Film vs. Scanning resolution

[grat]

So last night, I got to thinking about how much resolution is actually needed to scan film.

On the LF forum, the statement was made that the scanning resolution needs to be high enough to see the film grain, and I'm not sure of the "why" for that statement.

I went looking, and Fuji very helpfully lists the resolving power of their Pro 400H film on their datasheet, and they list it as:

17. RESOLVING POWER
Test-Object Contrast: 1.6:1 ............. 50 lines/mm
Test-Object Contrast: 1000:1 ........125 lines/mm

OK, so high contrast is going to be easier to distinguish, low contrast not as much, that makes sense. Did some hunting around for "what does lp/mm mean in terms of ppi", and got a number of answers, but eventually tracked down an equation, only to discover that you can specify lp/cm as a scanning resolution in SilverFast, and it helpfully can translate to ppi-- and 1250 lp/cm is 3175 ppi, and just inside the "yellow" area on the ppi slider.

So to make a long story into a short question, if Fuji rates their film's resolving power at 125 lp/mm, which apparently translates to 3175 ppi, what's the benefit of scanning at a higher resolution?

 

[Les Sarile]

No doubt that film type, conditions, scene, equipment and settings used ultimately determine what is needed. As a practical comparison this is what 4000dpi looks like compared to DSLR scanning using K20D 14.6MP and D800 36.3MP.
Film was 35mm Kodak Techpan shot at ISO25 and developed in Kodak Technidol.
Bottom left crop shows the whole target 4 X 4 arrangement of 12233 charts. Center in red is shown as 100% crops above.
Above it the 14.6MP Pentax scan.
Above it the Coolscan 4000dpi scan.
Above it is the 36.3MP Nikon scan. Even though the D800 applies more pixels then 4000dpi, they are pretty much equivalent,
Big crop on the right is that red center portion optical magnification at about 4.5X. Clearly there is much detail unresolved by any of the above methods.

Full res version -> http://www.fototime.com/8372250EA44CB06/orig.jpg[/QUOTE]

 

[Alan Edward Klein]

The real limitation is the scanner. I found that 2400 is best with Epson V600 and V850. I shoot Velvia 50 and Tmax 100.

 

[grat]

But why scan over 3200 PPI if the film resolution doesn't support it?

 

[MattKing]

Film's resolution is a lot more complex than a single number.
And scanners essentially sample the information in a piece of film, so scanning resolutions are a lot more complex than a single number.
If the optics and mechanics and electronics in a scanner actually support higher resolutions - the flatbeds probably don't - then there is more more to gain. Whether it is worth the extra time and file size may be a more important question.

 

[Bud Hamblen]

A good 35 mm negative is traditionally said to resolve 30 lines/mm. You need 30 x 2 (a line takes 2 pixels to resolve) x 1.414 (assuming the lines are at 45 degrees to the pixel pattern) = 85 pixels per mm = 2159 pixels per inch. You want set your scanner's software to an even fraction of the specified optical resolution to keep from dealing with fractions of a pixel, so if your scanner claims 4,800 pixels per inch without interpolating, set your scanning software to 2,400 pixels per inch. In round numbers that is 8 megapixels for a 24mm x 36mm negative.

The systematic way is to scan a known good resolution target and set the scanning software no higher than you need to get the resolution you need.

Bud

 

[138S]

So last night, I got to thinking about how much resolution is actually needed to scan film.

On the LF forum, the statement was made that the scanning resolution needs to be high enough to see the film grain, and I'm not sure of the "why" for that statement.

I went looking, and Fuji very helpfully lists the resolving power of their Pro 400H film on their datasheet, and they list it as:

17. RESOLVING POWER
Test-Object Contrast: 1.6:1 ............. 50 lines/mm
Test-Object Contrast: 1000:1 ........125 lines/mm

OK, so high contrast is going to be easier to distinguish, low contrast not as much, that makes sense. Did some hunting around for "what does lp/mm mean in terms of ppi", and got a number of answers, but eventually tracked down an equation, only to discover that you can specify lp/cm as a scanning resolution in SilverFast, and it helpfully can translate to ppi-- and 1250 lp/cm is 3175 ppi, and just inside the "yellow" area on the ppi slider.

