Scanner comparisson page and drum scan limits?

[SCHWARZZEIT]

The text came up in English for me! Guess something out there knows whether we speak English or German?

Sandy

That's great. Somehow I must have missed that they have the review text up in different languages.

-Dominique

[SCHWARZZEIT]

More generally, the expert eye has always been the best judge. Scanner SNR is directly related to so many factors that providing a broadly applicable quality metric would be very difficult. I do not recall ever seeing a thorough analysis back when PMT and CCD scanners were relevant to some lucrative markets. People bought what made their data look good - as they do today I guess.

It probably wasn't necessary to do a scientific analysis outside of an engineering level. There was the Seyold Report where a bunch of prepress scanners were evaluated.
I agree that the expert eye is the best judge. Looking at the scans from different machines and playing around with the files, trying to see where the limitations are is probably the best you can do to make a buying decision.
The scanner comparison page on this site is also a great resource to get an impression. But I guess some differences would be more obvious if the scans were done at a higher resolution. One more suggestion would be to host the original scans somewhere for those interested to play with the files. Today hosting space is much cheaper than it was when this comparison started.

Practically, I believe a PMT-based system can be engineered to have better SNR at small apertures than any area-based detector. LN2-cooled or amplified CCDs can actually detect single photons and have some advantages (primarily speed), but a PMT-based detector is more sensitive (better SNR) at low flux levels.

I am sure the engineers on this board can put it better but here's my simple summary. At anything under NASA price levels, PMTs maintain better SNR at small sampling apertures. This yields better MTF = perceptually better rendition of intensity variation overall and less visible "grain" in light image areas. Drum scanners are a fast way to position a PMT over large areas so we like 'em for LF. Wish I had a drum to play with today.

Thanks for your input, Peter. As Lenny pointed out the influence of the scanner operator should not be underrated.

-Dominique

[Lenny Eiger]

Any idea how the old Linotype-Hell scanners performed at their stated maximum resolution of 24,000 spi?

Sure. They didn't. For starters, I don't mean to pick on you, or make you defend yourself. Just in case you feel that way, this isn't personal at all.

I used to be very upset at Imacon's quoting of certain numbers that were actually out of range. They suggested their DMax was 4.8, I believe, which exceed the max on film, of 4.6. How did they actually get that result? I railed against them in forums for lying to us about their capabilities. That is, until one fellow took the time to explain to me how they got their result. There was a mathematical formula based up the probable capabilities of such a device with its components, etc. It was theoretical number. They were quoting theoretical while others were quoting tested values. It got worse, when Epson suggested it can do 6400 real resolution, which we all know is too high. Now its all just a mess of one kind of disinformation vs another.

In conversations with Aztek they said that their stepper motor was capable of 18,000 steps around the circumference of the drum. That means that they could have taken samples and produced a much higher number of pixels. Phil Lippincott said to me that he decided against this because he felt the other components could only do 8,000 and he would rather have delivered 8,000 real samples vs some other result, theoretical or otherwise. The long winded point I am trying to make is that there are two sets of numbers. One is the theoretical capabilities of the scanner and the other is the real resolution in real terms as measured by say, a piece of film.

I had occasion to work on an old Hell 299 (I think that's the number) and was probably the first person to make enlarged negs for platinum printing with a scanner - back in 1980. I'm familiar with that scanner. I also am familiar with the 3400 series, and their scans aren't up to par with later drum scanners.

I was comparing X5 scans with my own ICG scans at the highest optical resolution of 8,900 spi. I really have to say that the X5 was on par. I don't know if a Premier could do so much better when even the X5 is quite close to its 8000 spi spec. If you could show some sample how a Premier scan looks at 8,000 ppi maybe the difference compared to other scanners is obvious.

I have an assistant who did a lot of work with 4x5 scans on an Imacon before he found me and we spent some time comparing the scans. They aren't bad. But they aren't drum scans. They were a bit soft. I am curious as to what ICG you have. I'd love to see a scan off a 380... just to see how good it is.

I am happy to supply some samples. I'm not sure what format. It appears that one doesn't see it unless one is looking at the whole thing.... As much as I appreciate the efforts here on the scan comparison, I find it very difficult to make real sense out of it. When things are compressed that much, the differences disappear.

May I ask where you got those number about the 20K steps per PMT channel?

Haddon Stevens Lippincott. Phil passed away a couple of years ago and Haddon is now the head of the firm.

One thing though that made me wonder was a comment by Karl Hudson in the ScanHi-End newsgroup where he made a statement about the Tango being able to resolve down to 4 microns with a 15.9 micron aperture.

This is what's frustrating. There are 24,500 microns in an inch, roughly. One might not imagine that at 15.9 microns you would not resolve to 4,000, but only 1597.5. The micron settings are different for different scanners. I usually do chromes and b&w negs at 13, they start to get soft at 16. I wouldn't claim that I could do my best work at 16. But Karl might.... because the way the settings they have things are different. I have been told that it is fixed at 11 microns, then told it isn't. I don't know Karl personally and have no reason to doubt him.

