How many "pixels" does film consist of?

Niklas · July 12, 2004

Does anybody know how many pixels you would need in order to simulate a 35 mm movie frame? I've heard something like 80 Mb of data for one frame of Jurassic Park, but don't know if it's correct. Does anybody know? 10K * 10K ? Or even more? Less?

Alvin Pingol · July 12, 2004

Since film is a truly analogue format that has no fixed pixel "grid," it is impossible to determine an exact pixel equivalence for a MP film frame.

However, when it comes to useable image data (i.e. not wasting "pixels" on film grain that is nearly invisible to the naked eye), a 35mm MP film frame is equivalent to approximately 12 megapixels.

David Mullen ASC · July 12, 2004

It's generally considered that you need to scan at approx. 4000 x 3000 pixels (depending on the aspect ratio) minimum -- 12MP or "4K" -- to capture all the information (grains) of the original, although some now say it's more like approx. 6000 x 4500 -- 27MP or "6K".

However, working at "2K" (approx. 2000 x 1500 pixels) for scanning and recording back to film has been considered an "acceptable" industry compromise for years, especially for interior and night scenes. Recently the trend has been to scan at 4K but downrez to 2K for all the work done to the image.

Since film does not have pixels, it does not have a pixel resolution, so the only question is how much resolution is necessary in order to not be losing any information in the film frame.

By comparison, Sony HD cameras are generally 1920 x 1080 pixels (2MP) although recording formats like HDCAM reduce this to 1440 x 1080 (1.5MP). You could consider this "sub 2K" resolution.

Nate Downes · July 12, 2004

Any "pixel" rating given is almost always wrong for a few reasons.

1) No two film stocks are exactly the same.

2) A film project could process their film in a "non-standard" manner in order to achieve a special look

3) A film could be shot in a "non-standard" manner, again, affecting the "resolution" of the stock.

So, someone could say "Film is 80Mb/sec" and be right.... for one stock processed one way and exposed one way. Change *ANYTHING* and suddenly the chart goes out the window. And new stocks are arriving, and arriving, and arriving, so once you calculate out every possible manner with stocks... suddenly new stocks, new solutions and new techniques arrive changing the whole game yet again.

John Pytlak RIP · July 12, 2004

Almost everyone agrees it takes at least a 4K scan to capture most of the information contained on a 35mm frame. Hence the 12 megapixel figure for a 4K by 3K frame scan. Generally, slower films are sharper and finer grained.

If you consider film grains to be random "pixels", grain size varies from sub micron (less than 1 micrometer across) to about 5 microns (the fast yellow layer in the fastest color negative films). There are hundreds of millions of individual silver halide grains in each 35mm frame.

If you are an SMPTE member, there are several techical papers on the subject in the on-line SMPTE library:

http://www.smpte.org/members_only/library/...file=morton.pdf

http://www.electronicipc.com/journalez/det...=45390011120508

http://www.electronicipc.com/journalez/det...=45390011120705

John Sprung · July 12, 2004

There are hundreds of millions of individual silver halide grains in each 35mm frame.

And each grain is either exposed or not -- they don't have in-between states. So each grain is basically a binary digit. If we had a more precise handle on how many there are in a frame, that would give us an upper bound on the real information content of a frame. For instance, 800 million grains would be 100 Mega Bytes.

4k x 3k would be 12 Mega Pixels, which would mean we have about 8 Bytes worth of film grain data per pixel.

I don't see any immediate useful conclusion from that, but it sure is interesting. It feels like maybe some day somebody much smarter than me might connect some dots here.....

-- J.S.

Phil Rhodes · July 12, 2004

Hi,

Normally a 2K 10-bit log DPX is about 12Mb. Four times that for 4K.

Phil

Filip Plesha · July 12, 2004

Phil, he is talking about something else..

John. I see what you mean, but that would require you to have

multiple size and multiple shape pixels, not on a fixed grid but scattered

all around.

John Sprung · July 12, 2004

that would require you to have multiple size and multiple shape pixels, not on a fixed grid but scattered all around.

Yes, if the idea were to create an electronic/digital sampling system that worked more like film. That's a different subject, and I think ultimately a really good idea.

TV, with its horizontal scan lines, and digital with its regular array of equal sized and equally spaced little squares like screenwire are both subject to aliasing which causes moire patterns. That's why you have a Nyquist limit for them. But film doesn't have a Nyquist limit because its sampling structure is random in size shape and location, within reasonable bounds.

Chips are made by a photo-resist etching type of process, so there's nothing that limits them to the screenwire pattern of squares that everybody uses now. If we could create a pseudo-random sampling array that adequately mimics the randomness of film grain, we could give ol' Nyquist his walking papers.

The problem is that the pseudo-random pattern would have to be standardized so that everybody's equipment would work with the same data. It's like the size, shape, and location of the sprocket holes in film. It's that standard that makes it possible to shoot film from Japan in a camera from Germany.

What would have to be standardized is a sort of "tile" of this pseudo-random array, and the "tiles" could be linked together to make an imaging system of whatever size you want.

Clearly the randomness of film grain is a good thing that we should want to bring forward into an advanced digital system. While we're at it, we should look at the rest of its characteristics, like sensitivity being proportional to sample site size, having only one bit per grain, grains having only one of three colors, etc.