So to make a long story into a short question, if Fuji rates their film's resolving power at 125 lp/mm, which apparently translates to 3175 ppi, what's the benefit of scanning at a higher resolution?

First, consider that in practice you won't be able to project something on film having 125 lines/mm detail at 1000:1, which is 10 stops difference between light on some point on film and light over another point that is 8 microns far. Lens MTF, subject's microcontrast and flare won't allow that 1000:1 contrast in 125 lines/mm detail, this simply does not happen in real photography.

Portra 160, 400, and Fuji 160 and 400 are totally well scanned with a bare Epson 700-850 sporting 2300 to 2900 effective dpi (depending on the axis). A recent and reliable side by side demonstrated that.

What is true is that Proffessional scanners deliver very well digitally optimized images, while consumer and prosumer scanners do require some optimization job in Ps.


Regarding if scanning BW grain is necessary or not, this is about personal taste. Historically some preferred diffusion or condenser enlargers to make grain more or less evident.

 

[Lachlan Young]

On the LF forum, the statement was made that the scanning resolution needs to be high enough to see the film grain, and I'm not sure of the "why" for that statement.

It's fairly simple & very complex at the same time... You want to transmit the resolving power and (more importantly) the MTF characteristics of the film, and in so doing, you will end up transmitting some of the visible granularity/ sharpness etc of the film - the successful transmission of which is very important for how we perceive image content. To take an example, if I made a 30x enlargement from a 24x36 neg with a regular 50mm enlarging lens (nominally optimised 2-15x) and one with a specialist 50mm enlarging lens (nominally optimised 15-50x), you'd pick out the latter as being 'better', not because it resolved more of the detail of the image, but because the granularity, sharpness etc seemed better translated, making the image not seem 'soft' and slightly unsharp with a lack of microcontrast (and this is independent of whatever light source you are using - which, unless it's a point source, will only affect the gross overall contrast). Does this make sense? RMS Grain is tested by the manufacturers at about the equivalent of 529ppi, so you can see that if a scanner cannot begin to resolve granularity well at even quite low resolutions (usually from poor MTF in the optical/ mechanical imaging path - and/ or poor decisions involving pixel offsets) while you may get (to one extent or another) resolution of the image, the granularity becomes a strange un-sharp mess, not helped by the often extreme unsharp masking that people then apply to try & salvage the wreckage. It's also perfectly possible to overdo the scanning resolution (in a high MTF system) & end up with noisier images than in an equivalent darkroom print, but that's where stuff starts to get complicated... Quite a bit also has to do with the problems of output/ print resolution (which fades out rather than coming to a hard stop at 5, 8 or whatever cyc/mm various people claim based on squinting at a resolution chart). Generally, anything over 1200-1600ppi at high native MTF will start to communicate some degree of visible granularity.

 

[Ariston]

I would like to see someone scan a frame of film, then scan a silver print of the same frame, and post it so we can see the difference in resolution for scanning vs printing. I'll have my darkroom up, one day, so I can do this.

 

[Lachlan Young]

then scan a silver print of the same frame

Unless you do this using a good repro setup or a really good scanner, all you'll be testing is the scanner's performance.

 

[Helge]

So last night, I got to thinking about how much resolution is actually needed to scan film.

On the LF forum, the statement was made that the scanning resolution needs to be high enough to see the film grain, and I'm not sure of the "why" for that statement.

I went looking, and Fuji very helpfully lists the resolving power of their Pro 400H film on their datasheet, and they list it as:

17. RESOLVING POWER
Test-Object Contrast: 1.6:1 ............. 50 lines/mm
Test-Object Contrast: 1000:1 ........125 lines/mm

OK, so high contrast is going to be easier to distinguish, low contrast not as much, that makes sense. Did some hunting around for "what does lp/mm mean in terms of ppi", and got a number of answers, but eventually tracked down an equation, only to discover that you can specify lp/cm as a scanning resolution in SilverFast, and it helpfully can translate to ppi-- and 1250 lp/cm is 3175 ppi, and just inside the "yellow" area on the ppi slider.

So to make a long story into a short question, if Fuji rates their film's resolving power at 125 lp/mm, which apparently translates to 3175 ppi, what's the benefit of scanning at a higher resolution?

