Can Super 8 look as good as 35mm?

[Will Montgomery]

...posts from John Pytlak. I still miss his posts on film topics.

It was great having John on the forum. He was always kind, enthusiastic and knew what he was talking about.

 

I've often wondered why Kodak doesn't encourage more of their engineers and scientists to troll these boards and help out when ridiculous answers are given.

[Chris Burke]

It was great having John on the forum. He was always kind, enthusiastic and knew what he was talking about.

 

I've often wondered why Kodak doesn't encourage more of their engineers and scientists to troll these boards and help out when ridiculous answers are given.

 

 

they probably do. they still keep the super 8 format alive and it is forums like this that are part of the proof.

 

I think that in the age of YouTube, a feature shot entirely on super 8 is completely valid. Why not. for home video and the web, again why not?

[Carl Looper]

My answer of a simple "yes" was, of course, provocative, but not without reason.

 

Lets assume. for the sake of the argument, you could process Super 8 film in such a way that could reconstruct a signal with the same technical quality as that of 35mm. Well then you could always apply that same process (whatever it is) to 35mm to make it look better than the processed Super 8.

 

BUT 35mm film is typically scanned at 2K. The 2K scan acts as a low pass filter, or barrier to any further improvement of the scanned 35mm signal. This barrier is purely historical (arbitrary) but it helps to interpret the question in a fairer way - where the goal post is the same:

 

Can Super 8 look as good as 35mm film scanned at 2K?

 

The reason 35mm film is scanned at 2K (rather than higher) is because a. it is cheaper than doing it a higher defintion, and b. it is generally felt that scanning it at any higher definition won't achive much. If you scan 35mm film at a higher definition (4K, 8K etc) but don't do anything to improve the signal in that domain then you won't achieve much by having scanned the signal at that higher definition other than to enhance the noise.

 

The same goes for Super 8. If you scan Super 8 at a higher definition, but don't do anything to improve the scanned signal, then you likewise do no more than enhance the noise.

 

The difference is that Super8 scanned at 2K has plenty of room for improvement, whereas 35mm does not. Like the loser of last year's 100 yard dash, the loser has more chance of doing a whole lot better in next years race, than the winner.

 

The question of whether Super 8 can look as technicaly good as 35mm can be interpted in the same vein. What can be done to improve a Super8 signal scanned at 2K.

 

Well there is quite a lot.

Edited October 15, 2010 by Carl Looper

[Nicholas Rapak]

This question is extremely ambiguous. If we were watching VHS quality scans of 35mm and Super 8 on a 12 inch TV screen, it would be difficult to tell the difference. However, if you were to get 4K scans of both and project them onto a movie screen, the difference would be mind-blowing.

 

All film has a certain amount of silver that can capture a certain amount of information. 35mm has more silver in it than Super 8, so it can capture more information. In addition, the factor by which an image is enlarged is very important. 35mm academy is a 13x larger frame than Super 8. When you are enlarging an image 500x or more, that 13x size difference will make a tremendous difference in apparent resolution

[Carl Looper]

Continued.

 

When a motion picture is released in digital format the actual data is in a so called "compressed" format such as MPEG4. But what makes compression possible in the first place is not removal of useful information (which would degrade the signal) but removal of redundant information, which would not. For example, noise is redundant information. We don't need to store it. If we want noise (for a grainy look!) we can just synthesise it at runtime. And from one frame to another there is a lot of information that undergoes gradual transformation (rather than abrupt ones) that can be stored in a more efficient manner. For example MPEG4 uses motion vectors in order to evolve optical patterns over time.

 

The result of so called compression is a smaller file size.

 

But it's not really compression as such. The data storage medium we call film is simply wasteful in the way it stores information. The compressor is rearranging the information in such a way that cuts down on the waste. But the best way to do this is not a given. it is an ongoing research project.

 

The ongoing task is improvement of bandwidth (smaller file sizes) without degradation of the signal.

 

Now Super 8 can be interpreted as already in a small file size ("compressed"), but the signal is degraded. So the task here is to;

 

a. decompresss the signal

b. improve the signal

 

The first part is easy. The Super8 is scanned at a higher definition. However if nothing else is done this will simply reconstruct that which will be removed during subsequent recompression. The task is to rearaange the information in such a way that won't be removed during subsequent recompression. The task is step 2 - to improve the signal.

[Carl Looper]

This question is extremely ambiguous. If we were watching VHS quality scans of 35mm and Super 8 on a 12 inch TV screen, it would be difficult to tell the difference. However, if you were to get 4K scans of both and project them onto a movie screen, the difference would be mind-blowing.

