16mm scan test

[carllooper]

Full size test can be seen here:

http://members.iinet.net.au/~carllooper ... ocused.jpg

[Nicholas Kovats]

Congratulations, Carl!

I take it this is from the NAC 16mm analytical projector. Did you stitch together 2x adjacent scans? The top frame is softer than the bottom. I am intrigued by your intentions. Is this UltraPan "n" ?

[carllooper]

In the previous post, the top image is the original scan. The bottom image is where I've done some post-refocusing on the scan.

Here's a preview of the full frame:



And here's the same at full presentation rez, with some additional work on post-refocusing:
http://members.iinet.net.au/~carllooper ... mFilm2.jpg

I've found that you need to scan film at a rez at least twice that of the digital presentation rez, irregardless of whether the film is 8mm, Super8, UltraPan8, 16mm, or 35mm.

Those attributes we call resolution and dynamic range are combined in film, in a way that is very different from digital. In digital images these terms are exact, but in film the terms (rather than the film) are not exact. The film holds the physical meaning of these terms in a non-localised way, requiring non-localised methods in order to translate the physical meaning of these terms in film, into the digital domain.

Carl

[Nicholas Kovats]

What do you mean by the reference "non-localised way"? In fact how are the film attributes resolution and dynamic range combined?

[grainy]

Hi Carl - why would you not scan at the highest possible rate, besides to save on storage?
Also, your cow scan looks GREAT!!!
G

[carllooper]

What do you mean by the reference "non-localised way"? In fact how are the film attributes resolution and dynamic range combined?

Non-local processing involves using information from an area larger than a pixel in order to determine what should ultimately go into the rendered pixel. The original pixel of the scan holds only localised information regarding the signal. But there are attributes of the signal, such as it's resolution, that are counter-intuitively distributed between pixels.

The additional pixels provide a better approximation of the otherwise quasi-continuous (and quasi-infinite) signal inscribed in the film. But at first glance, no matter how many pixels you use, the signal looks soft. So you might erroneously allocate less pixels (to avoid overkill). But by working with more pixels, you have a better approximation of the film and non-local processes are better able to extract a stronger version of the original signal that is otherwise "hidden" in the film.

When looking at the original film projected I don't think this information is hidden, or not as much. Your brain can see it or otherwise reconstruct it. But when you digitally scan the film, the digital system acts as a kind of filter between you and the film, disrupting the flow of information that would otherwise go on between the film and your brain. So you have to use more pixels plus digital algorithms in order to recover what you would otherwise see when watching the film with the naked eye.

Or to put it another way, you have to use digital algorithms to extract a good signal from a digital scan, and such algorithms work increasingly better with the more information they have about the film (ie. more pixels in the scan). But just having more pixels in a scan, on their own, doesn't necessarily achieve much.

I think a well implemented film to digital pipeline could even improve on naked eye perception of the film and what it's mediating.

Carl

[PyrodsTechnology]

well Carl, i am becoming impatient ! when will be available to us your magic software ?
Roberto
p.s. remember that it must be Mac compatible

[wado1942]

By "refocus" are you meaning optical alignment or some kind of digital sharpening?

[Nicholas Kovats]

Carl,

Your oversampling theory bears great fruit. From the forthcoming and stupendous 70mm film "SAMSARA",i.e.

http://barakasamsara.com/samsara/about

"TECHNICAL DETAILS

SAMSARA is one of only a handful of feature films, including its predecessor Baraka, to be shot in the 70mm format in the last forty years.

The 70mm to digital conversion process required that each frame be scanned at 8k resolution, creating a massive data file (in excess of 20 terabytes!) which was then reduced in size to create the final film. The filmmakers extensively tested this method to confirm the benefits of oversampling in the scanning of film to digital. The result is an unparalleled viewing experience with extraordinary levels of detail, clarity, and vibrance.

This method was originally pioneered in the creation of BARAKA's acclaimed Blu-ray DVD, and has since become a widely adopted industry practice. The achievement of the BARAKA Blu-ray marked the first time 70mm analog quality was successfully introduced into a digital format."