Line pairs per mm is one black and one white. You can’t have a line without something to differentiate it.
So you should double the number.

Also “cycles per mm” is quite different. Sine curves have lower contrast per default.

You need at least 125 (line pairs) x 2 (to get pixels or dots) x 36mm = 9000 potential pixels and 6000 for the short side of the frame.
So that’s 54 megapixels.

That is with perfectly aligned pixels to straight lines.
Nyquist would like more.
Double would be good.

To avoid grain aliasing you also need the resolution, whether you used a pinhole or a Rokkor 24mm.

A ten stop difference isn’t too unusual. For example something silhouetted against the sky, water reflections, human made objects etc.

Keep in mind electronic sensors have many of the exact same problems with lowered contrast. And when it comes to colour contrast (as opposed to near black and white), as in nearly every real world situation, electronic bayered sensors have it much worse.

 

[Adrian Bacon]

So last night, I got to thinking about how much resolution is actually needed to scan film.

On the LF forum, the statement was made that the scanning resolution needs to be high enough to see the film grain, and I'm not sure of the "why" for that statement.

I went looking, and Fuji very helpfully lists the resolving power of their Pro 400H film on their datasheet, and they list it as:

17. RESOLVING POWER
Test-Object Contrast: 1.6:1 ............. 50 lines/mm
Test-Object Contrast: 1000:1 ........125 lines/mm

OK, so high contrast is going to be easier to distinguish, low contrast not as much, that makes sense. Did some hunting around for "what does lp/mm mean in terms of ppi", and got a number of answers, but eventually tracked down an equation, only to discover that you can specify lp/cm as a scanning resolution in SilverFast, and it helpfully can translate to ppi-- and 1250 lp/cm is 3175 ppi, and just inside the "yellow" area on the ppi slider.

So to make a long story into a short question, if Fuji rates their film's resolving power at 125 lp/mm, which apparently translates to 3175 ppi, what's the benefit of scanning at a higher resolution?

To put it simply, 125 lines/mm translating into 3175 ppi is the minimum resolution you should aim for, simply because that ppl number is only if you manage to get every theoretical line pair to exactly match up with every sampling pixel. In real life, that never happens. Just like in the audio world, even though technically, we only need a sampling rate of 44.1Khz for human hearing, and that does make for pretty high quality sound, sampling at a much higher sample rate will sound a lot better for a significant amount of musical material. The same goes for visual material. Even though technically 3200dpi would be enough for 400H, sampling at a much higher spatial frequency will tend to provide significantly more visual fidelity when enlarged. The smaller the film format, the more important that is, the larger the film format, the less you're enlarging anyway so having a high sampling rate is less critical. You only need to sample at a high enough sample rate to see the grain if you actually want to see the film grain. One of the benefits of larger format film is less film grain for the same composition.

 

[Alan Edward Klein]

Adrian I was looking at the bit rate of a couple of songs I bought downloaded from Amazon music. One was at 212 kbps the other 216 kbps. Then I looked at a Dixie Chicks song off a disk. It was at 128kbps. I don;t recall if that was the original disk rate from the original disk or the bit rate I selected in the program to copy it into my computer.'

In any case, what is the recommended bit rate for use in your car, phone for hearing with earphones, and a top rate stereo system in your home?

What bit rates are recommended when recording video in your cell phone or digital camera, assuming they're selectable?

 

[138S]

A ten stop difference isn’t too unusual. For example something silhouetted against the sky, water reflections, human made objects etc.

Human made objects won't usually deliver 1000:1 microcontrast, instead 10:1 it can be found if white paint is next to black paint, a black paint reflects around 5% to 15% of the light, also the detail in the subject has to go from 1000 to 1 in 0.008 mm on film, which won't happen in water reflexions.

So single situation it remains is contrasty silhouettes. But it happens that silhouettes are well optimized with sharpening algorithms anyway.

Then add that in a real scene nothing is in perfect focus but in the DOF, so the 1000:1 difference in those 0.008mm on film is quite difficult to happen. And now add that the lens by far won't have 100%MTF at 125 cy/mm, so most of the contrast at that frequency is destroyed. Then add flare... also most of shots are handheld...

In reality the vast majority of the shots are not to surpase 30 or 40 cy/mm efective resolving power.

