Flat bed vs DSLR scanning

[MattKing]

The "good enough to share on the web" thinking needs to go away.

Depends on what you intend your scans to do.
Those of us who don't intend to discard our negatives scan for particular utilities. In my case, reasonable sized colour prints (12" x 18" as a practical maximum) is the most demanding use of the scan. I always keep my negatives, and I can always in the future re-scan with more expensive options, so my interest in speed and efficiency is higher than some.
I'm end use focused - but for particular end uses.
My black and white scanning needs are even less demanding - in my life I've made (actually had made) one single digital print from a scan. Plus I once got prints from scans made at the time of processing for a roll of XP2. All the rest of my black and white prints are darkroom prints.
I'm all for maximizing the quality of the results within the constraints of practicality and expense. But if I need an extraordinary level of quality, I'll pay for it only when needed.
Give me a reasonably priced option that offers speed, efficiency, small footprint and flexibility along with higher quality, and I'll seriously consider it. But for now, my used flatbed fulfills my needs.

 

[Adrian Bacon]

No ergo.
It’s line pairs.
You need to at the very least multiply the lines pairs per millimeter rating by two.
In reality you need to multiply by three or more, because the real world is not aligned with the pixel array or consists of square waves.

For bunch of reasons (that are really easy and quick to research or deduce yourself) for optimal results (even at quite small reproduction ratios), you’ll want to sample an analog medium to its full maximum potential resolution whether the lens, microphones or pickups potential resolution warrents it or not.

One of them being that low resolution interaction with the substrates physical high frequency structure (that BTW almost never caps the maximum frequency range of the medium), will send artifacts down the frequency range, that will be quite visible/audiable at smaller reproduction ratios.

This is why I always advise to digitize at the maximum native sensor resolution, regardless of if it actually resolves at that detail level, then scale down to whatever resolution you intend to do post processing at (if at all), then scale down to the resolution you’re going to archive at. From there, if you need to output, make a work copy of the archive, make any compositional cropping, do any tweaks or touch ups, scale it to your intended output resolution, apply sharpening specific to your output, then output.

In the case of the Epson flatbeds, the optics tend act as an antialiasing filter to the sensor, so if you have the time, acquiring at the native sensor resolution will get the maximum the optics can deliver, then just scale down from there. I’ve seen pretty good evidence in my testing that even if you scan at a lower resolution, the Epson (at least the v850 Pro) still acquires at the native sensor resolution and internally scales it down before sending it over the wire, though I believe it just does a very naive add the samples and divide by two, implementation, so you’ll get better results by handling the down scaling yourself, but for less critical work, it’s fine to just scan at half or quarter the native sensor resolution if you faster scan times.

 

[McDiesel]

Depends on what you intend your scans to do.

Matt, I think you are misreading Helge's advice. The entire point is that when scanning, it doesn't matter what your intent is. You will "throw away" the excess later, but the end result - regardless of intent - always benefits from scanning at maximum resolution available to you. Flavio was showing a great example illustrating this.

[EDIT] I see Adrian saying the same thing in the comment above. This truly deserves to be a pinned advice in the "scanning" forum, because the misconception of "scanning depending on intent" is extremely common.

 

[MattKing]

Matt, I think you are misreading Helge's advice. The entire point is that when scanning, it doesn't matter what your intent is. You will "throw away" the excess later, but the end result - regardless of intent - always benefits from scanning at maximum resolution available to you. Flavio was showing a great example illustrating this.

With all due respect, I think my response was influenced as much by this part of your post, as by Helge's.

On the other hand, "maximum potential resolution" is such a rabbit hole... quickly leading to microscopes, $10K process lenses, and other exotics that incriminate that solid advice above as unrealistic or even insane.

I totally agree that it makes sense to make best use of what you have available already. If "maximum potential resolution" requires taking out your credit card for the maximization, then your intention needs to guide you.

 

[McDiesel]

Related, I just shared my result from last night scanning session. This is what I'd consider to be pretty close to insane, available for about $600 (that's how much I paid for that macro lens), but I already had the camera.