But what I was actually thinking about is that the raw uncompressed data per frame of a digital system could be compared with the number of film grains per frame, because grains are in a sense binary -- something more like the idea of information content used in compression and encryption.

-- J.S.

Filip Plesha · July 12, 2004

It is all logical, but very Sci-fi in a way..

First of all, you say that you would have to mark every grain with

1 or 0 , it would be one bit large, yes but what would you mark as

a zero? If the grain is not 1 than it does not exist. How would you

tell a computer that there is no grain? By specifying an area and saying

that it is zero. Which again comes back to a fixed grid of small surfaces or

pixels.

But there could be another way. You could specify the grain with its coordinates. But then you would not have to describe the grain with 1 and 0. Because since it is specified by its coordinates and size, it means that it does exist. There is no point of telling someone coordinates of something that does not exist.

I think this all would lead to something similar to how a ink jet printer works,but sort of reversed. A file format with specified coordinates of dots. Sort of how vector graphics work. But it would be very very complicated and I am not sure wheather it would take less space than a normal bitmap.

And then another problem is how to display it. You would have to convert it to normal pixels to display it. You couldn't even print it to film, unless you could make a laser that will change shape of the beam and turn the beam any whay the computer wants.

It is all very complicated. Quite frankly, it would be much simpler to make a 6K bitmap file that would solve all the grain structure problems and aliasing.

The lines on film that reach such small widths as one or two pixels in 4K or 6K are so low in contrast and "eaten" by grain that you youldn't really get any major aliasing problems.

4K is marginal in terms of ethics of capturing ALL the information. But in realitty, on screen nobody would ever notice the smallest lines.

First of all they would be like 10% response on the MTF curve, and secondly they SHOULD NOT be visible in the first place because if they are that would mean that you can see the grid on screen, and that is a big NO-NO for digital projection.

John Sprung · July 13, 2004

First of all, you say that you would have to mark every grain with
1 or 0 , it would be one bit large, yes but what would you mark as

a zero? If the grain is not 1 than it does not exist. How would you

tell a computer that there is no grain? By specifying an area and saying

that it is zero. Which again comes back to a fixed grid of small surfaces or

pixels.

And then another problem is how to display it. You would have to convert it to normal pixels to display it

All the grains in a film frame exist until the film is developed. They're all there when the shutter opens and closes. The ones that get hit by enough photons (IIRC, about six) get changed from unexposed to exposed. It's the same number of photons no matter how large or small the grain, that's why big grains make film faster, and small ones make it slower -- for the same number of photons per square micron, the big ones get more hits. Then in the lab the exposed ones get converted into metallic silver, and the unexposed ones get dissolved and washed out.

To do a pseudo random digital system, what you'd need to do is design a sort of "tile" pattern that would be standardized and used for the mask sets of both CCD's and LCD's and all other imaging surfaces. Instead of the regular screenwire pattern, this would be more of a jigsaw puzzle kind of thing, with the size, shape, and location of the sampling areas loosely based on looking at highly magnified film grain. The idea is kinda like that Pergo flooring, where you're actually walking around on photo lithographed pictures of real wood. Look carefully at any piece, and you can find identical copies of it all over the room. The standard tile image would have to be designed so that right edges fit with lefts and tops with bottoms, like jigsaw puzzle pieces, so you could make arrays of any size.

Ideally this data would be displayed on the same kind of pseudo random array it was shot with, but some sort of digital downconversion to the "screenwire" system would probably also be done.

-- J.S.

David Mullen ASC · July 13, 2004

Of course, the advantage of film is that the grain structure changes frame by frame, which is why projecting a series of frames quickly creates a moving image that looks finer-grained than each individual frame.

John Sprung · July 13, 2004

Of course, the advantage of film is that the grain structure changes frame by frame, which is why projecting a series of frames quickly creates a moving image that looks finer-grained than each individual frame.

Ah, yes. this would make it interesting to try alternating two different pseudo random arrays.

When you do a simple freeze frame, the stopping of the grain motion is very noticeable. "Rocking" a small piece of film, even just alternating two frames, solves that problem. You just need to get the actors to stop well enough that you can find a pair of frames with almost no motion.

-- J.S.

Filip Plesha · July 13, 2004

John you are talking about camera sensors, and I was talking about film scanning.

I was talking about storing actuall scans of film images in this format of yours.

John Sprung · July 13, 2004

John you are talking about camera sensors, and I was talking about film scanning.
I was talking about storing actuall scans of film images in this format of yours.

Not just camera sensors, a complete integrated system including cameras, telecine, direct and projection displays, and all the post stuff. It's a whole wierd little universe of hypothetical theoretical blue sky pseudo random array imaging technology.... ;-)

The way these things turn out in the real world is always stranger that we could ever imagine. F'rinstance, as recently as 1952, nobody could even dream that there would some day be time code. We wouldn't have 59.94 and drop/non-drop if they had.

Flying spot (Rank) and line array (Spirit) telecines wouldn't work in this system, you'd have to use an area array like that Sony camera based telecine, but with a pseudo random array. Either that or do a grid type scan with so many pixels that a conversion to the pseudo random array system could be made.

-- J.S.