 

[Adrian Bacon]

Adrian I was looking at the bit rate of a couple of songs I bought downloaded from Amazon music. One was at 212 kbps the other 216 kbps. Then I looked at a Dixie Chicks song off a disk. It was at 128kbps. I don;t recall if that was the original disk rate from the original disk or the bit rate I selected in the program to copy it into my computer.'

In any case, what is the recommended bit rate for use in your car, phone for hearing with earphones, and a top rate stereo system in your home?

What bit rates are recommended when recording video in your cell phone or digital camera, assuming they're selectable?

the bit rate of music is a totally different thing as it’s usually compressed. Uncompressed CD sound is ~1.5Mbps. If it’s a compressed mp3, if you want reasonably good sound, 160Kbps is pretty much the minimum for most things though 128 is passable, if it’s AAC compressed, 128 is the minimum, with 96 as passable. Both are lossy, so it will depend on the content.

for recording video, here in the US, broadcast TV mastering minimum is MPEG 2 video at 30 frames per second with 4:2:2 color sampling and 50Mbps. ATSC digital TV over the air is ~19Mbps MPEG 2. Most cable TV is that at best.

MP4 (with h.264 compression) is approximately twice as efficient as MPEG 2 compression, so ~24Mbps for a minimum is fine, though many cameras encode at much higher bit rates than that. This is all for full HD 1080 video. If shooting 4K, multiply by 4 for that minimum, and by 4 again for 8K.

 

[grat]

I appreciate everyone taking the time to answer, particularly Lachlan and Adrian. Useful information, although I think it confirms what I've suspected-- for my current usage (just mine, not anyone else's), 2400-3200 PPI is fine for scanning. Should I take a photo I like enough to turn into a giant print, I'll revisit the topic.

As for audio compression, an audio CD's native sampling rate is 44.1 KHz, or 96 KHz if it's "high resolution". Most compression algorithms are producing, quite literally, a reasonable facsimile of the original uncompressed stream-- something to trick the ears into believing they're hearing the original. Below 128kbps this will be noticeable, and depending on your hearing and your sound system (all of which are now optimized to make the best of 128kbps compressed audio), you might be able to hear artifacts up to 160kpbs, or 192kpbs.

Personally, if I'm transcoding files myself, I prefer FLAC (compressed, but doesn't lose any information), or if I have to do compression, I like VBR, or variable-bit-rate, which means for the simple parts, the bitrate will plummet, and for complex parts, the bitrate will climb. Similar adaptive strategies are used for video compression.

 

[flavio81]

But why scan over 3200 PPI if the film resolution doesn't support it?

because:

1. 3200 ppi would be a bare minmum to get those 125lp/mm

2. B/W films have even higher resolving power

and, most importantly

3. very few film scanners can give real 3200ppi optical resolution (or more).

 

[Adrian Bacon]

I appreciate everyone taking the time to answer, particularly Lachlan and Adrian. Useful information, although I think it confirms what I've suspected-- for my current usage (just mine, not anyone else's), 2400-3200 PPI is fine for scanning. Should I take a photo I like enough to turn into a giant print, I'll revisit the topic.

As for audio compression, an audio CD's native sampling rate is 44.1 KHz, or 96 KHz if it's "high resolution". Most compression algorithms are producing, quite literally, a reasonable facsimile of the original uncompressed stream-- something to trick the ears into believing they're hearing the original. Below 128kbps this will be noticeable, and depending on your hearing and your sound system (all of which are now optimized to make the best of 128kbps compressed audio), you might be able to hear artifacts up to 160kpbs, or 192kpbs.

Personally, if I'm transcoding files myself, I prefer FLAC (compressed, but doesn't lose any information), or if I have to do compression, I like VBR, or variable-bit-rate, which means for the simple parts, the bitrate will plummet, and for complex parts, the bitrate will climb. Similar adaptive strategies are used for video compression.

I like flac quite a lot, but wish there was a lot more support for it in the popular music playing software stacks. I stopped doing MP3s a really long time ago and basically standardized on AAC at 256Kbps.

