Jump to content

How to reduce film grain


Marisa Aurora V

Recommended Posts

Film has grain, but for reasons not well understood by film to digital systems, the transfer to digital actually increases the grain. This can be demonstrated by comparing, by eye, a film projected on a wall side by side with it's digital transfer. The grain is enhanced in the transfer. Where does it come from?

 

The reason is a little complex but it has nothing to do with film or the digital sensor but rather in the relationship between the two: an "interference" (for want of a better term) between the way film encodes an image and the way digital encodes an image. The digital camera has a cutoff frequency whereas film doesn't. The high frequencies in film tail off into "noise" (or grain) but this "noise" is statistically correlated with the original signal (the original image) which the digital sensor can't see but is there in the film. There is an necessary exchange that occurs at this boundary where the information still in the film gets traded for a corresponding increase in noise in the digital copy.

 

Fortunately there is a simple remedy that future datacine's will no doubt possess. One will be to increase the defintion of the capture sensor (4K or more), and/or the simpler remedy is to mutli-sample the film at a number of offsets (that are decorrellated from the pixel pitch of the sensor), and to re-register the captures, and then integrate.

 

Film should not look as bad as it does in a transfer. It should look as good as it does when projected. This is not yet the case because the problem remains not well understood

 

Denoising algorithms are not the real solution. They might decrease the noise but they also decrease the resolution of the image. The above method will properly minimise any increase in noise (the more samples the better) while maintaining the resolution of the result at the same resolution of the capture device (in fact it actually increases the resolution of the result from that which the sensor would otherwise suggest).

 

C

  • Upvote 1
Link to comment
Share on other sites

Carl,

 

I believe the scanity (and possibly Northlight 2) come close to what you are talking about. Sadly, they do not have a Super 8 gate. :(

 

For real good Super 8, you've basically got 3 options:

 

1.) The Spirit at Spectra... but it has it's draw backs where the spirit can't handle jitter, etc.

 

2.) The Millenium at Pro8mm, but they sometimes are hit or miss... it seems to be more the colorist than then machine.

 

3.) The LaserGraphics ScanStation... but no one really has one yet. Hopefully here in Boston within a few weeks at Gamma-Ray Digital.

 

There may be two other good machines, but I have no personal experience with them. The Goldeneye III can do Super 8 as well as the MWA Nova Flashscan 2K+. I'm still looking for a good opportunity to test these machines.

Link to comment
Share on other sites

Yes, it is still very much early days in the art of film to digital transfers. High end systems will no doubt have put more thought into the problem, if somewhat neglecting the small gauge medium. They have perhaps, fallen into the erroneous idea that smaller film requires less pixels. Indeed the reverse is true. The smaller the film the more grain it exhibits and the more pixels per image area and/or more multi-sampling you would need to stop such from being amplified.

 

Nevertheless one can come to an understanding of the problem even if there are not yet the best solutions for the Super8 filmmaker.

 

There are, however, DIY solutions, that can benefit from this information and be factored into new DIY solutions.

 

The most effective DIY system, I'd recommend, will be one in which the film is transported continuously in front of a capture device, and that uses an RGB LED flash to capture the film. But instead of just one capture per film frame one does many - the more the merrier. Even just four captures per frame will provide a significant difference. Depending on your skills this may or may not be the easy part. The second part may be the more difficult part. It involves digitally re-registering all of the captures and stitching them together.

 

I'd eventually like to put such a DIY system together myself. For the time being I'm relying on a DIY single capture per film frame, at 5K resolution, even though I know (from theory and single frame testing) that an even better result would occur if I were to re-engineer it for multi-sampling.

 

But for me that will just have to wait for another day.

 

Carl

Edited by Carl Looper
Link to comment
Share on other sites

  • Site Sponsor

There is the Kinetta scanner which is now 3400x2400 for Super-8mm full frame color CCD arguably better than any of the other solutions Image Quality wise.

 

http://www.creativeplanetnetwork.com/dv/feature/airs-cnn-august-our-nixon/63162

 

 

Cinelab's Xena servo scan with realtime OpenGL perf stabilization is currently scanning in Alpha out in LA and we should have that machine running in Dec-Jan at Cine it has a similar resolution to the Kinetta but slightly less because of the area used for machine vision and stabilization.

 

-Rob-

  • Upvote 1
Link to comment
Share on other sites

  • Premium Member

Carl,

 

Are you saying capture separate Red, Blue and Green (plus possibly an IR) scans of the one film frame and combine them? This is something I would like to do with a DIY small format scanner using a mono CCD camera that can do 12 or more bits of depth but I have been warned that residual vibration from advancing a film frame would make it very difficult.

Link to comment
Share on other sites

  • Site Sponsor

Our current Xena system has a 4K monochrome area sensor and the transport is registered by Oxberry pins for 35mm and 16mm so when capture is taken the film is stationary.

 

The scanner has a R,G,B and IR LED lamp house with the latest surface mount LEDs and it fires each color sequentially to capture color. The engineers at DCS just sent us a software upgrade which allows for HDR so the LED can be set to flash twice for each color and you can control the blend of shadow to hilite flash. The pin registered machine weighs about 900lb.

