Does multi-scan really reduce noise?

[dh003i]

I'm scanning using Vuescan with the Epson V700. I scan at 6400 dpi and then have the scanning software downsample. I scanned a small test swap of a 4x5 file, comparing 3x- and 16x-multi-pass. It seems to make no difference. Nor does either offer an advantage over 1 pass (not shown). [all of these include multi-exposure too].

So it seems to me like multi-pass isn't doing anything. It is supposed to reduce noise. Thoughts?

[redu]

I think that the multipass thing must be something theoretical and is just waste of time for the results obtained (none noticed). For me the best noise reduction tool is Neat Image. I have tried both Noise Ninja, Kodak Digital Gem and Neat Image. Eventhough i am not happy with their default settings (too dramatic noise suppression yielding loss of information along with noise) with custom settings they shine. They all have different user interfaces and parameters to tweak but if you use them correctly they all perform their job fantastically. I am used to Neat Image's controls quite well so that's my choice.

[David de Gruyl]

I have only seen a reduction in quality resulting from multiple passes.

[8x10 user]

Multiscan technology should reduce scanner noise but will do nothing for film grain.

I use Noiseware for grain removal. I haven't tried neat image yet. I like Noiseware because it is so customizable.

[domaz]

I haven't noticed any difference with my Epson scanner. I did notice a difference with my Minolta and previous scanner. Perhaps the noise the Epson generates is so consistent/non-random Vuescan can't determine what's noise and what's not.

[Juergen Sattler]

I use Kodak's Digital GEM for grain reduction/removal. It is so far the best software I have found for grain. Most of the other noise reduction programs just smear the grain around and result in a less than sharp image.

[rdenney]

I haven't noticed any difference with my Epson scanner. I did notice a difference with my Minolta and previous scanner. Perhaps the noise the Epson generates is so consistent/non-random Vuescan can't determine what's noise and what's not.

The idea behind multiscanning is that it takes several samples of the same spot. When the scanned material is thick, the signal is close to the noise floor of the CCD, and so the noise, which is always random in nature, has a greater influence on the resulting pixel. By sampling it several times and taking an average, that noise is, to some extent, filtered out.

But I would only expect an improvement in highlight areas of negatives or shadow areas of transparencies, where the light reaching the sensor was dim enough to be close to the sensor's noise floor. It should never affect the detail in the scan (but see below).

Noise removal software, on the other hand, looks for random color variations at the pixel level after the image is already scanned, and attempts to smooth those by averaging them with their neighbors by some algorithm or other. Determining what is random color variation caused by noise and what is information requires a fine touch, of course.

The problem with multiscanning is mostly mechanical. On consumer flatbeds such as an Epson, the CCD is squirming around between the samples, and you get blur just as you would if you thumped your camera between multiple exposures. The mechanicals have to be precise enough so that all samples are precisely aligned with each other. The greater the resolution, the harder this is to do. I suspect it is the mechanicals of the Epson that limit it to an effective maximum resolution of about 2000-2400 spi. Therefore, at optical scan resolutions higher than that, I would expect multiscanning to cause blur from movement between samples, when viewed at the pixel level before downsampling.

Scanning at 6400 and then downsampling is a different strategy for dealing with noise, and works within the mechanical limitations.

Rick "who has seen some improvement in highlights when multiscanning overexposed color negatives, but only with film scanners" Denney

[8x10 user]

Film grain is an issue will all high resolution scans. I don't have any experience with Epson scanners but the best CCD scanners do have a multiscan type option to improve scanner noise. In addition to Viewscan you might want to check out SilverFast. I could not tell you which one is better.

I'd like to try out the Gem tool from kodak. I believe it is said to be good for volume orientated workflows no? I'm thinking it would be good for more consumer level slide digitizing in medium to high volumes.

For critical work Noiseware isn't bad about sharpness loss depending on what settings you use. Sometimes it can miss a bit of grain here or there and you have to do a little manual retouching or it looks unnatural. You can control the percentage of noise reduction in many criteria so you can still have a some grain in the image if you want. It just doesn't need to be empathized as much as it is in a high resolution scan. You can use multiple layers with different settings and use masking to selectively apply some of an extra smooth version on your big low details areas (like blue sky's). I tend to set the opacity of the layer to medium low so that there is some grain left in the image to make the it look more natural. Also it is good to use a fathered layer or to paint it on with a soft brush tool.

[dh003i]

Ok, I would note that I'm not concerned with "grain reduction". In fact, I think that for prints 50x40, the grain you'd see from a 4x5 wouldn't be a big deal -- at least not for my E100VS Kodak film.

