Comparing digital and film resolution using MTF

[Overrank]

From a practical ie. real world perspective the question is can you actually achieve 2k resolution /HD from Super 16 film and the answer appears to be yes ( or very close.) 4K from Super 35 scans is achievable. In theory, scanning resolutions of up to >11000 PPI might be required but in practice it may be lower but >8K. I did once bookmark some Panavavision lens data but can't find it at present but seem to remember that these had high MTF , relative to lenses for 35 mm full-frame still photography, when measured at 80 cycles/mm. For a Super 35 negative, at 24.9 mm wide, this would mean 3984 horizontal pixels at 80 cycles/mm or 5278.8 pixels at the ITU limit of 106 cycles/mm and 6% MTF.

This is a 2K scan of Super8



65mm can be scanned at 18K, so I suspect greater than 4K resolution from 35mm

 

[iljitsch]

I found this 2009 paper that indicates that higher scanning resolutions are required than the figures above.

Very good paper. However, they do at some point argue that 4k is good enough, probably because that's the resolution of the scanner they are selling.

An interesting point:

"Scanning a first-generation OCN is also the extreme case. What the ITU tests also showed (Figure 2) was that after just one film generation (the answer print), the MTF fell to zero well before 106 lp/mm and even at 80 lp/ mm was only about 4%; in fact, 20% modulation level was maintained only to about 50 lp/mm; this second generation‘s information content could therefore be cap- tured adequately with far less scanning resolution than 4K; probably a 2.5K scanner would suffice."

So: 4k for 35 mm movies is only really worth it if you go back to the original negatives. Otherwise, you're getting barely more than 2k.

From a practical ie. real world perspective the question is can you actually achieve 2k resolution /HD from Super 16 film and the answer appears to be yes ( or very close.)

More practically: does 16 mm have more resolution than SD? The answer is definitely yes as apparently 8 mm does, too, be it not by a super wide margin.

I used this as an excuse to find my old Buffy the Vampire Slayer DVDs. The first and second seasons were filmed on 16 mm, the later ones on 35 mm. I didn't immediately see a difference on the 480i DVDs.

For those who can't get enough: here's an older thread discussing much the same thing.

Oh, and why did we photographers never get that very high resolution film that the movie guys got??

 

[Jack Hogan]

I pretty much agree with your findings on how many Mpix are needed for a fair digital reproduction of film images: Once you have the grainy film response across the image field, you can expect that higher resolution will mainly show more detail of the film capture noise. I think that already 6 Mpix maybe OK for Kodachrome II 35mm slides. I used a 24 Mpix crop camera and on accidentally archived the captures at 6 Mpix with little regret as the grainy film base shows consistently. I expect that the 45 Mpix of D850...Z9 cameras will be good enough for for 4x5 and more. But, I probably will not have a chance to inspect such images.

By that reasoning, 200MP should be enough for 8x10. Looking at the onlandscape comparison, I don't think that would be the case:

https://www.onlandscape.co.uk/2020/02/8x10-film-vs-iq4-150mp/

Interesting Chris, though looking at the image of the lens captured on 8x10 Velvia 50 vs the 80MP IQ180, the call could be closer than shown with a 200MP sensor.

The traffic sign crop was shot with 150MP:

Yes, undersampled, that's why I was suggesting that in this context we need to also talk about pixel size and format (which inevitably brings lenses into the equation).

For instance it may very well be that 6MP is 'enough' for Kodachrome II 135 film with a crop sensor and about 8um effective pixel pitch per Bernard above (say a Nikon D40x with a good FF lens). However, when scaling that to cover an 8x10" area everything should be scaled, otherwise the result may not be applicable:

8x10*25.4^2/24/36 ≈ 60x the area

60*6MP ≈ 360MP

go figure, right in the 300-600MP range guesstimated by onlandscape. Scaling the area of the taking sensor proportionately

(15.7x23.5mm)*sqrt(60) ≈ 122x182mm ≈ 4.8x7.2" sensor size.

The 150MP IQ4 is only 40.5x54mm = 1.6x2.1", so it suffers a major penalty in terms of magnification - hence sampling and aliasing, as clearly seen (some folks compensate by stitching).

That may be a valid comparison for their purposes (is state-of-the-art MF definition comparable to 8x10 film's?) but not in the context of this thread: a MP by itself is incomplete information. They suggest that 300-600MP should do it on MF. Would they suffice on Micro Four Thirds? How big would those pixels need to be? Does such a sensor exist?

Jack

 

[JimKasson]

Oh, and why did we photographers never get that very high resolution film that the movie guys got??

Kodak Tech Pan wasn't good enough for you?