Normally I wouldn't even keep this file, it's just too huge for any practical purpose and far exceeds what I consider to be "acceptable enlargement ratio" of a 6x6 negative. I would normally downsample it to 6000x6000 and that is my personal definition of "maximum practical resolution" for a negative of this size.

 

[Adrian Bacon]

With all due respect, I think my response was influenced as much by this part of your post, as by Helge's.


I totally agree that it makes sense to make best use of what you have available already. If "maximum potential resolution" requires taking out your credit card for the maximization, then your intention needs to guide you.

For the record, I don’t disagree with scanning for intended output, if you have the negative and a way to scan it. There’s nothing wrong with that, if that’s what works for you.

personally, I’d prefer to handle the camera original material as few times as possible simply because every time you take it out of the sleeve and handle it, it’s going to collect crap floating in the air, so as soon as it’s dry, make a digital archival copy of it at the highest resolution available to you, within reason, that you can base further outputs on, then get the camera original in a protective sleeve as quickly as possible after that and get it archived. From there, only touch the camera original when you have to, such as when making darkroom prints, or if getting a higher resolution archival copy.

 

[Helge]

I am conflicted. On one hand, this advice needs to be pinned somewhere. The "good enough to share on the web" thinking needs to go away.

On the other hand, "maximum potential resolution" is such a rabbit hole... quickly leading to microscopes, $10K process lenses, and other exotics that incriminate that solid advice above as unrealistic or even insane.

Exactly. And that is why you never throw out your film.
That’s barbarism.
The original film (neg or pos) will always have the most data.
There is a reason Hollywood keeps rescanning film.
4K is only scratching the surface of even Super 35.

There is such a thing as diminishing returns though.
When you have resolved the grain really well you’ll probably need help from “deep learning” networks to extract more.

Grain is not binary, overlaps and is approaching stochastic distribution, so it’s absolutely possible that a really fantastically good lens will be able to write in that 3D structure.
Extracting resolution like that is probably the hard part.

 

[Alan Edward Klein]

This is why I always advise to digitize at the maximum native sensor resolution, regardless of if it actually resolves at that detail level, then scale down to whatever resolution you intend to do post processing at (if at all), then scale down to the resolution you’re going to archive at. From there, if you need to output, make a work copy of the archive, make any compositional cropping, do any tweaks or touch ups, scale it to your intended output resolution, apply sharpening specific to your output, then output.

In the case of the Epson flatbeds, the optics tend act as an antialiasing filter to the sensor, so if you have the time, acquiring at the native sensor resolution will get the maximum the optics can deliver, then just scale down from there. I’ve seen pretty good evidence in my testing that even if you scan at a lower resolution, the Epson (at least the v850 Pro) still acquires at the native sensor resolution and internally scales it down before sending it over the wire, though I believe it just does a very naive add the samples and divide by two, implementation, so you’ll get better results by handling the down scaling yourself, but for less critical work, it’s fine to just scan at half or quarter the native sensor resolution if you faster scan times.

For some reason, I found that scanning a 4x5 Tmax 400 at 2400 with my V850 sharpened better in Lightroom than when I scanned the same negative at 3200.

 

[Alan Edward Klein]

Matt, I think you are misreading Helge's advice. The entire point is that when scanning, it doesn't matter what your intent is. You will "throw away" the excess later, but the end result - regardless of intent - always benefits from scanning at maximum resolution available to you. Flavio was showing a great example illustrating this.

[EDIT] I see Adrian saying the same thing in the comment above. This truly deserves to be a pinned advice in the "scanning" forum, because the misconception of "scanning depending on intent" is extremely common.

I found that you can't get more than 2400dpi with a V850 and somewhat less with a V600. So scanning at let's say 6400 just makes bigger files and much longer scan times. So I standardized on 2400 with both machines.