The Epson V700 is absolutely not scanning down to the level of film grain. I wish it were. It seems to scan at a level of detail about an order of magnitude less than that. I.e., if film grain were like pixels, the V700 would be getting to the level of a 4-pixel square. Maybe you get the vague perception of some grain, but you aren't actually seeing film grain. No way. (at least not with fine-grained film, I don't know about larger grained film). I saw on test on some older film showing the V700 resolving down to the level of film grain, but I suspect that was very large grain. This test shows what I said (http://www.flickr.com/groups/466058@N21/discuss/72157622922559537/) -- it isn't getting film grain.

What I was concerned with was noise in the shadows and shadow areas. This is a scan of a bridge/waterfall (http://www.tabblo.com/studio/stories/view/1791520/) that I did, with crops at 50% zoom; there doesn't appear to be any noise in the shadows. (nor is there at 100%). This is the comparison of 3x and 16x multi-scanning (http://www.tabblo.com/studio/stories/shared/34702/2eczmd6gulpsfw4), and it includes shadow areas. I see no difference, except maybe a little less sharp on the 16x multi-scanning. What about you guys?

[Richard M. Coda]

I find it humorous that people using 8x10 cameras are concerned with grain :)

[dh003i]

Richard, I'd think they'd be more concerned with wanting to resolve down to the level of film grain ;-)

[urs0polar]

Are we talking about multi-sampling or multi-passing? I thought that in multi-sampling, the CCD doesn't move but makes multiple exposures and the results are averaged to get rid of noise. In multi-passing, the CCD goes over the whole piece of film multiple times to try and extend the dynamic range. This can obviously make things blurry as it's hard to line up exactly on subsequent passes.

I think it also depends on the scanner from what I recall of the vuescan manual online. Some scanners can't multi-sample in place; they can only make passes, while others can multi-sample in place.

If the CCD is not moving but merely firing 4x or 16x or whatever, how would that make the resulting pixel blurry?

Am I totally off base here? (it won't be the first nor the last time...)

-Mark

[dh003i]

Are we talking about multi-sampling or multi-passing? I thought that in multi-sampling, the CCD doesn't move but makes multiple exposures and the results are averaged to get rid of noise. In multi-passing, the CCD goes over the whole piece of film multiple times to try and extend the dynamic range. This can obviously make things blurry as it's hard to line up exactly on subsequent passes.

I think it also depends on the scanner from what I recall of the vuescan manual online. Some scanners can't multi-sample in place; they can only make passes, while others can multi-sample in place.

If the CCD is not moving but merely firing 4x or 16x or whatever, how would that make the resulting pixel blurry?

Am I totally off base here? (it won't be the first nor the last time...)

-Mark

Mark, I'm not sure, but I think my Epson V700 does multi-sampling, where the CCD stays in place. See the example of 3x vs. 16x here (http://www.tabblo.com/studio/stories/view/1791834/). These were scanned at 6400 dpi, and are 100% crops. It seems to me there is absolutely no difference in the level of shadow noise, so the multi-sampling is hence useless.

In those two, I'm strained to see any difference in sharpness, but I would suspect that multi-sampling can reduce sharpness by exposing the film to the lamp for much longer periods of time, hence possibly causing warping.

[urs0polar]

Mark, I'm not sure, but I think my Epson V700 does multi-sampling, where the CCD stays in place. See the example of 3x vs. 16x here (http://www.tabblo.com/studio/stories/view/1791834/). These were scanned at 6400 dpi, and are 100% crops. It seems to me there is absolutely no difference in the level of shadow noise, so the multi-sampling is hence useless.

In those two, I'm strained to see any difference in sharpness, but I would suspect that multi-sampling can reduce sharpness by exposing the film to the lamp for much longer periods of time, hence possibly causing warping.

yeah that makes sense. I also have a V700, and I've used the multi-pass for more dynamic range, but I kind of agree with you about the multi-sampling not making any difference from what I remember from last time I did a comparison. I leave it at 4x anyhow for psychological reasons :)

As for blurriness, I guess sharpness could be lost from heat expansion / warping, I suppose. Scanning is crazy; you're damned if you do and damned if you don't.

On my Nikon 9000, however, the multi-sampling does seem to help out (at 4x vs 1x), especially with things like Velvia 50 blacks. I can't tell 4x vs 16x but sometimes I do 16x anyhow if it's going to be printed big. I don't have an example right now as I did the compare when I first got the scanner then have been either 4x or 16x ever since. I haven't seen a loss of sharpness at 16x on the 9000, for what it's worth.

[8x10 user]

I shoot 8x10 but I'm thinking about switching to 5x7 for portable applications.

I scan whatever people give me... I just did a 12,000 DPI scan from a Velvia 50 35mm slide. Oh yes, there was grain, but after digitally processing the image you would think it was medium format if you didn't know better.

Cheers,
Ed


I find it humorous that people using 8x10 cameras are concerned with grain :)