nyquist theory

[Filip Plesha]

I was just thinking.

 

A lot of people think that if film can "store"/resolve let's say 120 line pairs,

or 240 pixels per one millimeter than this means that you should use 240 pixels per mm to "describe" the film image. But it really doesn't go in both ways.

 

You can say that film's resolving power equals that of a 4K/6K/8K electronic sensor, but that doesn't mean that you can preserve all the information by using 4K/6K/8K scanning. This would work if the details on the image were in perfect phase with the position of pixels. But what if the lines on film are moved by one third of the width of one pixel? You don't get a proper digital representation.

 

That would mean that you need AT LEAST 4 pixels for one line pair. Allthough perhaps even 6 pixels would be nice. These are huge resolutions. That would mean 15K to show fines details of 1000:1 contrast target lines on a slower film of maximum resolving power of 150 lp/mm for which we would usually use some 7K scanning to capture "ALL" the details.

 

Of course this is all imposible, now or in 10 years, so it is worthless to think like that, but I wonder where did this idea that you need 2 pixels for a line pair came from? It is misleading.

 

Not to mention the problems of capturing diagonal lines. On film they will be diagonal, but if you use 2 pixels per line pair, you will get a chess field instead of lines.

[DanStewart]

That's exactly the thinking behind over-scanning at say 4k for a 2k job, or now maybe 8k for a 4k job. As you say the amount of data quickly gets very silly if you try to keep an 8k+ scan, but as long as the scanner has the buffer (and you have the time) a higher res scan is evident even after downres, which makes the point nicely.

[K Borowski]

Amen. I feel that a 2K scan is suitable for maybe S8, not 35mm. Someone was saying a while back on this forum that 6K is all that is needed to capture full detail in a 65/70mm movie such as Lawrence of Arabia. I sincerely doubt that such a downgrading in image quality would stand up on the big screen compared to a true 65mm print. We are lucky though that the industry is now moving into 4Ks. I don't think we'll see much higher resolution than that for a long while. Kodak seems to think that that is all that is necessary to capture the information written on film, as they describe in their new system that uses DIs to reduce generation loss. What a shame. I definitely think 4K is losing a lot still.

 

Regards.

[Filip Plesha]

I'd say that the use of double resolution scanning for a certain output resolution is more of a demand resulted by the nature of CCD's (they are never perfect, and at this time never fill the entire potential of a digital file), and of course 8K scanning will solve a lot of "conflicts" that a sensor might have when trying to "understand" the light patterns that it is capturing. A perfect example of this is grain aliasing.

Aditional noise apperas because the sensor is somewhat "confused" by the little variations in density that are smaller than the sensor itself. This is why 8K is better. That way you have exeeded the size of a typical grain cluster and there are no "conflicts" A sensor is able to describe a grain cluster with few pixels and this order of things is translated into a cleaner 4K file when downresizing.

 

But this does not solve the nyquist problem as you said. The nyquist theory fails in theorethically perfect 4K files (perfect 4K files would come from oversampling from 8K or at least 6K) Why? Well, just for the sake of argument let's say that film emulsion will have th eresolving power of 4K pixels. No matter from what resolution you have downsampled, you still have one pixel for one line because you are operating on the limit. If the line is not in phase, you won't get accurate results. This is not so important when you are dealing with real images because the image will really be filled with 4K details, no matter how perfectly accurate or not they might be. But even though this may be irrelevant, the problem still exists in theory.

 

And using 8K or 16K for 4K output won't solve the problem, because the problem is not in the sensor, but in the sampling rate of the FILE itself. And this is where Nymquist has been mostly misquoted a number of times.

People say things like:

The Nymquist theory says that you need to use the twice the resolution of your output file to get a perfect file at that resolution. First of all. When you are scanning at pure 4K, you are allready using the Nymquist theory, considering that you need two pixels for each line pair. (at 85lp/mm) And secondly, mr. Nymquist

died long before scanners existed and the theory does not deal with oversampling or anything similar.

The theory simply states (which is logical) that you need two sampling points (pixels) for a modulation from low to high (in this case black and white) which perfectly makes sense.

The only thing 8K downsampling solves is the grain aliasing issue and the elementary problems of CCD sensors.

[Filip Plesha]

Film4ever...

 

What you are saying is true for the emulsion itself, but this is a romantic version of film imaging. The reality is a bit different.

 

Consider that you need 6K for 35mm only for very steady images. The exposure time is 1/50sec. That means that no matter how small the move of the camera is, the loss is huge. But of course, scanning at 6K is worth even if there are only few steady shots in the film.

 

Secondly, you haven't considered the lens. People often think that there is no need to consider the lens resolution if it is higher than the one of film (for example, film resolves 160lp/mm and lens 300 lp/mm) But that is not true.

The formula is 1/max resolution of system = 1/ max resolution of film + 1/max resolution of the lens.

 

Here is an example. You have a film that resolves 120 lp/mm (which is probably the resolving power of today's fast films). This is about 6K resolution. Your lens resolves some 300 lp/mm. So you can ignore it? Nope. If you put all that in the formula you get that your final resolution of the system will be only 85 lp/mm which equals 4K.

 

The bottom line is that no matter how sharp your lens is, it will ALWAYS limit

the resolution of the image on film below the one you can see in the MTF chart for the film. You can trully consider the MTF data in cases of contact printing or laser recording.

 

4K is suitable for most faster emulsions in real world conditions. You can consider all the significant data to be captured at that resolution (unless you are a pain in the ass like me that starts to question the whole sampling theory :) )

 

 

And as for slower films, well it depends, depends on the aperture. For apertures of f/11 and beyond, you can be sure that 4K is enough for ALL the modern emulsions, even those that reach 200lp/mm resolution (10K !!)

And this is not fixable with technological advancement of glass manufacturing because it is limited by the pure wave nature of light.

[Phil Rhodes]

Hi,

 

Filmlight's superb Northlight scanner scans at 6K and then optionally downsamples to lower resolutions for these reasons. It is a world-class piece of machinery without which any modern postproduction establishment would clearly be lost.

 

And yes, I actually AM in the paid employ of Filmlight at the moment...

 

Phil

[Filip Plesha]

Northlight seems like an impressive piece of technology.

And i find it a big promise for thiongs to come.

 

If anyone suddenly decides to digitize an entire film at 4K, he would prbably use the northight, and that means 6K downsampling, which is great. Future scanners sure won't go a step back, they will probably use 6K downsampling as a minimum (the new imager will use 8K downsampling). That means the industry has finally entered an age where 4K files will be filled with maximum data possible.

I am really currious what could a possible northight2 bring (10K downsampling to 8K??)

[John Pytlak RIP]

Kodak image scientist Dr. Roger Morton and his team have published several recent SMPTE papers on this very topic:

 

SMPTE Journal, May 2003:

 

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

 

SMPTE Journal, July 2003:

 

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

[Filip Plesha]

I am not a subscriber so I can't read the article, but i suspect that he concluded by giving a different proposal of pixel sampling that the nyquist theory proposes.

(something like that is mentioned in the "synopsis")

 

What I wonder is, how come even Kodak sticks to this whole 4K-6K scanning

ideal? It is based on the nyquist theory of sampling.