 

There are a few factors that make great scans which resolve grain more finely:

 

1. Oversampling your target resolution i.e Nyquist so for example when we scan S-16mm for 2K the film is scanned at 3.2K resolution and down sampled.

2. Monochrome camera with individual color layer/channel capture.

3. Mechanically robust transport.

Link to comment
Share on other sites

There is the Kinetta scanner which is now 3400x2400 for Super-8mm full frame color CCD arguably better than any of the other solutions Image Quality wise.

 

http://www.creativeplanetnetwork.com/dv/feature/airs-cnn-august-our-nixon/63162

 

 

Cinelab's Xena servo scan with realtime OpenGL perf stabilization is currently scanning in Alpha out in LA and we should have that machine running in Dec-Jan at Cine it has a similar resolution to the Kinetta but slightly less because of the area used for machine vision and stabilization.

 

-Rob-

 

 

I saw the Kinetta transferred Super 8 footage on the Nixon show that aired on CNN. The footage looks great !

 

I caught a few glimpes of what looks to be Canon 814 or 1014 camera being used by the Nixon aides.

 

Canon814_1014.jpg

 

 

http://vimeo.com/45079849

 

.

Edited by David Nethery
Link to comment
Share on other sites

Carl,

 

Are you saying capture separate Red, Blue and Green (plus possibly an IR) scans of the one film frame and combine them? This is something I would like to do with a DIY small format scanner using a mono CCD camera that can do 12 or more bits of depth but I have been warned that residual vibration from advancing a film frame would make it very difficult.

 

 

That wasn't what I was getting at but that is certainly another improvement - anything which increases the transfer of information rather than blocks it, is a good thing. An advantage with a mono camera (as distinct from a bayer filter camera) is that you will get a much better colour signal, because you are doing the colour filtration yourself separately, rather than interlaced with every second or third pixel (so to speak).

 

But what I was specifically talking about is the creation of noise that occurs in the transfer - ie. noise that is in addition to that already understood as in the original film.

 

But insofar as electonic sensors have been the traditional means by which noise has been quantified in the first place, any noise actually created in a transfer (in a measurement) can be missed. There isn't anything to compare it with other than one's eyes. So for all those people who have insisted Super8 has always looked better projected than in a transfer, they have been right. There just wasn't anyway of measuring it because the very measurement tools, the electronic sensors, have been a contributing factor in the creaation of noise.

 

The sensor itself is not the problem. The most expensive, cleanest, noise free sensor in the world can be used and it will still happen.

 

The only solution, as we're discovering, is to increase the resolution of the sensor and/or perform multi-sampling, ie. over and above the delivery definition one is aiming at. The reasons put forward against this have been due to an erroneous understanding of the relationship between film and digital. Many will say that, for example, the resolution of film is X pixels/mm and so that scanning any higher than this is redundant. But if scanning higher than this improves the image (but not necessarily the resolution) one has to ask what else it is improving. What eles it improves is the noise level. The nosie level is reduced.

 

But why is it reduced? This is not some magical thing. It's not as if a higher rez or muti-sampling approach reduces the noise already in the film. It reduces the amount of noise being created during the act of digitally measuring the film in the first place!

 

C

Edited by Carl Looper
Link to comment
Share on other sites

Also it's important to know that this isn't really anything to do with Nyquist limits. The Nyquist theory argues correctly that you should sample at twice the frequency of your target defintion. But this in realtion to a an ideal noiseless signal. In relation to a noisy signal you will want to sample beyond this limit, as far as you can. How far is just a question of practical and economic limits rather than any theoretical limit. The only reason it's hard to pin down a limit is that there is no such thing as noise in mathematics. You have to use statistics to get some sort of idea of it and even then it's more art than science.

 

C

Link to comment
Share on other sites

  • Premium Member

Our current Xena system has a 4K monochrome area sensor and the transport is registered by Oxberry pins for 35mm and 16mm so when capture is taken the film is stationary.

 

....... The pin registered machine weighs about 900lb.

 

..........

3. Mechanically robust transport.

Yeah, thought so. Not sure how I can do a DIY pin registered system but was thinking of screwing the whole system down to something like a cement or marble top including the camera and lens (plus fixing that to the projector) and then looking at trageting damping vibration close to where it it affects things.

 

Another thing would to not transfer in real time but at a rate perhaps less than ten frames a second.

My idea would

Link to comment
Share on other sites

Yeah, thought so. Not sure how I can do a DIY pin registered system but was thinking of screwing the whole system down to something like a cement or marble top including the camera and lens (plus fixing that to the projector) and then looking at trageting damping vibration close to where it it affects things.

 

Another thing would to not transfer in real time but at a rate perhaps less than ten frames a second.

My idea would

 

A DIY alternative to pin registration is to use open source computer vision algorithms, such as OpenCV. They provide a way to register individual captures, not just one film frame to another (in time), but multi-sample captures of the same frame (in space) to each other. The benefit of adopting such an approach is that the mechanical side of the DIY system can then become a little more relaxed, ie. cheaper.

 

C

Edited by Carl Looper
Link to comment
Share on other sites

You are lucky. Here in London I haven't figured out where to buy 50D on super 8 though a mate has used it.

 

You would think Kodak would try and get as much 50D out there as possible or am I just being naive.

Try The Widescreen Center, in London

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...