I edit quite a lot of video, and see all manner of stuff come across my desk. You’d be amazed how many people go and shoot a tv commercial and edit It themselves in an effort to keep costs down then get their commercial totally rejected by the local TV station because it doesn’t meet broadcast specs. Very few TV stations are going to accept an MP4 file that you made to put over the air. Their equipment isn’t designed to handle it and their in-house staff isn’t going to fix it for you. They just don’t have the time. That’s when my phone rings with somebody trying to get their video file in a format that will be accepted.

 

[alanrockwood]

To put it simply, 125 lines/mm translating into 3175 ppi is the minimum resolution you should aim for, simply because that ppl number is only if you manage to get every theoretical line pair to exactly match up with every sampling pixel...

A slight correction... well actually a fairly big correction to what you wrote. The Nyquist limit to resolve 125 lines per mm is 6450 ppi, which is double the figure you quoted. That is a lower limit to the sampling rate required, and as a practical matter a substantially higher rate would be required.

By the way, there is a subtle and seldom appreciated point with regard to the sampling theorem. The theorem says that if the sampling rate is at least as high as the Nyquist limit then it is possible to reconstruct the original sample without error. The theorem does not say that the sampled result is, in and of itself, an accurate representation of the original signal. To reconstruct the signal actually requires interpolating the raw data with a certain interpolating function to give more points. The correct interpolation function will produce a continuous function, but to get a good representation of the original function at discrete points requires resampling the reconstructed continuous function at discrete points, and the density of sampling points for the resampled result needs to be much higher than the Nyquist sampling rate. Otherwise, signals that are close to the Nyquist limit (but within the Nyquist limit) will be severely distorted. The correct interpolating function is based on the sinc function.

 

[138S]

These are Portra 160 (120 film) 6400 dpi crops from an Epson V700, two top notch Creos and an Scanmate 11000 drum, V700 is top-left:







While the scanmate drum has a way, way better resolving power no practical benefit is found from that with Portra 160 because the film itself it's the limiting factor. The V700 capability at around 2600dpi effective is outresolving Portra. But to obtain true 2600 effective with an Epson a careful (flatness/height) scanning is required, and we have to oversample well beyond those 2600, and also a careful edition is required to get sharp look if print is to be relatively large.

https://www.largeformatphotography....Epson-Flatbed-Eversmart-Flatbed-Drum-Scanners

If anyone has a doubt about what the scanner is resolving or not then he may invest $6 in a x60 loupe to see what's in the film, this solves any doubt.

 

[Helge]

These are Portra 160 (120 film) 6400 dpi crops from an Epson V700, two top notch Creos and an Scanmate 11000 drum, V700 is top-left:

View attachment 253113

View attachment 253115



While the scanmate drum has a way, way better resolving power no practical benefit is found from that with Portra 160 because the film itself it's the limiting factor. The V700 capability at around 2600dpi effective is outresolving Portra. But to obtain true 2600 effective with an Epson a careful (flatness/height) scanning is required, and we have to oversample well beyond those 2600, and also a careful edition is required to get sharp look if print is to be relatively large.

https://www.largeformatphotography....Epson-Flatbed-Eversmart-Flatbed-Drum-Scanners

If anyone has a doubt about what the scanner is resolving or not then he may invest $6 in a x60 loupe to see what's in the film, this solves any doubt.

View attachment 253114

I have inspected Portra 400, Ektar and Tmax 100 through a pro microscope (where lens was a 60mm 2.8 Nikkor macro stopped down to 8) and let me just unequivocally state that 2600 dpi does not even begin to out-resolve the potential resolution of those films.
2600 dpi is around 9 megapixels for a Barnack frame FFS! Don't be ridiculous!
You can only blame operator error, equipment error (those old, big-metal scanners need some expensive CLA every year. It can be tempting to let it slide, if there is no complaints), or we can speculate about just plain fabrication of results, for whatever reason.

 

[Helge]

A slight correction... well actually a fairly big correction to what you wrote. The Nyquist limit to resolve 125 lines per mm is 6450 ppi, which is double the figure you quoted. That is a lower limit to the sampling rate required, and as a practical matter a substantially higher rate would be required.