 

All film has a certain amount of silver that can capture a certain amount of information. 35mm has more silver in it than Super 8, so it can capture more information. In addition, the factor by which an image is enlarged is very important. 35mm academy is a 13x larger frame than Super 8. When you are enlarging an image 500x or more, that 13x size difference will make a tremendous difference in apparent resolution

 

If you do nothing to improve a Super 8 signal scanned at 4K, the difference won't be mind-blowing. The difference will be what you would expect. The 35mm signal would look better than the Super8 one. However the point is being missed. There is a lot that can be done to improve the Super 8 signal but not much that be done to improve the 35mm.

 

The difference between 35mm academy and Super 8 is, as you say, 13X, but in the area of signal enhancement it is the square-root of this value that is more often used to describe the task. The Super8 signal needs to be improved by 3.6X (in width and height) to achieve the target definition.

 

Here is an example of 2X improvement:

 

http://en.wikipedia.org/wiki/File:Super-resolution_example_closeup.png

 

And the accompanying article:

 

http://en.wikipedia.org/wiki/Super_resolution

 

Note that the original question is not: DOES Super 8 look as good as 35mm, but CAN Super8 look as good as 35mm.

[Carl Looper]

I should also add that if the goal post is moved to 4K the task is a lot harder because at 4K there is more room to improve the 35mm signal (2X that of 2K), so the required enhancement of the Super8 signal would also have to go up by 2 (from 3.6X to 7.8X).

 

But that is moving the goal post.

 

A 2K reference is justified by contemporary practice and the idea that 35mm scanning might migrate to 4K in order to combat experiments in Super 8 signal processing is very funny.

 

Carl

[Carl Looper]

As mentioned in previous posts the required improvement factor for Super 8, to bump it up to that of 35mm, both scanned at 2K, is 3.6X.

 

Here is a description of obtaining 4X (apparently) using currently available software:

 

http://www.infognition.com/articles/what_is_super_resolution.html

 

The most common use for this software (and the papers that have motivated it) is in the area of enhancing digital video - but digital video is more diifcult to enhance than well scanned film. With the same algorithms, and well scanned film, one can improve a film signal more than the equivalent video signal. This is due to the random distribution/size of particles/dye clouds in film.

 

As algorithms improve the situation will, of course, get better.

 

Carl

[Nicholas Kovats]

Carl.

 

Full steam ahead. This is fascinating stuff. I must say that reviving 4-6 year old threads

is quite the entrance to this concept of "Super 8 signal processing". Wow. Unification and aggregation of terminology. Love this.

 

Have you examined Avisynth? Would you say this is a variation of Super Resolution

methodology? Do you have access to 2k Super 8 scans? Where are going with this? Alpha software?

 

Let's not forget that actual motion (film transport) is also part of the perceived experience of resolution. It isn't enough to compare a static noiseless HDSLR photograph to a typical "untreated" film frame.

 

Your reference affirms the "superior" information available in the fundamental random pattern of silver particles, i.e.

http://www.infognition.com/articles/what_is_super_resolution.html

 

"Super-resolution (SR) works effectively when several low resolution images contain slightly

different perspectives of the same object. Then total information about the object exceeds

information from any single frame.

 

...which is distinct from film grain as per your excellent reference doc on the revived thread "True Resolution", i.e.

http://cool.conservation-us.org/coolaic/sg/emg/library/pdf/vitale/2006-03-vitale-filmgrain_resolution.pdf

 

That "film grain is a perceived property; due to clumping of these fundamental silver particles" and that the "resolution of film is related to the size and distribution of fundamental particles in the emulsion."

[kevin jackman]

super8 can look as good as 16mm from about ten years ago. just make sure you use a really really good camera and the slowest speed film you can get. if you have interchangable lenses as an option then male sure you use the best lenses you can get. 16mm was a consumer format but the grain technology caught up. the same has happened for super8. just remember that the sharpness will be another issue but you can make it look great.

[Chris Fernando]

I was wondering. Is it possible?

 

Sure it is...(channel radio advert pitchman voice here), on the new Samsung Galaxy S, with it's massive 4.3" widescreen display!

 

Keep in mind; Lawrence of Arabia, conversely, will look like it was shot on 284.375mm, though :-(

[Carl Looper]

Have you examined Avisynth? Would you say this is a variation of Super Resolution

methodology? Do you have access to 2k Super 8 scans? Where are going with this? Alpha software?

 

I've used Avisynth (and VirtualDub). These contain numerous plugins for enhancing signals in the time domain and are an excellent benchmark. Indeed I'm using such as benchmarks for my own software. Regarding scanning I've got an experimental 4K scan setup (using a Canon 10MP with bellows and microspcope objective) which yields very good results.

 

Let's not forget that actual motion (film transport) is also part of the perceived experience of resolution. It isn't enough to compare a static noiseless HDSLR photograph to a typical "untreated" film frame.

