Getting the most out of the Epson 4990

[Emre Yildirim]

Hello,

I've been reading through the archives but couldn't really come up with concrete answers to some of my questions. I have an Epson 4990 which I'm trying to use for large prints (30x40 being the largest) from 4x5 negatives/trannies. So my questions are:

1) Has anyone made 30x40 prints from this scanner? Some people in this forum believe that 16x20 is the largest acceptable print size. I'm wondering if anyone has made prints that are larger than that.

2) Is there any reason to scan above 3200dpi? or above 2700dpi? I guess what I'm asking here is, will a 4800dpi scan result in more detail/better quality than, let's say, a 2700dpi scan upsized to 4800? What's the 'optimal' resolution that I should be scanning at for the least noise and the most detail?

3) Does Silverfast's multisampling option actually work? I haven't purchased the Ai version yet (still using SE), but if noise is actually reduced by multisampling, I'll be willing to buy the professional version to make use of this feature.

4) Digital ICE - does it degrade image quality? I've heard this rumor somewhere, I'm not sure if it's true. I usually fix hairs, dust, scratches etc. in Photoshop, but if ICE will take care of it without degrading image quality, I should probably use it.

5) And finally, what's the best way to improve sharpness? I usually use USM or Focalblade in Photoshop after I scan. Since this scanner has fixed focus, there isn't really anything I can do physically to improve sharpness. Or is there? Aside from laying the film flat of course.

Any other tips or suggestions are highly appreciated.

[Ron Marshall]

I have the Ai version, and multisampling does reduce noise. But, you absolutely must do a single pass scan before the multi-scan to heat the film; otherwise it will move during the scan.

I will be shooting 5x7 soon and will use mat board to make a frame and tape the negs to it. I will experiment with different thicknesses to improve focus. I have read that taping the film to mat board, instead of using the holder, improves flatness. But I have never noticed any soft areas on my scans.

Scan in color and choose the sharpest channel.

I have only scanned b/w so I have not tried Ice.

[Ted Harris]

Emre, as one of those that doesn't believe this is the right scanner to use for prints larger than 16x20 here is my take on your questions:

1) Acceptable quality of a print is in th eeye of the beholder. What I think is unacceptable you may think is great. You just ahve to make a big one and see how you like it. IMO you can go up to 11x14 and no one will object to the print quality, after tht the arguments start.

2) There really is no reason to scan at 3200 actually as the real opticalresolution of the scanner is between 2000 and 2400. You should experiment in that range. Some will ell you to scan at a higher spi but we have just never seen the value of it.

3) Yes

5) Sharpen only after you have set the final image size in Photoshop.

The best tip I can give you is to work as hard as you can to be sure the image is properly exposed and processed. Just as in a traditional enlargement a negative or transparancy that is created right on th emoney will require far less manipulation.

[Ed Richards]

> 2) Is there any reason to scan above 3200dpi? or above 2700dpi? I guess what I'm asking here is, will a 4800dpi scan result in more detail/better quality than, let's say, a 2700dpi scan upsized to 4800? What's the 'optimal' resolution that I should be scanning at for the least noise and the most detail?

A 4800 DPI scan downsampled to 2400 has the effect of doing a 4x sample for each pixel, which reduces noise. On the Canon 9950, this makes the best scan. You will have to try it on your machine, but it does not have the problem of the possible film movement of doing a 4x scan with four passes. If you use Silverfast, you need to downsample in Photoshop. If you use Vuescan, it will do it for you. My guess is that your native resolutions are 2400 and 4800, and that the intermediate resolutions are not going to be as good as scanning at 4800 and downsampling. The scanner tests seem to indicate that the native resolution of Epson is less than 2400 (the Canon looks like about 2000-2200), so there is no reason to work with a final file higher than 2400.

If you are doing color, a 4800 x 16 bit scan may give you memory problems in Photoshop in a PC, unless you are using Windows 64. That is one of the reasons I use Vuescan - it has good virtual memory management in the current version, so I can do a color scan and get a downsampled file out.

[Emre Yildirim]

Ron & Ted - it's good to know that multisampling actually works. I guess I'll have to buy the Ai version of Silverfast, since Epson Scan doesn't have any of these fancy features.

I just made two scans of a B&W 4x5 (Delta 100) slide using Epson scan. One was at 2400 dpi, the other at 4800 dpi. I used 16-bit grey scale, no sharpening or any other manipulations. The only thing I've done is upsize the 2400 dpi scan to match the same dimensions as the 4800 dpi scan. I've used "bicubic" with no sharpening whatsoever.

After looking at both at 100% magnification I could see no difference. Not even a slight difference. Maybe this will be different if I used color slides, but now I'm having doubts.

Unfortunately I don't have any color slides laying around. I'll probably test some later tonight. It's a real shame that 4800 dpi isn't the real optical resolution as Epson claims. That said, what's the Microtek 1800f's real optical resolution? There seems to be a general consensus that this scanner is better than the Epson.

