Digital grain reduction for film--"GEM", etc versus Canon 1Ds?

[alex___4]

There's been something of an uproar about the new Canon EOS-1Ds

digital camera, precipitated by Luminous Landscape web site where in

addition to outright blowing away film (Provia 100F in this test),

the tester also claimed and demostrated initially that it edged out

Provia 100F shot with a Pentax 645 as well. The later statement was

later retracted and the superior sharpness of the Canon 1Ds picture

over the the 645 scan was attributed to sharpening artifacts of some

sort. The thread in question is:

 

http://www.photo.net/bboard/q-and-a-post-new?topic_id=35

 

While the Luminous Landscape test report of the Canon EOS1Ds is an

excellent report, I feel the major factor of software grain reduction

(for both imaging sensor and film) has been overlooked. To begin

with, exactly how much of the noise reduction in the new EOS 1D is

due to better hardware (less intrinsically noisy imaging chip) and

how much is due to improved software (grain detection and reduction

by software). Of course, this is a question for the original Canon

engineers of the camera, as any tester/user can only view the final

image, since it's likely a fully closed system. However, we might be

able to infer improvement in software, if the new Canon G3 compact

digital zoom software offers such over it's predecessor the G2--

seeing how Canon did not increase it's pixel density (this is the

first time, a later generation product has not increased pixel

density, indicating that Canon things that 4 megapixels more than

suffices for the typical consumer).

 

And then, of course is the fact that NO grain reduction processing

was applied to the film's (35mm or 645) digital output. Despite being

on the look out for it, I have yet to see ANY analysis/review of the

Applied Science Fiction's "GEM" (Grain Equalization Management)

offered as yet, only on the top Nikon and Minolta scanners. So I

can't begin to say how much more improvement there might have been,

had "GEM" been used. Moreover, this is the first such

product/software that I know of for film, while EVERY digital camera

released nowadays, has it built in! Clearly, even Fuji and Kodak (let

alone Canon and Nikon), are not concerning themselves much with

possible digital improvements to film--given the steep decline

everyone is forecasting for it. For those of us with immense backlogs

of film and/or still shooting film, hopefully the motion picture

industry will come to our rescue as they may need to heavily

reprocess their old backlog for HDTV.

 

The previous 2 paragraphs I posted in the initial EOS1Ds vs 100RDF

thread, but no one noticed or replied to it's specifics. So I ask the

question here--has anyone conducted an evaluation of ASF's "GEM"

product or something similar? If so, can you please comment?

 

 

Alex

[marbing]

If they said on Luminous-Landscape that it was dark at night and bright during the day...I would ask for a second opinion.

[jay_.]

I think it's probably pointless to belabor the 1Ds vs MF issue until enough reliable people have their hands on production 1Ds's.

[ilkka_nissila]

I think the two cases are a little different. Noise reduction in a digital camera is concentrated on improving the appearance of shadows, where the noise is most visible. This is analogous to an underexposed color negative, which prints with muddy and grainy shadows. If these areas are enhanced, information is obviously not increased (unless it is done by increasing the exposure time or averaging multiple exposures), but the shadows appear smoother.

 

On the other hand, grain is the very fabric of a film-based image. If you remove it, you loose the image. So, by necessity a general-purpose automatic grain-remover causes the softening of detail in the image. Obviously, the situation is not hopeless. By selecting certain areas of the image with the magic wand, you can filter out noise in certain critical areas such as the sky and in faces. This may be a lot of work though.

[gannet___]

I'm not following this. Digital captures don't have grain, and digicams don't have grain reduction software. They don't need it. They do have noise reduction software, which is an entirely different and unrelated thing.

 

GEM, like ICE, is reported to have a noticeable negative impact on resolution.

 

Why in the heck would any of the major players be worrying about "digital improvements to film"? Film will always look like film, even when scanned. That is neither good nor bad, it just is. Improvements to digital are being driven by the need to improve (and sell) the new, not the old. The people who can invest the kind of money new digital equipment costs mostly aren't all that concerned about marrying it to their film archives, or about shooting more film with it in future. Just imo.

[alex___4]

>>I'm not following this. Digital captures don't have grain, and digicams don't have grain reduction software. They don't need it. They do have noise reduction software, which is an entirely different and unrelated thing.

 

GEM, like ICE, is reported to have a noticeable negative impact on resolution.<<

 

I disagree Gannet, that they digital noise and film grain are entirely different! Visually they are similar in effect and I strongly suspect that once a film image is digitized, the residual grain can be substantially reduced.

 

As for GEM, I have heard that it is very film specific (works well with only certain/selected films). Moreover, it's brand new technology and may improve substantially in the coming months/years--I hope so in any event.