 

Ah yes - the pressure plate on Super8 isn't as good as it could be - causing possible loss of definition along the vertical axis. However a very good signal can be reconstructed from motion blur (blur along a single axis) using an appropriately calibrated weiner filter. Full circle blur is a lot harder to filter.

 

Your reference affirms the "superior" information available in the fundamental random pattern of silver particles, i.e.

http://www.infognition.com/articles/what_is_super_resolution.html

 

"Super-resolution (SR) works effectively when several low resolution images contain slightly

different perspectives of the same object. Then total information about the object exceeds

information from any single frame.

 

...which is distinct from film grain as per your excellent reference doc on the revived thread "True Resolution", i.e.

http://cool.conservation-us.org/coolaic/sg/emg/library/pdf/vitale/2006-03-vitale-filmgrain_resolution.pdf

 

That "film grain is a perceived property; due to clumping of these fundamental silver particles" and that the "resolution of film is related to the size and distribution of fundamental particles in the emulsion."

 

Yes that's right. Film grain is so much easier to process than the fixed frequency of video pixels.

 

One problem I haven't mentioned is the lens. Unlike film grain the lens is not in a random location from one frame to another so the limits imposed by the lens are more strict than that of the limits imposed by the film grain. If a signal is moving with respect to the lens then the limits of the lens can be pushed back, but for a locked off shot on a non moving signal, the same rated lens (in MTF lp/mm) will introduce a lower defintion signal on Super 8 than it does on 35mm because there there are less millimeters available.

 

Or to put it another way, a lens with the same resolution limit of, say, 80 lines/mm will resolve four times more lines per width of 35mm film than it does per width of Super8 film.

 

For a standard 2K scan of 35mm film, where the film width is 22mm, the lens only needs to resolve 93 lines/mm (46 line pairs/mm), but for Super 8, where the width is a lot smaller (5.79mm) the lens needs to resolve 353 lines/mm (176 line pairs/mm) if it is reproduce the same signal.

 

Now the resolution of a lens is limited by microscopic-defects in the lens, in which the mutiple paths of the light wave, through the lens, fail to meet at a precise point.

 

So one possible approach to this limitation is to obtain the signature of a given lens. The defects will not be symmetrical so the resulting point spread function will also be non-symmetrical meaning that the signal is ammenable to weiner filter correction.

 

Carl

 

 

 

Carl

[Carl Looper]

Sure it is...(channel radio advert pitchman voice here), on the new Samsung Galaxy S, with it's massive 4.3" widescreen display!

 

Keep in mind; Lawrence of Arabia, conversely, will look like it was shot on 284.375mm, though :-(

 

Yes - that's very true. Whatever you can do to make Super 8 look better you can also do to make larger formats look better. This is not a sad thing.

[Rick Palidwor]

I think that in the age of YouTube, a feature shot entirely on super 8 is completely valid. Why not. for home video and the web, again why not?

 

Absolutely. Here is a link to some frame grabs from a feature I shot in super (duper) 8 with Mitch Perkins:

http://friendlyfirefilms.com/sleepalwaysframegrabs.html

 

Rick

[K Borowski]

If you're shooting for Youtube, or a low-resolution viral, maybe. For anything else, no S8 will not look as good as 35mm even on SDTV. It simply does not have adequate resolution.

[Carl Looper]

If you're shooting for Youtube, or a low-resolution viral, maybe. For anything else, no S8 will not look as good as 35mm even on SDTV. It simply does not have adequate resolution.

 

This comment doesn't engage with the previous discussion on super-resolution and reflects the usual knee jerk simplistic characterisations with which this problem has been treated by those who think the answer is too obvious. But even as early as the 1970s Lenny Lipton was recognising the promise that digital signal processing had to offer in this area. What was completely madness then is not so crazy now.

 

Digital signal processing can improve a signal considerably. And while native 35mm definition may be difficult to achieve for the humble Super 8, it is not completely out of the question, and is a worthy target for those working in the area of digital imaging.

 

The reason for testing these algorithms on Super8 is that Super8 is cheaper than testing it out on larger formats. But a nice byproduct is that a. the algorithms can be used to reduce the cost of shooting a film that might otherwise require 16mm or 35mm, and b. the same algorithms can also be applied to larger formats without any change.

[Friedemann Wachsmuth]

Keep on explaining, Carl (NOT Karl!). This is truly fascinating and interesting.

I poked around a lot with avisynth by now and, even if it it not (really/solely) applying SuperResolution algorithms, it it amazing what it can do to involve the time axis. Don't give up!

[Nicholas Rapak]

Or to put it another way, a lens with the same resolution limit of, say, 80 lines/mm will resolve four times more lines per width of 35mm film than it does per width of Super8 film.