On the other hand, some would say that you resolve on the amount of microns you have and I would say not. When I scan at 13, I shouldn't be able to get more than 2,000 resolution on anything. But that isn't how it works. I get much more. As you stated, one can't scan at 3 microns or you will anti-alias. But how does one explain the results? Either there is no more after 2,000 or there is a lack of understanding on the actual process. I would suggest the latter.

If you're working with 16 bit/channel, is there any advantage applying the tonal separations in the scanner software compared to what you could do with a raw scan in Photoshop?

I think there is. Especially with DPL. Can I prove it? Probably with time I could. However, I can tell you that if I get things right in the scan, things separate well and the prints are fairly easy to make. I just did something off a raw scan and I ended up with 14 adjustment layers. There is a huge loss in going to Photoshop. It's going from 20K separations down to 250. If you give it tonal ranges that are separated in the way you want it to be, there is a real benefit when the conversion happens. I am stating my opinion, rather than stating it as a real fact. It's whats worked for me so far. I'm still learning.

Lenny

[SCHWARZZEIT]

Sure. They didn't. For starters, I don't mean to pick on you, or make you defend yourself. Just in case you feel that way, this isn't personal at all.

I didn't take it as personal offense. On the contrary, I do welcome real honest opinions, and hopefully we can all learn from each other.

I used to be very upset at Imacon's quoting of certain numbers that were actually out of range. They suggested their DMax was 4.8, I believe, which exceed the max on film, of 4.6. How did they actually get that result? I railed against them in forums for lying to us about their capabilities. That is, until one fellow took the time to explain to me how they got their result. There was a mathematical formula based up the probable capabilities of such a device with its components, etc. It was theoretical number. They were quoting theoretical while others were quoting tested values.

Yes, it is the theoretical Dmax value for any 16 bit/channel workflow. It doesn't take into account any CCD noise issues. I haven't tested a Flextight scanner for real Dmax. I heard they are doing quite ok compared to the Nikon Coolscan film scanners. FWIW, a client of mine who owns an Imacon 646 told me that the drum scan I did for him of a Velvia 50 slide had much better shadow detail than he could get from his Imacon. But I haven't compared it myself.
Aztek has an honest approach by giving the real Dmax value in their public specs for the Premier.

It got worse, when Epson suggested it can do 6400 real resolution, which we all know is too high. Now its all just a mess of one kind of disinformation vs another.

I see your point, and of course I'd prefer the manufacturers to state some true optical resolution for their machines but it seems they rather give the specs for their stepper motors and let the customers in the dark. But drum scanner manufacturers are no real exception on this.
So in the end we have to see for ourselves how these machines perform and draw our own conclusions if the quality is worth the money.

I had occasion to work on an old Hell 299 (I think that's the number) and was probably the first person to make enlarged negs for platinum printing with a scanner - back in 1980. I'm familiar with that scanner. I also am familiar with the 3400 series, and their scans aren't up to par with later drum scanners.

These scanners were probably made for a different set of computer hardware. It's only for a few years since we're able to process huge files. It's still very time consuming.

I have an assistant who did a lot of work with 4x5 scans on an Imacon before he found me and we spent some time comparing the scans. They aren't bad. But they aren't drum scans. They were a bit soft.

I guess there are differences between the newer and older Flextight models.

I am curious as to what ICG you have. I'd love to see a scan off a 380... just to see how good it is.

I have a 370HS. When I bought the scanner from ICG I asked them about the difference between the 370HS and the 380. I was told that the 380 has a slightly faster top speed at 2,000 rpm on lower res scans. But I wouldn't see a difference in the quality of the scans. Exactly the same optics and components are used to build both machines. This came directly from ICG.
I could supply some scan samples if you like.

I am happy to supply some samples. I'm not sure what format. It appears that one doesn't see it unless one is looking at the whole thing....

I think a crop where you personally think that it's representative in showing how sharp the scanner sees the grain at 8,000 spi would be great. But I know what you mean that it's more impressive to see the whole scan.

This is what's frustrating. There are 24,500 microns in an inch, roughly. One might not imagine that at 15.9 microns you would not resolve to 4,000, but only 1597.5. The micron settings are different for different scanners. I usually do chromes and b&w negs at 13, they start to get soft at 16. I wouldn't claim that I could do my best work at 16. But Karl might.... because the way the settings they have things are different. I have been told that it is fixed at 11 microns, then told it isn't. I don't know Karl personally and have no reason to doubt him.

On the other hand, some would say that you resolve on the amount of microns you have and I would say not. When I scan at 13, I shouldn't be able to get more than 2,000 resolution on anything. But that isn't how it works. I get much more. As you stated, one can't scan at 3 microns or you will anti-alias. But how does one explain the results? Either there is no more after 2,000 or there is a lack of understanding on the actual process. I would suggest the latter.