 

[Helge]

This is why I always advise to digitize at the maximum native sensor resolution, regardless of if it actually resolves at that detail level, then scale down to whatever resolution you intend to do post processing at (if at all), then scale down to the resolution you’re going to archive at. From there, if you need to output, make a work copy of the archive, make any compositional cropping, do any tweaks or touch ups, scale it to your intended output resolution, apply sharpening specific to your output, then output.

In the case of the Epson flatbeds, the optics tend act as an antialiasing filter to the sensor, so if you have the time, acquiring at the native sensor resolution will get the maximum the optics can deliver, then just scale down from there. I’ve seen pretty good evidence in my testing that even if you scan at a lower resolution, the Epson (at least the v850 Pro) still acquires at the native sensor resolution and internally scales it down before sending it over the wire, though I believe it just does a very naive add the samples and divide by two, implementation, so you’ll get better results by handling the down scaling yourself, but for less critical work, it’s fine to just scan at half or quarter the native sensor resolution if you faster scan times.

You sound like you are doing a bang on job Adrian. If I'm ever in Petaluma I know where to send my film.

Problem with poor/suboptimal optics is that they don't just work as a straight anti aliasing filter. They treat certain frequencies differently, causing what is probably the equivalent of ringing in audio.

 

[Adrian Bacon]

I found that you can't get more than 2400dpi with a V850 and somewhat less with a V600. So scanning at let's say 6400 just makes bigger files and much longer scan times. So I standardized on 2400 with both machines.

It depends on what you're scanning and how you scan it. For full color material, 2400-2700 is the upper limit simply because the red channel maxes out at about that, but if scanning BW material and using vuescan, you can do a monochrome scan using the blue channel and the optics pass more resolution than that because blue frequencies are shorter wavelengths. The difference in resolution between doing a monochrome scan with just the red channel and then the same scan but with just the blue channel is very real and visible. Think about it, red to infrared is 620+ Nm, blue and UV less than 500 Nm. This is why optical disc standards went from red lasers with DVD to blue lasers with Blu-Ray. The shorter wavelengths allowed higher data densities on the disk. Scanning BW material with shorter wavelengths passes more resolution through the optics and lets you get more detail.

 

[Adrian Bacon]

You sound like you are doing a bang on job Adrian. If I'm ever in Petaluma I know where to send my film.

Problem with poor/suboptimal optics is that they don't just work as a straight anti aliasing filter. They treat certain frequencies differently, causing what is probably the equivalent of ringing in audio.

thanks. True, it's not a perfect AA filter, but it's generally good enough.

 

[Paul Verizzo]

Me too, me too.

Me too, too.

 

[brbo]

I’ve seen pretty good evidence in my testing that even if you scan at a lower resolution, the Epson (at least the v850 Pro) still acquires at the native sensor resolution and internally scales it down before sending it over the wire, though I believe it just does a very naive add the samples and divide by two, implementation, so you’ll get better results by handling the down scaling yourself, but for less critical work, it’s fine to just scan at half or quarter the native sensor resolution if you faster scan times.

I really wonder which manufacturers went to extra trouble (for zero gain) to actually NOT read the entire CCD line sensor when scanning at lower resolutions?

 

[Adrian Bacon]

I really wonder which manufacturers went to extra trouble (for zero gain) to actually NOT read the entire CCD line sensor when scanning at lower resolutions?

Don't know, but there's a lot of gain to read the whole sensor, namely, reduction of aliasing artifacts.

 

[Alan Edward Klein]

It depends on what you're scanning and how you scan it. For full color material, 2400-2700 is the upper limit simply because the red channel maxes out at about that, but if scanning BW material and using vuescan, you can do a monochrome scan using the blue channel and the optics pass more resolution than that because blue frequencies are shorter wavelengths. The difference in resolution between doing a monochrome scan with just the red channel and then the same scan but with just the blue channel is very real and visible. Think about it, red to infrared is 620+ Nm, blue and UV less than 500 Nm. This is why optical disc standards went from red lasers with DVD to blue lasers with Blu-Ray. The shorter wavelengths allowed higher data densities on the disk. Scanning BW material with shorter wavelengths passes more resolution through the optics and lets you get more detail.