 

For a standard 2K scan of 35mm film, where the film width is 22mm, the lens only needs to resolve 93 lines/mm (46 line pairs/mm), but for Super 8, where the width is a lot smaller (5.79mm) the lens needs to resolve 353 lines/mm (176 line pairs/mm) if it is reproduce the same signal.

 

The problem here is not the lens, but the film. The highest resolution microfilm is only capable of 140 lp/mm, so it is impossible to get 176 lp/mm out of a frame of Super 8 film.

[Nicholas Kovats]

Carl,

 

By film transport I meant to say the actual motion, speed, velocity, etc necessary to appreciate the actual and increased resolution inherent in random silver particle based film.

 

Hence my point regarding the comparison of noiseless HDSLR originated digital frames and static film frame captures.

 

Would you consider initiating the concept of Super 8 digital signal processing over at the Filmshooting Forum? A slightly more technical bunch who react quickly to posts in general, i.e.

 

http://www.filmshooting.com/scripts/forum/viewforum.php?f=1

 

Cheers!

[Will Montgomery]

How about 35mm looking like Super 8? Just tested my first Eyemo after years of 16mm and 8mm...

 

 

[Carl Looper]

The problem here is not the lens, but the film. The highest resolution microfilm is only capable of 140 lp/mm, so it is impossible to get 176 lp/mm out of a frame of Super 8 film.

 

Yes - the resolution of the film is a problem. However the solution is the previously discussed approach of "super-resolution". See previous posts. The native resolution (if you do nothing about it) is limited by the grain of the film. However, unlike photographic images, motion picture images encode a large number of correlated images (per shot) whereas the grain of each frame (from one frame to another) is uncorrelated (independant of the signal). Digital signal processing techniques can be used to recover a signal with a higher definition (resolution) than otherwise recoverable from a single frame.

[Carl Looper]

Carl,

 

By film transport I meant to say the actual motion, speed, velocity, etc necessary to appreciate the actual and increased resolution inherent in random silver particle based film.

 

Hence my point regarding the comparison of noiseless HDSLR originated digital frames and static film frame captures.

 

Would you consider initiating the concept of Super 8 digital signal processing over at the Filmshooting Forum? A slightly more technical bunch who react quickly to posts in general, i.e.

 

http://www.filmshooting.com/scripts/forum/viewforum.php?f=1

 

Cheers!

 

Not sure what you mean by "film transport". Perhaps you mean the motion of the objects in front of the camera - ie. that which is notionally being tracked by the motion signal processor.

 

HDSLR sensors are not actually noiseless. But typically the sensor noise is filtered. A significant ammount of noise is filtered by the simple act of spatial integration during downsampling to HD (the input sensors have more pixels than the output signal). And simple compression algorithms can smooth out noise due to the motion vector techniques of such.

 

Indeed it is the motion vector techniques of compression algorithms that can act as a useful starting point for understanding the idea of super-resolution. But instead of using the fast block matching/error correction codecs, super-resolution uses slower (computationally expensive) signal matching techniques. One thing not to use is the "error correction" component of compression codecs as that can reconstruct (to some extent) the very thing one is trying to suppress: the noise.

 

Regarding comparison of film to digital its difficult to establish a common frame of reference.

 

But the important point here is not so much whether one can improve the film signal to some notional digital standard - but just improve the signal full stop. How far the limits can be pushed back remains an open question. But the important point is that they can be pushed back.

 

I'll have a look at that other site.

[Carl Looper]

How about 35mm looking like Super 8? Just tested my first Eyemo after years of 16mm and 8mm...

 

 

 

Not sure I understand the comment but I love the clip. Its very beautiful. The interesting thing about 35mm is that you don't need a particulary good lens for a 2K scan, due to the physical size of the frame. For a 2K scan you only need a lens with a resolution limit of about 50 line-pairs/mm. Most modern lenses are a lot better than that.

[Carl Looper]

Can always try:

 

Super8 Kodachrome 40, circa 1984, scanned @ 2.5K (Canon EOS 1000D), with 2.2X SR processing:

 

Super8 K40 Super Resolution

 

Carl

[Roberto Pirodda]

many years by now i am datascanning S8 frames at 2.5k resolution with a 3ccd 2/3" progressive scan camera Sony DKC-ST5 2560x2048 pixel

here some stills

 

http://www.maraclub.it/still1.html

http://www.maraclub.it/still2.html

 

Hem 12 MB per frame sorry :(

these frames are as the camera output them, uncompressed TIFF, no digital manipulation at all.A 3ccd system outperform always , in terms of color rendition, a sigle chip sensor. I think that to improve S8 performance there is a simple solution : to shoot and project it at double speed ( see showscan)

Regards.

Roberto