[Ed Richards]

You are going the wrong way, you should be downsizing the 4800 scan to 2400 to make the comparision. Beyond that, try either Vuescan or Silverfast demos (Silverfast has a nice demo setup), to make the scans - there is no reason to believe that epson scan does a very good job.

[Ellis Vener]

It's a real shame that 4800 dpi isn't the real optical resolution as Epson claims.

I've talked with Epson about this and they say it is a 4800ppi scanner according to their tests and test targets. The talked me about how they test .etc..

My underlying take is that they are testing a completely optimized scanner in ideal conditions , and as we all know the film carriers for the 4990 leave somethign to be desiredas regards film flatness and focus distance. I wonder if film flatness and placement problms result in the lower "real world' resolution numbers people are quoting here have to do with film flatness. Also how wee the "2000-2400 ppi" numbers determined. I'm not say ing they are wrong for a particular scanner and test set up, I am just curious about the methodology these numbers that resulted in these numbers.

Assuming that you are getting best results by scanning at 2400ppi a 3.75 x 4.75 inch negative or transparency. that translates as a 9000 x 11400 pixel image.

Without any interpolation between scan generated pixels and dots of ink layed down by the printer (i.e., 1 pixel = 1 dot ) and printing at 360ppi (optimum for the Epson printers) you'll get a 25 x 31.666 inch print. Is anybody really putting down that much ink for that large of print? Are they really seeing a difference between a 360dpi print and say a 180dpi print at that size at a normal viewing distance? BTW: Printing the same 9000 x 11,400 pixel file at 180dpi yields a 50 x 63.333 inch print.

I am curious about your experiences with these very large prints.

[Ted Harris]

I have had the same conversations as Ellis with both Epson and Microtek. For the 'consumer' scanners not only are the tests done under optimized conditions they are done in a lab setting that in now way resembles real world scanning. Film flatness may have something to do with the numbers but I don't think much. For a detailed discussion of our methodology see the first of our scanning articles in View Camera. We tested real world scanners, took'em out of the box and calibrated them using an IT8 target and then scanned an AIG USAF flexible strate (read transparency) test target. We are doing the same thing now with the Microtek 1000XL and i800.

My take on the resolution numbers is that they are more a factor of advertising hype and the need to keep consumers happy.

The fact is that some scanners, such as the Microtek 1800f, which delivers ~a real 1800 spi, produce much superior scans because of other, less discussed features, such as more accurate stepping motors and better optics. The methodology is important in that there are a lot of 'so called' USAF test targets on the market that are not what they seem. AIG and other companies that produce resolution trgets for the imaging industry do so by precisely inscribing the lines on the base material not by making copies of copies on film, etc. The quality of the target used makes a big difference in the accuracy of the results. I know tht some here use the Edmounds target and since I have neither used nor seen one I can't comment on how it compares to the AIG arget. Suffice to say that it costs 75% less and,while that may mean nothing, it does call a question.

None of this should be teaken as a slam at the 4990 or comperable scanners from Microtek ... they do a fine job with 4x5 and 8x10 materials for smaller prints.

[paulr]

i think there's also something practical about the "true" resolution being around half the sampling frequency (the advertised optical resolution.).

if the optical resolution approaches the sampling frequency, your scans will be subject to aliasing and other ugly effects. drum scanner operators can adjust the aperture size, which allows them to adapt their optical resolution to the detail resolution of the original. this allows them to avoid aliasing. there's no real equivalent in the flatbed scanning world.

offering a much higher sampling frequency than optical resolution kills two birds with one stone: it doubles the marketing hype potential, and it eliminates many aliasing concerns.

i find that my 4800 can resolve about 2400 ppi ... around half the advertised number. this is with a negative wet mounted to float glass (perfectly flat) and set with shims at the optimum focal point, which is around a milimeter off from where it's supposed to be (your scanner may vary!).

oversampling still offers some advantages ... it allows noise reduction with single pass scanning. and 2400 ppi resolution is pretty impressive. there may be some detail on the neg that i'm not seeing, but not much. the scanner certainly sees more fine detail than my apo enlarging lens does, at least at the small to medium print sizes i usually work with.

i don't know if there's much to be learned from usaf test charts. what you're really concerned with is what the final image looks like, which is after it's been sharpened over the course of your workflow. square wave targets do not resemble photographic detail, and so are dubious at best for this purpose. what you'd really need is a sinusoidal test target, and image analysis software that's designed to interpret it. this is all available ... but your own negatives and you eyes will usually tell you what you need to know.

[Frank Petronio]

I've only been using a 4990 for a few days, after two defective Microteks 1800fs (then I gave up). It seems like a fine scanner using VueScan and I might spring for Silverfast. I never trusted multi-sampling or ICE, but I also try to scan clean, well exposed film.

As for the max size, considering that I have 30x40s printed from 35mm TMax 3200 hanging in my house, I'd say just do it. Of course an Imacon or better will be better but unless you decide to buy Imacon scans of your best work, why worry about it? There will always be >i>something</i> better than what you have now ;-)