 

 

Ilkka please see the following for an explanation of GEM:

 

http://www.asf.com/products/gem/

 

Nonetheless, to restate my question, how much noise reduction was done in camera with the EOS 1Ds (probably unanswerable at the moment) and how much "grain reduction" is possible via software of film images?

 

Alex

[bill_k1]

>> I disagree Gannet, that they digital noise and film grain are entirely different! Visually they are similar in effect and I strongly suspect that once a film image is digitized, the residual grain can be substantially reduced. <<

 

Is digital noise and film grain really visually similar to you? I do not see it this way at all. Grain is much more acceptable to me than noise and sometimes even is desirable in adding a feeling to the picture. Noise is never something, when noticeable, that I would want in an image.

[pete_andrews]

<i>"I strongly suspect that once a film image is digitized, the residual grain can be substantially reduced."</i> - You suspect wrong!<p>There is a phenomenon called 'aliasing', which is basically an interference effect between the fixed sampling frequency of digitisation, and the pseudo-random grain pattern. (I say pseudo-random, because grain is really closer to pink noise, in that it occurs in a limited bandwidth.)<p>Anyway. This aliasing cannot easily be undone after the digitising stage. ASF's GEM appears to me to simply be a sophisticated low-pass filter algorithm which removes image detail as well as grain.<br>To properly combat the effect of grain, the filtering would have to be physical, such as throwing the film slightly out-of-focus, or reducing the resolution of the scanning optics.<p>As Gannet and Bill have said. Grain and noise are two different things. Electrical noise is a good fit to being truly random, and therefore lends itself to statistical or additive cancellation. Film grain isn't truly random, it's engineered to be within certain parameters, and doesn't lend itself to statistical methods of reduction.

[gary voth]

GEM and ICE, at least as implemented on my LS-8000, are both pretty benign when it comes to affecting resolution. First thing I did when setting up the scanner was to do a batch of 16x 14-bit multipass scans straight and with ICE and GEM at various settings. There was a *very slight* softening of fine detail with GEM at its highest setting, but this was far more subtle than net gain from the grain reduction, and edge acutance can easily be enhanced with modest USM. I always use GEM (at 3) for images which can benefit from this.

 

That being said, the scan process is slowed dramatically by all the extra processing, and hardly rivals digital capture from a time efficiency standpoint...

 

As to the larger topic, I think digital is here now. Whether or not the 1Ds really does exceed scanned MF is not really the point, because surely digital will do this sooner or later. I say if you still prefer to shoot film, then go for it. It won't go away anytime soon. (But it will be harder to argue that film is superior for very much longer.)

[cliff_lesergent]

All I can say about the digital vs. film comparison at <i>The Luminous Landscape</i>, is that when <u>my</u> Pentax 645 lenses start giving me results as poor as those shown in the comparisons, I'll throw them away! The comparisons on this site are not about the absolute quality of digital vs. film, but rather digital vs. as much effort as the reviewer feels like putting into scanning film on a consumer-grade desktop scanner.<p>

 

I used to feel that Michael's reviews were quite objective, but lately there seems to be a lot more "hype" involved - I guess if reviewers don't hype the products, the manufacturers stopping giving them samples to play with.<p>

 

With respect to GEM, in this month's issue of <i>Photo Techniques</i> magazine, Ctein reviews the Minolta Dimage Scan MultiPro film scanner. In his review, he noted in test scans <i>"how effective GEM is at smoothing out the grain without blurring any of the fine detail"</i>.<p>

 

Although I've never gone to the expense of having drum scans made, it is my understanding that the effect of scanning through the mounting fluid is to greatly reduce the appearance of dust, scratches, and to a lesser extent, film grain. Can anyone with drum scanning experience educate me on this?

[alex___4]

In the July/August 2002 issue of Photo Techniques USA, there is an excellent article by John Paul Caponigro titled "Art of Photoshop: Noise/Grain" starting on page 49. In it, the author discusses the removing and/or adding of grain into an image via Photoshop in some detail. One of the things he mentions is a PS plugin called "Grain Surgery" (www.visinf.com). Consider it a must read, if this subject interest you.

 

I will quote a single line from his article:

 

"While digital capture doesn't have grain per se, it introduces noise into images that is quite similar to grain in appearance, although not identical".

 

Thanks to those who contributed.

 

Alex

[bobatkins]

"<em>Film grain isn't truly random, it's engineered to be within certain parameters, and doesn't lend itself to statistical

methods of reduction</em>"

<p>

If grain isn't random I wonder whether the grain in different films might have a predicatble frequency spectrum? If so, and if the spectrum (or the peaks in the spectrum) were sharp enough maybe some sort of fourier tansform and filtering could knock down grain without too much of a toll on overall sharpness? In this way it might actually be easier to remove than if it were random!