 

[The Ghost of Caravaggio]

...It’s been stated elsewhere that there are issues with Tim Vitale’s paper ( e.g. https://www.photrio.com/forum/threads/film-vs-scanning-resolution.177544/page-5 ) where Adox (film manufacturer, of among other things CMS-20 high resolution black and white film - https://www.fotoimpex.com/shop/images/products/media/30945_5_PDF-Datasheet.pdf ) state

“We can confirm that from our own countless resolution tests over the years.
The articles of Clarke and Tim Vitale [i.e. Film Grain, Resolution… ] unfortunately have many errors and give overall a wrong assessment of the performance of film. We cannot recommend them.”

Thank you for this information.

 

[iljitsch]

Hi all,

With DPReview closing down I want to thank all of you who participated in this thread for doing so. It was definitely my favorite discussion here in recent history.

Also, I used some of the material from my contributions in this blog post:

http://www.iljitsch.com/2023/03-29-comparing-digital-and-film-resolution-using-mtf.html

I dipped into my bag of Javascript tricks to add a slider to the film format overview so you can select a line pair per mm setting you feel comfortable with and the image width and megapixel numbers are updated in real time.

If there is another forum somewhere out there that hosts discussions like this, please let me know. If not, so long and thanks for all the fish.

 

[J A C S]

MTF does not make sense for film (or for a display, or even for a sensor). MTF requires a linear translation invariant system while film does not do that on macro scale. If your lens is soft enough and you are interested in not so high resolution, then you are fine, approximately. If you want to get to the bottom of it, you have all those random crystals which provide some kind of random locally averaged sampling. This is not linear, not translation invariant.

Now, you can measure something convincing yourself that you are measuring MTF but you are not.

 

[iljitsch]

I've read quite a lot of stuff on this subject before, during and after the discussion in this thread, and although you are technically correct that film MTF curves are not linear, they do seem to follow a rather predictable curve. Unlike lenses, BTW.

But does that matter? If we exclude the counter intuitive idea that MTFs can go up as lp/mm increases, then the only thing that matters is at what lp/mm number the MTF dips below a contrast that we find acceptable/usable. At that lp/mm we've found the minimum resolution needed for scanning film.

Currently, I'm in the process of watching Minority Report in 4K.

It does seem the movie was originally finished on film, but later remastered using a 4K digital intermediate. Some scenes have very nice sharpness, others not so much. But the grain is pretty horrible. It looks like they boosted the contrast, which helps sharpness, but also makes the grain much more prominent. (And perhaps they also explicitly increased sharpness.) I don't find this a pleasant tradeoff. (When I saw Ghostbusters in 4K on Netflix some time ago this was even worse.)

With the zooms of that building in the park I posted earlier you can see that although a 46 megapixel scan delivers a bit more detail than a 12.5 megapixel scan, it becomes quite hard to determine if something is just a random clump of grain or an actual detail. The 20.6 megapixel digital shot (with a lens that was likely sharper and of course focus with a mirrorless camera is going to win from focus on a 20-year-old low end film camera) at ISO 100, you know that when you see something, it's an actual detail recorded from the real world.

I think we can all reasonably agree that MTF50, which results in 2K resolution for 35 mm movies, leaves recoverable image detail on the table in virtually all cases. (Ignoring truly horrible film / processing / lenses.) But MTF06, which is 8K for 35 mm movies, surely must be the upper limit and no more image detail can reasonably recovered. (Perhaps unless using some insanely slow technical black-and-white film and exceptional lenses under perfect circumstances.)

 

[Jack Hogan]

I've read quite a lot of stuff on this subject before, during and after the discussion in this thread, and although you are technically correct that film MTF curves are not linear, they do seem to follow a rather predictable curve.

There is a central linear region in the curve - and non-linear regions at either end. One could say that it is mostly linear where it counts. Or that it is equivalent to being linear throughout but then followed by non-linear 'processing'

Unlike lenses, BTW.

But does that matter? If we exclude the counter intuitive idea that MTFs can go up as lp/mm increases,

If we stick to in-spec photographic lenses with clear circular apertures, used in their typical working range, the spatial frequency response transferred to raw data can only decrease monotonically to diffraction extinction.

then the only thing that matters is at what lp/mm number the MTF dips below a contrast that we find acceptable/usable. At that lp/mm we've found the minimum resolution needed for scanning film. [...]

I think we can all reasonably agree that MTF50, which results in 2K resolution for 35 mm movies, leaves recoverable image detail on the table in virtually all cases. (Ignoring truly horrible film / processing / lenses.) But MTF06, which is 8K for 35 mm movies, surely must be the upper limit and no more image detail can reasonably recovered. (Perhaps unless using some insanely slow technical black-and-white film and exceptional lenses under perfect circumstances.)

