scanning large format: to the limit (and beyond)

Discussion in 'Digital Cameras' started by Roger N. Clark (change username to rnclark), May 23, 2004.

  1. I thought I would pass on my experience so far scanning large format
    images with my new Epson 4870 scanner. Later, I'll post some comparison
    images so you can see how good this scanner is. So far my experience
    is that it is close enough to drum scans I've had done that I will
    use it for almost all my work. It is not quite as sharp on Fujichrome
    Velvia as the drum scans I've done, but careful sharpening overcomes
    the limit for the most part. However, because it can do 16-bit
    compared to the drum scans, I believe it is better because I can
    recover more shadow and highlight details.

    The problem is that at 16 bits/channel, no scanner software I've used
    can scan the full width of a 4x5 transparency at anywhere near the rated ppi
    of the scanner. My requirements are 3200 ppi minimum (the scanner does
    4800 ppi). In my testing, 3200 ppi gets information important to my
    images that 2400 ppi loses. I've tested Epson scan, Silverfast, and
    Vuescan, all the latest versions. The limit, as discussed in a recent and comp.sys.scanners thread, is due to
    a firmware limitation in the scanner limiting total bytes per line, and
    that limit does not allow the full width of 4x5 to be scanned at 16 bit
    except at 2400 ppi. Vuescan, for example, reads 4800 ppi
    if you request 3200 then downsamples. Epson scan was the one I used.
    I can do 3200 ppi, 16-bits/channel and a 3.4-inch line width.
    With ICE turned on, such a scan takes about 1 hour. Then I move the
    box over the the other half of the image, with lots of overlap, and
    scan a second time, another hour. I make sure the settings are exactly
    the same for the two scans. The resulting images are
    about 14,820 x 11,740 pixels and 1 GByte. This joining procedure
    went well in photoshop CS on a 1.8 GHz win XP box with 2 GB ram
    and 600 GB disk.

    I combine the two halves in photoshop. The intensity levels match
    essentially perfectly: you can't see the join line at all. I have
    found that some, but not all scans mis-register by about a half pixel
    in blocks of a few hundred scan lines, meaning one block will be dead
    on, the next off slightly. I erase the edge of the overlap
    image to so it is not straight, add some feathering, and follow
    darker portions of the image if possible and the the images go together
    without a possibility of finding a join line.

    Then, to really push the limits, I mosaiced two such 4x5 images
    into a panorama. The result: 23,380 x 11,820 pixels. But here is
    where it really became difficult. The combined file size in
    photoshop, keeping them as layers was over 2 GBytes and photoshop
    would not save the file when I tried. Fortunately it did not
    crash either, so I had to feather the join line and merge the
    images before the file size dropped below 2 GBytes. I'm using NTFS
    (file system) so files can be larger than 2 GB, but photoshop
    would do it in standard photoshop format. If someone knows of a way
    for photoshop to save such a file (and read it in again later),
    please let me know. The final image is 1.62 GBytes.

    The problem with mosaicing images is the lens distortions. If someone
    knows of software that will do the mosaicing on such large images,
    please let me know.

    Anyway, Thanks for those in previous threads who helped me work out
    the details and limits of the scanner. I will just have to sigh, and
    scan in pieces! So here is the final image:

    If people want more information on the procedure, I'll be happy to
    provide it.

    Roger Clark
    Roger N. Clark (change username to rnclark), May 23, 2004
    1. Advertisements

  2. Thanks, looking forward to the examples. The resolution numbers you reported
    earlier, seem higher than some of the other users reported. Maybe your
    methods of determination are more accurate, and maybe you also got an above
    average sample from the inevitable quality spread.

    That is actually pretty good for a flatbed scanner. The temperature rise
    when scanning (causing film expansion), play in the CCD/lens assembly, and
    lens distortion, can accumulate to several pixels on other scanners.
    In Photoshop it might be done quicker by setting the layer blending mode to
    "difference" (meaning equal RGB values are black), but maybe that's what you
    already did.

    Photoshop CS seems to be able to do that with the .psb "Large Document
    Format" (up to 300,000 pixels in any dimension, upto 56 channels per file,
    all PS features enabled), although the operating system seems to pose a
    process limit (not file limit as I presumed earlier) to 2GB, maybe Photoshop
    gets around that by dividing between processes.
    Haven't tried it on such big files, but I use the combination of Panorama
    Tools ( and the PTAssembler
    GUI frontend ( It also allows the
    stitching of plan-parallel offset images and morphing the residual
    differences. The author, Max Lyons, did a recompile of Helmut Dersch's
    source code to include some additional functionality, so I'd assume he's
    capable of addressing particular issues you might run into.

    Lot's of dynamic range! It looks pretty good, thanks for sharing.

    Bart van der Wolf, May 24, 2004
    1. Advertisements

  3. Yes, I did use CS. Thanks to you and Bart, I looked inder
    preference, file handling, and found the large format
    check box. I wish I knew about that earlier!

    Roger N. Clark (change username to rnclark), May 24, 2004
  4. Yes. That is the most accurate way in my opinion. It works
    very well.
    Thanks, I now have it enabled. I had assumed the existing psd
    format had been extended, bbut it is a whole new format.

    Thanks, I'll check them out.

    Roger N. Clark (change username to rnclark), May 24, 2004
  5. OK, I checked this out. Pretty cool program.
    The question I have about this, is how secure is it? It seems
    that if you know the structure of the program files, which is easily
    done with one test, then one can go to the image directory and
    simply download all the tiles and reassemble the complete
    image! It tried it and was able to access all of them

    Roger N. Clark (change username to rnclark), May 24, 2004
  6. Roger N. Clark (change username to rnclark)

    gsum Guest

    This is probably a difficult question to answer but, in a nutshell,
    how much better is this scanner than the Epson 3200? I use the
    3200 to scan 6x9 and 6x4.5 MF and have found that it extracts
    impressive amounts of detail but struggles with very dense or
    saturated slides. I had no luck at all with Velvia and usually use
    Provia 100 or Astia 100.

