Does Velvia really hold more detail than other transperency film?

[keithwms]

Marco, thanks for the link to your nice article.

I do think you will see a big difference in top resolution when you compare a drum to any scanner which is lens-based and in which detector noise itself becomes an issue. The difference is especially extreme for 35mm in which the detail is compressed onto a comparably smaller piece of film.

I also doubt very seriously that a lens-based scanner is ever truly scanning into the actual grain. These ppi ratings are massive overestimates for lens-based scanners. I haven't run the resolution numbers myself but I can attest that the scanned grain simply looks quite unlike the grain one sees at large traditional enlargement. I find that the grain in traditional-grain films like fp4+ and hp5+ really enhances the feeling of an image that is printed 100% traditionally, sometimes even giving a sort of 3D effect. But when I scan "into the grain" any format size the grain is very, very different and it just doesn't feel right. People have done detailed analyses of this effect and there is very real doubt that a non-drum scanner ever sees true grain. This is important because for years people have scanned film and said, gee, it's not as "clean" as digital, look how obtrusively grainy the film is.... when all of us know damn well that a straight traditional print looks much better and were deploying our brain cells to try to figure out why. Ah don't get me started!

Another factor, which is something that is rarely discussed, is how much "real" optical information is in the scan, per file size (megabyte not megapixel, because we have to think in terms of bit depth as well as megapixel). A drum scan is, I would argue, by far the most efficient way to pack real information into a moderate file size. With just about any other scanner of which I am aware, you really do need to overscan and sharpen and downsample to compete with what the drum gives in its raw file, per filesize. This is an effect that becomes very important when working with files into the hundreds of megabytes.

Anyway, enough about bytes and bits, anyone interested in all this should look at hybridphoto, and look in particular for the extended discussions with Lenny and Sandy over there. I don't have nearly as much exprience with different scanners as those blokes, but on the other hand I have tried just about everything and I do see a big advantage to drumming. Ultimately one must try all the different techniques and decide for oneself.

 

[Marco B]

Marco, thanks for the link to your nice article.

I do think you will see a big difference in top resolution when you compare a drum to any scanner which is lens-based and in which detector noise itself becomes an issue.

Keith, admittedly I have never used a drum scanner, nor had any film scanned on such a machine. The best scanner I have used is the pretty impressive Imacon Flextight 949 though, which does give very convincing high detail results.

However, I do not entirely understand your remark about "lens-based" versus "drum-scan". If you look at the schematic figure of a drum scanner at page 21 of the article of Tim Vitale that I referenced before, even a drum scanner has a focusing lens. How otherwise would it generate a sharp image???

http://aic.stanford.edu/sg/emg/library/pdf/vitale/2007-04-vitale-filmgrain_resolution.pdf

Or am I missing here something :confused:

I also doubt very seriously that a lens-based scanner is ever truly scanning into the actual grain. These ppi ratings are massive overestimates for lens-based scanners.

I also posted Tim's article to show that actually, what we "conceive" as grain, is a clustering of silver particles to small for the eye and most optical scanning systems to see... It means that even a drum scanner is unlikely to "see grain", in terms of true silver particles, because it needs electron microscopic resolutions to reveal them. Also, the emulsion, as can been seen in the article is actually a 3D matrix of stacked silver particles. As Tim writes (page 3):

"The fundamental image particles (silver), when rescaled into dimensions commonly used for the wavelengths of visible light, range from 200 to 2000 nm (nanometers). The size domain of visible light is 400-750nm; blue light ranges from 380-450 nm; green light ranges 450-550 nm; and red light ranges 550-750 nm. Ultra-violet light ranges 205-380 nm, while infrared radiation ranges 750-5000 nm. Note that the smallest silver particles (0.2-0.8 microns) are not visible unless clumped into larger agglomerates, because most are smaller than the wavelengths of light."

Note that the size ranges are for the yet undeveloped silver halide crystals, Tim doesn't entirely clearly state how big the developed particles are. But if we assume a not much bigger size, which is reasonable since any silver comes from the silver halide crystals themselves, even the drum scanner is unlikely to reveal them...

Actually, in another article I have here at home showing electron microscopic images of silver in developed photographic paper, as opposed to film, the silver "grain" in the paper turns out to be a kind of filamentous silver strings at the molecular and electron microscopic level.

With just about any other scanner of which I am aware, you really do need to overscan and sharpen and downsample to compete with what the drum gives in its raw file, per filesize. This is an effect that becomes very important when working with files into the hundreds of megabytes.

As I tried to explain on my webpage, whether to scan "into-the-grain" or not, is very much a personal preference. I do personally think though, that you are not adding any serious new information and *personally* prefer to scan at a slightly lower resolution, as I called the "non-obtrusive grain" scanning category on my website.

Ultimately one must try all the different techniques and decide for oneself.

I can not do anything other than fully agree with you on this point!

 

[keithwms]

If you look at the schematic figure of a drum scanner at page 21 of the article of Tim Vitale that I referenced before, even a drum scanner has a focusing lens. How otherwise would it generate a sharp image???

By rastering a spot and reading out spot intensity. The role of the lens in a drum scanner is quite different from in other scanners (especially flatbeds). A drum is basically collecting light from a very tight spot focus on the scanned material while that spot is rastered over the film. And note that there is a confocal aperture which further reduces background and enhances signal to noise on the signal collection side. The confocal aperture ensures that you collect signal only from that spot in tight focus.

Let me quote from the lovely article you linked, which says it all quite clearly:

"The major difference is that the lens used in the flatbed scanner must transmit image detail, while the lens in the drum scanner transmits no image detail, just light intensity."

