture.



As for eliminating noise through exposure, I get film grain noise in shadows no matter how much I blast Kodak UC 400 negatives with adjusted scanner exposure. My understanding is, if it's in the film, no amount of scanner light can bleach it out. I believe it's the angle of slope chosen for the gamma type tonal correction that tends to exaggerate film grain noise much like what some have experienced processing RAW digicam files.

Only my observations. I'm new to scanning negatives and still trying to figure out a way for getting consistant results. So far it's been hit or miss and I can't seem to understand what influences such results.

When you have a light source scanner projecting that light towards sensors with colour filtering over the pixels, some ranges will do better with noise control, though mostly due to current and charge levels. Without getting overly technical (since there is little point in that), the green filtering passes more light through to the sensor than the red filtered areas, and the blue filtered areas suffer the worst on light transmission. If you want a simple way of envisioning this, image the effects while viewing through a lens filter of similar colour to what is over each row of pixels. Nikon scanners with LEDs work somewhat differently, in that the wavelength of light is affected by the colour range transmitted by the LED, though the charge levels on the imaging chip still function in a similar manner, and you can often see that the green channel results come out slightly better than the others.

This is tough to explain, and much easier to see in action. The heaviest emulsion deposited on the negative blocks more light than the film base, or the thinner areas. That heavy area on a negative represents the lighter areas of a picture. The thinner areas and film base are the darkest picture areas. Scanning as a positive inverts those capture values, though if you get it wrong then the highlight areas of your picture will not look correct. Then you could get a situation of great shadow details with blown highlights.

Think of dark areas of an image during digital capture. Anything near white gets a full charge at the pixel, while the darker areas get less charge. Less charge means very little information from which to put a value on a hue. This is more troublesome with Bayer pattern interpolation due to blue and red filtering passing less light than green area filtering, leading to noise with colour. A scanner mostly avoided odd colour noise, but still suffers that lack of information problem. Inverted scanning simply gives you more information on shadow areas.

Okay, so practical uses, or when to use . . . . . I do this with everything B/W that gets scanned, unless I am scanning on a CREO (better software). I rarely shoot colour negative films, though when I do and need to scan them, the darker images get the inverted scan, while the more even or daylight images get straight scans. You need to decide if you want to bias highlights or shadow details.

So I hope I made some sort of sense with all this. There is probably too much to type in a short reply, and too easy to miss many aspect. Plus my coffee is not working that well this morning.

Ciao!

Gordon Moat
A G Studio (http://www.allgstudio.com)

QT Luong
26-Sep-2006, 12:28
In my experience Noise Ninja does a better job on native digital images than on scans. I'd second the suggestion of Bruce to get a better scan. There is a big difference in noise levels between a drum scan and a desktop scan.

chris jordan
26-Sep-2006, 12:54
Kirk, here's a suggestion, which I do to all of my images to eliminate shadow noise. First, go into the shadows at 100%, and apply the dust & scratches filter. Use a radius of 1 and a threshold of about 25 or so. You might need another pass with radius of 2; that depends on the image. With a few tries, you should be able to completely eliminate the noise in the shadows, but the whole rest of the image will be messed up too.

Make a copy of the image, and paste it back as a layer on top of the old version. Now go into the advanced blending mode, and pull the sliders on the right (the highlights) down, so that the layer is only applied in the shadows. The sliders should be on something like 50 or 60. Using the option key you can split the sliders so that they fade in.

That way, the denoised layer only applies in the shadows. Now you can go back into the shadows and fade the layer's opacity to taste. I usually fade it partially back in, so there is a tiny amount of noise, so the shadows don't look too smooth.

Then, when you sharpen, you can do the same thing-- make a copy of the sharpened layer, put it on top of the non-sharpened layer, and use the advanced blending mode sliders to pull the sharpened layer out of the shadows. That way, the small amount of noise you leave in the shadows will not get sharpened.

Does that all make some sense?

Pete Watkins
26-Sep-2006, 15:02
I'm not digital orientated in any sense, BUT when I have taken pictures in this type of lighting for scanning I've used Tungsten balanced film. This image seems to be too cold in my thinking. Don't tear me apart as I really have no interest in digital apart from creating a website.

Kirk Gittings
26-Sep-2006, 19:02
Interesting. Some good ideas to pursue. Thanks guys.

Qt, I agree about a quality scan but this is a quick turn around job. No time or money for a drum scan.

Pete, tunsten film would be a bad choice for this image as most of the interior lights are are daylight flourescent or metal halide. Daylight film is much closer, tunsten film would have made it much bluer (though this is color negative and it does not matter much either way).

Tim Lookingbill
26-Sep-2006, 19:18
That's a good one chris. I forgot about advanced blending mode. That saves me from doing my clunky selection method.

There's always a better way of accomplishing something with Photoshop if I can only remember when and where to use what method.

chris jordan
26-Sep-2006, 23:10
Yeh, that advanced blending mode is an amazing tool, one of the most power in Photoshop I think. I use it in all kinds of ways to affect only parts of an image: taking noise out of shadows; taking color (especially blue casts) out of shadows; making sure my sharpening doesn't blow any highlights; taking sharpening out of the shadows; sharpening the midtones more than the highlights and shadows, etc.

QT Luong
27-Sep-2006, 01:21
Great tip ! Let see if I understand how to sharpen without blowing the highlights with this method. Make a duplicate layer. Sharpen the duplicate layer. Double-click on the duplicate layer icon. Move the highlight slider of the duplicate layer down just a little (maybe to 240 ?). Flatten. Correct ?