    I overexpose by 1/2 stop and make use of the shadow/highlight
    tool in PSCS. Does the 4870 deal with oversaturated or dense
    slides more successfully than the 3200?

    gsum, May 24, 2004
  7. I missed the beginning of this thread (I see it's heavily cross posted). Just wanted to add
    something about large images: a while ago I developped an app that displays/manipulates large
    image files quickly. It's probably not what you are looking for but might give you an idea. I did
    it mainly to access scientific bitmaps.
    The freeware is online at
    For instance I used it to create the images on
    from a 2Gb image file.
    and there's even a cgi demo at but it's an
    excrutiatingly slow ADSL line.
    Guillaume Dargaud, May 24, 2004
  8. Roger N. Clark (change username to rnclark)

    No One Guest

    Is it possible in PS-CS to print files as large as the 300,000 pixel limit?

    I mean that if I have a file that is 12,000 x 72,000 pixels can I print that
    on my Epson 9600 as a 20 foot panoramic?
    No One, May 24, 2004
  9. Roger N. Clark (change username to rnclark)

    Paul Schmidt Guest

    The problem is a computer one, file pointers are usually signed long
    integers, on a standard 32bit PC that means the pointers can hold a
    maximum value of 2,147,483,647 or 2GB-1. Most modern operating systems,
    can handle larger files, Linux can be setup to use longer pointers, IIRC
    Windows uses a shifting mechanism but, in either case the software needs
    to be built to do this. Considering that even medium format digital
    cameras are only producing 20MB files at this point, 2GB is a reasonable
    software limit. Some of the stitching programs may be able to go
    larger, but your dealing with a pretty unweildy image size, none the less.

    Paul Schmidt, May 25, 2004
  10. SNIP
    Indeed, but the result is rewarding.
    This is another way to achieve resolution with a light kit to travel:

    Bart van der Wolf, May 25, 2004
  11. Yes, and though unweildy today, not necessarily tomorrow.
    It takes me several hours of work to dodge and burn, contrast stretch,
    fix film/dust defects on a large format film scan. I won't go back,
    so I'll do the best I can the first time. I've been doing ~3300
    ppi drum scans of large format for years, starting on a 100 MHz
    Pentium I with 256 megabytes of ram, working with 650 mbyte
    files. It's a piece of cake now compared to then! Once 64-bit
    systems and software are here these images will be easy.

    The 32-bit limit is a real pain. I deal with that at work
    most days, and retrofitting software takes a lot of time.
    I'm hoping for that 10,000 x 10,000 pixel display on my
    wall so I can work on my 25 gigabyte Mars image at one time.
    Its now in multiple pieces and I can see only a postage stamp
    section of it at one time!

    Roger N. Clark (change username to rnclark), May 26, 2004
  12. message SNIP
    You need a bigger screen ;-)

    You probably know this site:

    Bart van der Wolf, May 26, 2004
  13. Roger N. Clark (change username to rnclark), May 26, 2004
  14. Roger N. Clark (change username to rnclark)

    Tom Monego Guest

    It is interesting that his 5ftx10ft mural is only 360ppi, hardly heady stuff.
    The teacher at New York's New School that uses 2 gig files to print 16x20s is
    the one doing new stuff.

    Tom Monego, May 27, 2004
  15. Roger N. Clark (change username to rnclark)

    J Vee Guest

    I don't understand the comment either. I print to approx 50" X 65" with
    a ColorSpan 11 head/color (quad K) printer with each head rated at 600 dpi.
    The company insists that 300 ppi files are the proper size for this printer,
    no matter what the viewing distance. And, actually, the prints are
    absolutely gorgeous even on closest inspection.
    J Vee, May 27, 2004
  16. SNIP

    Cross-posted spam.

    Bart van der Wolf, May 28, 2004
  17. Roger N. Clark (change username to rnclark)

    No One Guest

    Likely the same as spam from the US - the country with no respect for
    anything other than expediency.
    No One, May 28, 2004
  18. One of my pet peeves is "trial ware" that is called "share ware".
    Used to be, "share ware" was something you could try and then send a
    contribution if so inclined, while trial ware was... well... a tiral
    you either had a limited version, or a full version that would time
    out after so many days unless the proper S#, or activation code was
    entered which you got by paying money.

    I have no problem with either, but I wish they would use the old
    naming conventions.

    Roger Halstead (K8RI & ARRL life member)
    (N833R, S# CD-2 Worlds oldest Debonair)
    Roger Halstead, May 29, 2004
  19. Roger N. Clark (change username to rnclark)

    RSD99 Guest

    Agree with that ...

    RSD99, May 29, 2004
  20. Roger N. Clark (change username to rnclark)

    Tom Monego Guest

    Trying to find the name, he wrote a couple of articles in Yahoo's Epson Wide
    Format group stating that he was printing at 720ppi and 1440ppi. Scans 4x5's to
    4000ppi (Aztek drum) and prints at 16x20 monochrome. He teaches at the New
    School in New York and tours with Cone Editions doing workshops. Been going
    through archives etc. trying to get the name. But he is questioning the status
    quo and backing up his theories with work. I find his work interesting and got
    me looking at prints at 240, 300, 360, and 720ppi. I personally saw a
    difference at 240 to 360ppi, 720 just didn't seem that much different and the
    file size got huge.But I'm working at 360ppi more now. Oh yes using an Epson
    7600 with the pieziochrome inksets.

    Tom Monego, May 30, 2004
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.