[carllooper]

By "refocus" are you meaning optical alignment or some kind of digital sharpening?

It's digital refocusing. The theory behind such is that an out-of-focus image is a composite of a large number of in-focus images, that are not aligned with each other, producing a soft (blurry) result. If you could separate out each of the in-focus images, and recombine them in a way where they coincided with each other, they would result in a sharp image. Something like that anyway. It's an idea that's been around since at least the early seventies (where I first read about it) but has an earlier history. Analog versions of the process (using specially prepared photographic plates as filters) were used in World War II on reconnaissance photography.

It's technical name is "deconvolution" ( http://en.wikipedia.org/wiki/Deconvolution ) of which there are a number of variants insofar as the base algorithm doesn't exploit knowable or computationally guessable properties of the image.

One of the problems I found was that the scratches on the film (which I like from an artistic pov) occur after the image is formed. So when refocusing the image (as distinct from the scratches), "shadows" appear adjacent to the scratches. Indeed shadows will appear along the edges of any foreground objects that are closer than the plane being refocused. In the example posted I was refocusing in relation to the trees in the background so everything forward of such generates some degree of shadowing. The shadowing reflects the inability of the algorithm to see what's behind an object in the scene, or see behind a scratch on the film. It can partially do so (due to the non local distribution of information) but the signal's amplitude drops quickly to zero causing a void to appear (black). Due to the sharpness of the scratches in the original scan being a lot sharper than the image, some high frequency scratch recognition could be done first, identifying pixels which are most likely scratches. Then once the refocus is done, you could use the identified pixels (representing the scratches) to infill such with the adjacent refocused image information. Or if you prefer the scratches to remain, just infill the void along the scratch edges.

Carl

[carllooper]

Carl,

Your oversampling theory bears great fruit. From the forthcoming and stupendous 70mm film "SAMSARA",i.e.

http://barakasamsara.com/samsara/about

"TECHNICAL DETAILS

SAMSARA is one of only a handful of feature films, including its predecessor Baraka, to be shot in the 70mm format in the last forty years.

The 70mm to digital conversion process required that each frame be scanned at 8k resolution, creating a massive data file (in excess of 20 terabytes!) which was then reduced in size to create the final film. The filmmakers extensively tested this method to confirm the benefits of oversampling in the scanning of film to digital. The result is an unparalleled viewing experience with extraordinary levels of detail, clarity, and vibrance.

This method was originally pioneered in the creation of BARAKA's acclaimed Blu-ray DVD, and has since become a widely adopted industry practice. The achievement of the BARAKA Blu-ray marked the first time 70mm analog quality was successfully introduced into a digital format."

I look forward to seeing the film.

Yes, downscaling improves the colour of the result. So the higher you sample the original, the greater the downscaling required to meet a particular presentation rez, and therefore the better will be the resulting colour.

My latest thinking is that small gauge film actually requires more oversampling than a large gauge film. The idea being that the transcendental image (the signal to recover) is more deeply hidden in small gauge film, requiring greater digital analysis, and consequently a greater number of samples to extract.

Carl

[carllooper]

well Carl, i am becoming impatient ! when will be available to us your magic software ?
Roberto
p.s. remember that it must be Mac compatible

The main challenge isn't the software as such since that's easily redistributable. It's chasing down physical hardware to capture the raw data at a high enough rez that the software has something substantial to work with.

Film transport system. Transformer. Macro lens. Camera. Bellows. Lens adapter rings. Clamps. Drills. LEDs. LED computer interface. Solenoids. USB breakout board. Wires. Soldering Iron. The list is endless. Without these the software does nothing.

Carl

[wado1942]

Sort of the like the proprietary Plangent Process for removing flutter from analogue tape. People keep asking him if he'll ever make the software available and he keeps telling them that the software is a very small part of it. It takes a special, high bandwidth tape deck, modded to run at half speed and still have the proper EQ (when returned to normal), high sample rates, special analogue to digital converter clocks and a custom built hardware to interface them together.

On a slightly different note, I just learned yesterday, that the low-frequency sharpening on Neat Video (noise reduction plugin) is good for reducing halation.