I don't have Vuescan and scan with Epsonscan. Does it allow blue channel scanning?

 

[Helge]

It depends on what you're scanning and how you scan it. For full color material, 2400-2700 is the upper limit simply because the red channel maxes out at about that, but if scanning BW material and using vuescan, you can do a monochrome scan using the blue channel and the optics pass more resolution than that because blue frequencies are shorter wavelengths. The difference in resolution between doing a monochrome scan with just the red channel and then the same scan but with just the blue channel is very real and visible. Think about it, red to infrared is 620+ Nm, blue and UV less than 500 Nm. This is why optical disc standards went from red lasers with DVD to blue lasers with Blu-Ray. The shorter wavelengths allowed higher data densities on the disk. Scanning BW material with shorter wavelengths passes more resolution through the optics and lets you get more detail.

Have you done the experiment?
I’m doubtful as to whether there is an actual visible effect for that reason.
The optics are most likely just bad and mainly corrected for blu - green.

 

[flavio81]

Related, I just shared my result from last night scanning session. This is what I'd consider to be pretty close to insane, available for about $600 (that's how much I paid for that macro lens), but I already had the camera.

And i'd say it's not THAT extreme: You're clearly resolving the grain, however this is HP5. With a film like Acros 100, the grain would be much finer and there would be even more detaill, not just grain.

Awesome.

For the ones who want to see it for yourself, this is a 100% crop of McDiesel's scan:

https://d3ue2m1ika9dfn.cloudfront.net/120-bw/insane-stitch-and-pixelshift.jpg

 

[flavio81]

It depends on what you're scanning and how you scan it. For full color material, 2400-2700 is the upper limit simply because the red channel maxes out at about that, but if scanning BW material and using vuescan, you can do a monochrome scan using the blue channel and the optics pass more resolution than that because blue frequencies are shorter wavelengths. The difference in resolution between doing a monochrome scan with just the red channel and then the same scan but with just the blue channel is very real and visible.

This is most likely due to chromatic aberrations on the scanner lens. Which means the two color extremes (red and blue/violet) don't focus at the very same spot.

 

[250swb]

This is most likely due to chromatic aberrations on the scanner lens. Which means the two color extremes (red and blue/violet) don't focus at the very same spot.

Probably true but kind of a harsh criticism given all scanner lenses are likely to be cheap and have the same problem. The point is it works, a situation when photographers can exploit 'faults' and use good software to create a better image, nothing to grumble about there.

 

[McDiesel]

BTW, these two blogs are good to follow if you're into camera scanning:

• Jim Kasson. He covers all kinds of digital imaging topics, and has a series of posts of scanning with Fuji GFX and Nikon platforms.
• Closeup Photography is dedicated to macro. Awesome resource.

 

[Adrian Bacon]

Have you done the experiment?
I’m doubtful as to whether there is an actual visible effect for that reason.
The optics are most likely just bad and mainly corrected for blu - green.

Yes. I can't remember if I posted results in another thread somewhere. If I get some time, I'll try to post results again if anybody wants to see them, though if you have a scanner and vuescan it's not that hard to try it out.

 

[Adrian Bacon]

Probably true but kind of a harsh criticism given all scanner lenses are likely to be cheap and have the same problem. The point is it works, a situation when photographers can exploit 'faults' and use good software to create a better image, nothing to grumble about there.

exactly. Scanning with the blue channel renders visibly more detail. I'll take it.

 

[Helge]

Probably true but kind of a harsh criticism given all scanner lenses are likely to be cheap and have the same problem. The point is it works, a situation when photographers can exploit 'faults' and use good software to create a better image, nothing to grumble about there.

A scanner lens could be very simple if you use a super strong backlight to get a small aperture and use a peaky RGB backlight exposing sequentially, combined with a monochrome sensor, so remaining CA is more easily corrected in software.

 

[MattKing]

Scanner lenses are probably much easier to design, because they don't need to work with a wide range of magnifications and they are only required to image to and from a really small field.
The film and scanning head transports are more of a challenge.