Can you remind us how you arrived at those equivalencies?

Jack

PS There is a site that is being setup to try to keep the DPR community together. It has been set up in a hurry so it is a bit rough around the edges for now, but it's early days and its objective is to mimic DPR functionality, check it out:
https://dprevived.com/

 

[Erik Kaffehr]

I've read quite a lot of stuff on this subject before, during and after the discussion in this thread, and although you are technically correct that film MTF curves are not linear, they do seem to follow a rather predictable curve.

There is a central linear region in the curve - and non-linear regions at either end. One could say that it is mostly linear where it counts. Or that it is equivalent to being linear throughout but then followed by non-linear 'processing'

Unlike lenses, BTW.

But does that matter? If we exclude the counter intuitive idea that MTFs can go up as lp/mm increases,

If we stick to in-spec photographic lenses with clear circular apertures, used in their typical working range, the spatial frequency response transferred to raw data can only decrease monotonically to diffraction extinction.

then the only thing that matters is at what lp/mm number the MTF dips below a contrast that we find acceptable/usable. At that lp/mm we've found the minimum resolution needed for scanning film. [...]

I think we can all reasonably agree that MTF50, which results in 2K resolution for 35 mm movies, leaves recoverable image detail on the table in virtually all cases. (Ignoring truly horrible film / processing / lenses.) But MTF06, which is 8K for 35 mm movies, surely must be the upper limit and no more image detail can reasonably recovered. (Perhaps unless using some insanely slow technical black-and-white film and exceptional lenses under perfect circumstances.)

Can you remind us how you arrived at those equivalencies?

Jack

PS There is a site that is being setup to try to keep the DPR community together. It has been set up in a hurry so it is a bit rough around the edges for now, but it's early days and its objective is to mimic DPR functionality, check it out:
https://dprevived.com/

Hi,

My take may be that the reasonable approach may be:

• Find the MTF curves or curves for a selection of film.
• Find the MTF curve for a reasonably good lens
• Find the MTF curve for a reasonable scanning system (*)
• Multiply the the three MTF curves
• Find a reasonable cut off, for some cy/mm, say 20% MTF
• Multiply that 'cy/mm' by image dimensions
• Multiply that number by 4, as it takes minimum 2 pixels / cycle and divide with MP

I would recall that the resolution of motion film, as showed in theatres, is a bit below 1080P.

Best regards

Erik

 

[iljitsch]

I think we can all reasonably agree that MTF50, which results in 2K resolution for 35 mm movies, leaves recoverable image detail on the table in virtually all cases. (Ignoring truly horrible film / processing / lenses.) But MTF06, which is 8K for 35 mm movies, surely must be the upper limit and no more image detail can reasonably recovered. (Perhaps unless using some insanely slow technical black-and-white film and exceptional lenses under perfect circumstances.)

Can you remind us how you arrived at those equivalencies?

I'm using MTF50 at 40 lp/mm based on:

Resolution and MTF curves in film and lenses by Norman Koren

KODAK VISION3 50D Color Negative Film 5203/7203 The world’s finest grain film

Then Koren's Lorentzian equation to convert MTF50 to other MTFs and ~ 25 mm wide Super35 film, so:

• MTF50 = 40 lp/mm * 25 = 1000 line pairs = 2000 pixels horizontally
• MTF20 = 80 lp/mm * 25 = 2000 line pairs = 4000 pixels horizontally
• MTF10 = 120 lp/mm * 25 = 3000 line pairs = 6000 pixels horizontally
• MTF06 = 160 lp/mm * 25 = 4000 line pairs = 8000 pixels horizontally

So adding real-world considerations like less than 100% contrast and lens limitations, I think it's reasonable to conclude there is no recoverable detail beyond 8K for this class of film.

Experience shows that 35 mm cinema film can definitely outperform 2K so MTF50 as a sharpness limit seems too conservative.

The real question is how much better Super35 and the like can be scanned at 8K vs 4K. Probably a little for really well-shot movies.

In practice, it's moot, as I don't see 8K becoming a consumer format. There is simply no workable viewing distance that lets you discern the extra level of detail over 4K.

Maybe 8K TVs will be a thing as apparently it's possible to build screens with tens of millions of subpixels with high yield today and on large TVs, these pixels are pretty large so shrinking them should be fairly easy. And as we camera people know all to well, larger numbers generate to more sales.

But broadcast TV hasn't even made the jump to 4K as far as I'm aware, and from the occasional necessity to carefully wipe down even brand new UHD blu-rays, I doubt there's a big capacity increase possible there without reducing reliability. And with declining physical media sales, nobody wants to add another format and start scanning their libraries all over again.