By the way, there is a subtle and seldom appreciated point with regard to the sampling theorem. The theorem says that if the sampling rate is at least as high as the Nyquist limit then it is possible to reconstruct the original sample without error. The theorem does not say that the sampled result is, in and of itself, an accurate representation of the original signal. To reconstruct the signal actually requires interpolating the raw data with a certain interpolating function to give more points. The correct interpolation function will produce a continuous function, but to get a good representation of the original function at discrete points requires resampling the reconstructed continuous function at discrete points, and the density of sampling points for the resampled result needs to be much higher than the Nyquist sampling rate. Otherwise, signals that are close to the Nyquist limit (but within the Nyquist limit) will be severely distorted. The correct interpolating function is based on the sinc function.

Exactly! As I alluded to in post #11 Nyquist is one of the most missused and poorly understood guys by people who really should know better.

 

[Helge]

Human made objects won't usually deliver 1000:1 microcontrast, instead 10:1 it can be found if white paint is next to black paint, a black paint reflects around 5% to 15% of the light, also the detail in the subject has to go from 1000 to 1 in 0.008 mm on film, which won't happen in water reflexions.

So single situation it remains is contrasty silhouettes. But it happens that silhouettes are well optimized with sharpening algorithms anyway.

Then add that in a real scene nothing is in perfect focus but in the DOF, so the 1000:1 difference in those 0.008mm on film is quite difficult to happen. And now add that the lens by far won't have 100%MTF at 125 cy/mm, so most of the contrast at that frequency is destroyed. Then add flare... also most of shots are handheld...

In reality the vast majority of the shots are not to surpase 30 or 40 cy/mm efective resolving power.

That depends entirely one the water and the reflection.
True that it will usually only be a small part of the image that benefits from the resolving power, but that is just down to the lens. A wide-angle lens stopped down will be able to deliver a lot of sharp detail.
Man made stuff can have a surprising amount of high micro contrast in bright light. Just think of a car, printed text, peeling paint in sidelight etc.
Of course the resolving power is not a simple matter of "is there/is not". Even non obviously used resolving power, will still have an impact on the overall look of the frame. Even when digitally scaled down to Instagram size.
This is not mysticism, but something that can be well described and explained by looking at signal theory as mentioned in the above posts.

 

[Lachlan Young]

although I think it confirms what I've suspected-- for my current usage (just mine, not anyone else's), 2400-3200 PPI is fine for scanning.

Possibly - it can depend a huge amount on the quality of the OTF/ MTF of the scanner - for example, it's possible to have a scanner - let's say 2400-2500ppi - that has quite low MTF (for arguments sake, let's say it drops below 50% response at just under 700ppi or equivalent resolution) and another scanner that is running at maybe 1200-1500ppi range, but has drastically higher MTF/ OTF. Take a piece of film that contains a systemic MTF of perhaps 1000-1200ppi at 50% response (that's combining film MTF, lens MTF, film plane accuracy etc) and scan it on both & you'll see that the nominally lower resolution but higher MTF scan will look significantly better - it may be close to lossless, whereas the higher resolution but much lower MTF scan gets more and more lossy as the MTF curve of the scanner drops under the MTF of the original. Added to that, a high native sharpness in a scan has systemic consequences both in terms of the file being able to maintain a higher information capacity (less noise because low/ no sharpening needed) with better overall contrast & by being able to scan at a resolution closer to that needed for the final result, less storage space is needed & scanning times are shorter. There are however significant opportunity costs involved in this decision. It is equally unfortunate that the marketing departments of certain well-known scanner manufacturers (Epson, Plustek) seem to believe that more rather than better pixels are the path to success. In both cases most of their machines' shortcomings seem to derive from onboard interpolation methods to try and overcome mechanical/ optical limitations in the cheapest possible ways. Undoing that at a firmware level would probably significantly improve their overall visual performance.

or we can speculate about just plain fabrication of results, for whatever reason.

There's no need for speculation: he was publicly asked to desist in posting those claims by the person whose scans he is distorting the results of.

 

[Adrian Bacon]

A slight correction... well actually a fairly big correction to what you wrote. The Nyquist limit to resolve 125 lines per mm is 6450 ppi, which is double the figure you quoted. That is a lower limit to the sampling rate required, and as a practical matter a substantially higher rate would be required.

yes I’m aware of that. I didn’t correct the OP simply because the actual required resolution is somewhat irrelevant given that it’s quite a bit higher than what anybody has any easy access to, and the point I was trying to make was that higher sampling frequencies will look better.