I'm sure there there's something else than just interpreting the aperture size as the smallest resolvable spot. The aperture is a hole in a disc, so how does the micron number describe the hole? It could be the radius, the diameter or something else that defines the spot size. I asked this question several times and was always told that it is the diameter. If it is the diameter of the analysis spot than the smallest resolvable detail could be as fine as half that size because the spots can overlap. This way smaller details are resolved with some loss in contrast, just like diffraction spots.
Another approach could be that the analysis spot is optically magnified before it is projected onto the resolution aperture, and the solution is probably different for each manufacturer. It seems that we cannot draw conclusions on the absolute resolving power of the scanner just from the aperture size.
BTW, in Phil Lippencott's scanner test the Tango came out with an optical resolution of 4096 ppi.

I think there is. Especially with DPL. Can I prove it? Probably with time I could. However, I can tell you that if I get things right in the scan, things separate well and the prints are fairly easy to make. I just did something off a raw scan and I ended up with 14 adjustment layers. There is a huge loss in going to Photoshop. It's going from 20K separations down to 250. If you give it tonal ranges that are separated in the way you want it to be, there is a real benefit when the conversion happens. I am stating my opinion, rather than stating it as a real fact. It's whats worked for me so far. I'm still learning.

I think the difference may be that some corrections can be done more easily in a dedicated scanner software beforehand. Thus it's much easier to get to the print file from a preadjusted scan than working from a raw scan file. On the other hand I like the idea that I have the full unmodified gamut of a raw scan to work with. I only made a few tests where I compared a preadjusted scan with a raw scan after both files went through Photoshop. The results were very close, and I couldn't really decide which I liked better.
It's not that you have only 250 separations per channel in Photoshop. Internally a 16 bit file has roughly 16K values per channel. If you have 20K separations to work with in DPL then your Premier must have a greater than 16 bit A/D converter. Otherwise you would lose some steps during A/D conversion.

-Dominique

[Lenny Eiger]

So in the end we have to see for ourselves how these machines perform and draw our own conclusions if the quality is worth the money.

Yes, "you're on your own" is the response we are getting because there is no independent testing agency. Aztek's testing was apparently independent, but it is old. Their efforts were not supported by the rest of the industry, in fact one scanner maker told them to take off the results or they'd be sued. FLAAR, which attempted to be such an agency, or at least promote it that way is corrupt. Their endorsements are bought and paid for.

This said, there are a lot of factors. I have a scan sample someone gave me from a Scanmate 11,000 that should not be possible. The samples are too tight for their "stated resolution". Did the fellow do some post-processing - I don't know. The scan project here undertaken by Leigh Perry was an attempt to make a real comparison, but there is quite a bit of downsizing and post processing done. Really mediocre (being polite) scanners show up as quite reasonable on the web.

It's not that you have only 250 separations per channel in Photoshop. Internally a 16 bit file has roughly 16K values per channel. If you have 20K separations to work with in DPL then your Premier must have a greater than 16 bit A/D converter.
-Dominique

They have told me its a 16 bit. Must be the effects of rounding numbers up and down... for discussion.

Regarding raw vs adjusted at scanner scans- one thing the software does (DPL) does that's great is that they have a "CMS" file attached to each scan. This gets loaded into the scanner at scan time and adjusts the range of the scanner. This is very different from doing a raw scan, like most scan software does. The pro version allows you to create your own cms file/profile for each image. It's not needed for most chromes, and only some color negs, but its great for b&w negs. It's like having a different scanner for each image. At least that's what I've been told and the premise I am working on. It appears to be working that way, but there's plenty I don't know...

Lenny

[Tim Povlick]

I am reading this thread with great interest. One point of discussion I do not understand is separation. How does a 16-bit number have less than 20,000 "separations" ?


_ .. --
Tim

[Lenny Eiger]

I am reading this thread with great interest. One point of discussion I do not understand is separation. How does a 16-bit number have less than 20,000 "separations" ?_ .. --
Tim

Tim,
I am curious myself. According to latest Photoshop we are in 16 bit - so there should be 65,536 possibilities, if my math is correct. However, when we move a curve up and down, we only have 0-255 as choices. This would appear quite sledge-hammer-ish as a tool. In terms of b&w, I also don't think my printer can separate out that many tones. My gray ramps are perfectly smooth, but the resolution is low - 1440x1440. So if I only have the tones that can be generated with that resolution, I believe its quite limited. I don't know what the number would be... do you?

Lenny

[Ken Lee]

One has to ask whether any monitor can accurately render that many shades per channel, and whether we could even see them if it could

[Lenny Eiger]

One has to ask whether any monitor can accurately render that many shades per channel, and whether we could even see them if it could

Agreed. The key, IMO, is to get as many shades on the paper as possible. (I try not to imagine the monitor is of any use at all. I only look closely at the print, which is what open would do in a darkroom.) To get as many shades, one needs good ink, and good eyes. It helps a great deal if the scan's zones are separated... in my experience so far...

Lenny

[Ken Lee]

If I recall correctly, printer drivers for inkjet are 8-bit only.

This means that we have lots of room to get things right in the deep color space - before things get collapsed by the printer, the driver, the RIP, and ultimately, by the even more limited gamut of the paper/ink combination.