Futhermore the PMT detector used in the drummer offers extremely low noise- another big advantage.

Also, it is important to realize that the drum aperture (which you see in Fig.22 of the article you referenced) is controllable down into the microns. I know that at least for a high-end Aztek drummer, you can optimize that aperture and get clean resolution into the [real] grain at a few microns.

But.... you know, at the end of the day we have to set these technical things aside and compare actual results. And I can attest (as can others) that the output from a top-of-the-line drum scan, when optimally operated, is really in a class by itself... especially for smaller formats in which the ratio of detail to grain is smallest.

As I tried to explain on my webpage, whether to scan "into-the-grain" or not, is very much a personal preference. I do personally think though, that you are not adding any serious new information and *personally* prefer to scan at a slightly lower resolution, as I called the "non-obtrusive grain" scanning category on my website.

For flatbeds and non-drum scanners, I think you are quite right that scanning at higher and higher ppi is not necessarily yielding more information- mostly what you are doing in that case is interpolating. But it can be fairly argued that a high end drum like an Aztek is scanning down to ~3 microns, which really is "into the grain," without interpolation. That is a big difference.

Another thing,which I tried to say above and it perhaps wasn't clear is that the amount of real (noninterpolated) information that you gain by going to higher and higher ppi with a non-drum scanner is progressively smaller at higher ppi. Scanning at 9600 does not give you twice as much real info as 4800; not even close, which agrees with your observations. So you wind up with huge filesizes just to get tiny improvements in scanned detail. This is not so with a drum scanner- the scanned image is pretty damn close to 100% real information, right up the res maximum of the instrument. It is something you just have to behold with your own eyes.

All I can say is, you have to see for yourself. N.b. I am not saying that everybody should drum everything or that this is the solution for everyone. But for 35mm slide, I think the case for high end drumming is particularly strong.

 

[Marco B]

Let me quote from the lovely article you linked, which says it all quite clearly:

"The major difference is that the lens used in the flatbed scanner must transmit image detail, while the lens in the drum scanner transmits no image detail, just light intensity."

OK, you got me there!

Yes, I can see where you are going with this explanation...

Re-reading the article's pages and your comments, as I understand it well, the lens of the drum scanner more or less captures pixels on a "pixel-per-pixel" or "pixel-per-pixelgroup" basis (the latter when the aperture is considerably bigger than than the ppi resolution or pixel pitch as it's called in the article, and each pixel in the group receiving the same RGB values based on the read-out of the PMT).

Opposed to this, a flatbed design, like also the Imacon has, captures an entire "line-of-pixels" at once, as transmitted by the lens in the flatbed scanner, thus, as you and Tim say, "the lens used in the flatbed scanner must transmit image detail".

Futhermore the PMT detector used in the drummer offers extremely low noise- another big advantage.

Well, again I must take your word for this (I believe you ), as I have never seen a drum scan... Although I do think the significance of noise in dark parts of the images is sometimes overrated (I am NOT talking about the horrible 1600 ISO cheap d******l camera's noise, which IS a problem)... With the light amount of noise I have seen in the Imacon scans, there is no real problem, as this noise will never show up in a real world print as it will be hidden by the printers "interpretation / rendering" of the images in ink dots, and that's the ultimate thing that counts for me, not a 100% view on a computer screen.

Still, I think the Imacon Flextight scanners do an impressive job as well, both in terms of resolution and overall scan quality and look. Before going the tedious root of drum scanning, having some Imacon scans made and judging those, may be a more cost effective method for acceptable image quality for most users.

 

[RobC]

Marco,

Let me ask you a question.
Firstly a ccd scanner including flatbed scanners have a fixed size array of sensors. These are in a line and have fixed spacing and a fixed number of sensors in a line. This fixed number determines the native resolution of the scanner.

What do think you think actually happens when you specify a scan resolution of less than that fixed number. i.e. how do you think your scanner actually manages to get from a fixed hardware number to a lower number, and what effect on the detail you extract form a source do you think that process will have?

 

[Marco B]

Marco,

Let me ask you a question.
Firstly a ccd scanner including flatbed scanners have a fixed size array of sensors. These are in a line and have fixed spacing and a fixed number of sensors in a line. This fixed number determines the native resolution of the scanner.

What do think you think actually happens when you specify a scan resolution of less than that fixed number. i.e. how do you think your scanner actually manages to get from a fixed hardware number to a lower number, and what effect on the detail you extract form a source do you think that process will have?

Your suggestion doesn't hold for the Imacon's design, as the whole lens rig is moved up and down for different film formats (and also for different resolution settings, it will just use less pixels on the pixel array, there are a few true "optical" resolution you can set in the scanner's software). As long as you keep to the setting of maximum quality for the film format, no down sampling or interpolation will take place.

Of course, with true flatbed's, things are different.

And in the end, yes, choosing a lower ppi resolution WILL effect image detail, but as said, above about 4000 ppi I *personally* do not like what is extracted as extra "image detail". Again, this is based on taste, if you want to see every detail of dye clouds or clustered silver particles (grain), than it's your call and do whatever you like. It's just not the way how I like to scan... (and of course as Keith has explained and Tim's article shows, and as I am now aware, drum scanners operate differently and can handle "grain" better).

Anyway, due to the fixed size of the pixel array, as you also say, an Imacon can not extract more than 2040ppi from a 4x5 image, 8160 ppi out of a 35mm neg. Still, that compares to a huge 80MP image, which is more than enough for my purposes.