PS There is a site that is being setup to try to keep the DPR community together. It has been set up in a hurry so it is a bit rough around the edges for now, but it's early days and its objective is to mimic DPR functionality, check it out:
https://dprevived.com/

I will.

 

[iljitsch]

I would recall that the resolution of motion film, as showed in theatres, is a bit below 1080P.

Hm, I think the relevant resolutions are 2K and 4K, which in the cinema world means 2048x1080 and 4096x2160 , respectively. (So everything gets shrunk 6% for home consumption, reducing sharpness, sigh.)

I don't know how common 4K projection is in movie theaters. I was at a press screening for a Dolby Vision theater back in 2015 and I remember that their twin projectors were 4K. But I didn't write that down, so I can't be sure.

 

[finnan haddie]

...
I would recall that the resolution of motion film, as showed in theatres, is a bit below 1080P.

Yes, the film distributed to film theaters is the result of several film duplication generations, and mostly ends up somewhere between 720p and 1080p.

IIRC Spider-Man 2 (2004) was the first blockbuster movie where digital post production was partially done in 4k.

 

[TheGrammarFairy]

The problem with movie scans is always that dozens of people were involved in the quality decisions.
And there’s the skill of the team doing the scanning and what their objectives, and the condition of the print or negative they chose to scan.
Of course consumer televisions do all kinds of behind the scenes image enhancement, or disenhancement as we saw in the final episode of Game of Thrones.
So theres a high probability that in the booth when the scan is being done the show probably looked several times better on their $50,000 monitors than it looks at home or in the theater.

Why is the universe so unjust to truth seekers?

I've read quite a lot of stuff on this subject before, during and after the discussion in this thread, and although you are technically correct that film MTF curves are not linear, they do seem to follow a rather predictable curve. Unlike lenses, BTW.

But does that matter? If we exclude the counter intuitive idea that MTFs can go up as lp/mm increases, then the only thing that matters is at what lp/mm number the MTF dips below a contrast that we find acceptable/usable. At that lp/mm we've found the minimum resolution needed for scanning film.

Currently, I'm in the process of watching Minority Report in 4K.

It does seem the movie was originally finished on film, but later remastered using a 4K digital intermediate. Some scenes have very nice sharpness, others not so much. But the grain is pretty horrible. It looks like they boosted the contrast, which helps sharpness, but also makes the grain much more prominent. (And perhaps they also explicitly increased sharpness.) I don't find this a pleasant tradeoff. (When I saw Ghostbusters in 4K on Netflix some time ago this was even worse.)

With the zooms of that building in the park I posted earlier you can see that although a 46 megapixel scan delivers a bit more detail than a 12.5 megapixel scan, it becomes quite hard to determine if something is just a random clump of grain or an actual detail. The 20.6 megapixel digital shot (with a lens that was likely sharper and of course focus with a mirrorless camera is going to win from focus on a 20-year-old low end film camera) at ISO 100, you know that when you see something, it's an actual detail recorded from the real world.

I think we can all reasonably agree that MTF50, which results in 2K resolution for 35 mm movies, leaves recoverable image detail on the table in virtually all cases. (Ignoring truly horrible film / processing / lenses.) But MTF06, which is 8K for 35 mm movies, surely must be the upper limit and no more image detail can reasonably recovered. (Perhaps unless using some insanely slow technical black-and-white film and exceptional lenses under perfect circumstances.)

 

[J A C S]

I pretty much agree with your findings on how many Mpix are needed for a fair digital reproduction of film images: Once you have the grainy film response across the image field, you can expect that higher resolution will mainly show more detail of the film capture noise. I think that already 6 Mpix maybe OK for Kodachrome II 35mm slides. I used a 24 Mpix crop camera and on accidentally archived the captures at 6 Mpix with little regret as the grainy film base shows consistently. I expect that the 45 Mpix of D850...Z9 cameras will be good enough for for 4x5 and more. But, I probably will not have a chance to inspect such images.

By that reasoning, 200MP should be enough for 8x10. Looking at the onlandscape comparison, I don't think that would be the case:

https://www.onlandscape.co.uk/2020/02/8x10-film-vs-iq4-150mp/

thanks for the link. I had forgotten about Velvia 50, with almost MTF50 at 50c/mm .

Compare those two crops:

https://www.onlandscape.co.uk/wp-content/uploads/2020/02/meeting-810-detail.jpg (the Velvia)

meeting-iq-detail.jpg (2048×1207)

So much aliasing and distortions in the second one! BTW, the second one is scanned, and has some faint staircase aliasing as well but mild.