Sigma SD10 Fringing

Discussion in 'Digital Photography' started by Mike Engles, Oct 29, 2003.

  1. Mike Engles

    Mike Engles Guest

    Hello

    I do not own a digital camera, but I follow these arguments closely.

    I have beem looking at images from the SD10, both on the Sigma site and
    DPReview.

    On DP review image 17 from the right has curious Magenta/Green fringing
    on the wet stones at the bottom. Also in the trees at the top right.

    http://img.dpreview.com/gallery/sigmasd10_preview/originals/IMG07171.jpg

    The same effect can be seen on the stones in image 21 bottom.
    http://www.foveon.com/X3_images_21.html

    Any idea what this might be? It looks like a lens problem.(micro lenses)


    I have to say that in general I am most impressed by the sharpness of
    Sigma images. Dodgy colour can be corrected to some extent.
    As far as I can see Canon images are superficially good, but when looked
    at closely are soft and blurred.
    At the moment I will stay with my Nikon/Olympus film cameras and Nikon
    LS40 scanner.

    Mike Engles
     
    Mike Engles, Oct 29, 2003
    #1
    1. Advertisements

  2. Mike Engles

    JPS Guest

    In message <>,
    With sharp lenses, Canon DSLRs have enough sharp detail to revive with
    sharpening in software. The trick is to use a tiny radius, and a high
    strength to the sharpening; stuff like 0.5 pixels @ 300%. I see people
    talking about radii of 5 pixels, or high-pass filters of 10 pixels
    overlayed on the images; that's why most people assume that sharpening
    these images is just a gimmick. Sharpening lower frequencies like that
    generally *is* a gimmick.
    --
     
    JPS, Nov 1, 2003
    #2
    1. Advertisements

  3. Even fairly cheap D/SLR lenses are razor sharp, as is easily demonstrated
    with low ISO film. If your DSLR sensor cannot image basic sharpness (none
    of the 6MP DSLRs have enough sesnor resolution to image the sharpness
    obtainable with average 35mm glass), spending money on pro lenses is a
    complete waste of money--unless you are after more obscure lens
    characteristics like speed or richness of color. Though, interestingly,
    brightness/color is one area where digital can genuinely help, as color is
    fixable, optical sharpness is not. Optical sharpness cannot be created or
    restored digitally without doing major irreversable damage to the image.
     
    George Preddy, Nov 1, 2003
    #3
  4. Mike Engles

    Chris Brown Guest

    The artifacts you are seeing are caused by two different things. The cyan in
    the trees, against the blown out sky is sensor blooming, and is caused by
    charcge leakage from a blown-out pixel into adjacent pixels. All digital
    cameras suffer from this to some extent.

    The purple highlights are almost certainly due to the colour interpolation
    performed by the Sigma/Foveon software that converts raw camera images into
    something usable. I've been experimenting with raw SD9 files (I don't expect
    the SD10 would be hugely different), and this sort of artifact is not
    present in the unprocessed images, AFAICS. However, unprocessed raw data
    from an SD9 (again, haven't looked at SD10 files, but I'd expect the issues
    to be similar) is somewhat soft and the colours are very muted. It's a
    little optimistic to say that these sensors record red, green and blue at
    each point - the red channel appears to contain quite a lot of green, the
    green channel seems to contain a lot of red and blue, and the blue channel
    seems to contain significant amounts of green, and to a lesser extent, red.

    The raw conversion software (Sigma Photo Pro) performs colour interpolation
    and sharpening on the raw images, but it seems that sometimes it gets the
    colour interpolation wrong, and the odd purple highlights you noticed are
    one result.

    I'm working on an alternative flow which involves processing the raw sensor
    data in Photoshop, using the channel mixer to try and "correct" the colour.
    This leads to slightly less oclourful results than the Sigma software, but
    the resultant images are less noisy, and contain fewer artifacts. I'm hoping
    to post details of my results here and on dpreview soon, but I'm doing this
    in my spare time and am having to rely on donated Sigma raw files, so it's
    not the fastest process in the world.
    Would you be terribly disappointed to hear that they're actually quite
    aggressively sharpened during raw conversion? My experiments suggest that
    the amount of sharpening produces a visual effect similar to unsharp mask
    settings of 0.5 pixels, 200%, threshold 0 in Photoshop. The raw data out of
    the camera is a lot less sharp (looks a lot like unsharpened Canon output to
    my eye).
    Try running them through Photoshop unsharp mask using the settings described
    above - they suddenly look a lot like SD9 images at the pixel level. The
    sharpness of the images from different DSLRs would appear to be almost
    entirely down to decisions taken in raw convertors.

    One should also not confuse sharpness with high resolution. An image with
    abrupt high contrast luminance changes on edges will look sharp, but may
    well actually contain less image detail than a similar image which doesn't
    look as sharp. Sharpening an image to make it look more detailed is a bit
    like turning up the volume on an amplifier to make it "sound better".
    Ultimately, however, what looks good in your desired output medium is what's
    important. If you don't want to do any post-processing on a picture, and
    prefer images which look sharp out of the camera, but don't mind the issues
    with colour that you mention in your post, then the SD9/SD10 may well be
    good choices. If you intend to process the majority of your images before
    use anyway, sharpness out of the camera isn't actually very interesting -
    any DSLR can produce images which look just as sharp as SD9/SD10 images with
    trivial use of Photoshop.

    If you want to seen an example of what I've been talking about, have a look
    at my post on dpreview:

    http://forums.dpreview.com/forums/read.asp?forum=1027&message=6473768

    That contains two fragments of images - one processed using Sigma's software
    and one where I have colour-corrected and sharpened the image myself, using
    Photoshop. You can see the raw camera output for the same image in the
    following post. Notice how it's actually a lot less sharp, and the colours
    look rather muted. That is what the SD9 sensor actually "saw" for that
    image.
    Good luck with whatever purchase you end up making. If you have a good
    collection of Nikon lenses, however, you could do a lot worse than look at
    the D100. My view is that, ultimately, the lenses you end up investing in
    are a far more important consideration than the body (which you'll probably
    end up replacing in a few years anyway) in a DSLR system. Any differences
    between the image quality produced by the SD10, *ist, D100, 10D and S2, are
    ultimately dwarfed by the differences between the lot of them and
    zoom-compacts, or even a good and a bad lens at the end of the day.
     
    Chris Brown, Nov 1, 2003
    #4
  5. Mike Engles

    eawckyegcy Guest

    eawckyegcy, Nov 1, 2003
    #5
  6. Mike Engles

    Chris Brown Guest

    I don't think what the OP has noticed is chromatic abberation - it looks
    much more like sensor blooming.
     
    Chris Brown, Nov 1, 2003
    #6
  7. Mike Engles

    Samuel Paik Guest

    I think it is a software problem in the raw to jpeg conversion.
    Aliasing is a major enemy of good digital imaging. If you do not low
    pass filter to remove frequencies above the Nyquist frequency before
    sampling, you will have aliasing and it is irreversable. Aliasing
    converts signal frequencies above the Nyquist frequency into signals
    below. This "creates" "detail" which does not exist in the scene.
    Stairstepping on edges is one example of aliasing. Moire is another.
    False textures can be created by aliasing.

    Several people have noted that a picture of a ruler taken by a Sigma SD9
    shows 9 millimeter marks per centimeter, rather than ten (if there is
    sufficient resolution in the imaging device) or indeterminate (if
    there isn't enough resolution). If you look carefully at the SD9
    resolution chart on Digital Photography Review, you can see where the
    resolution lines go from the correct number to the wrong number before
    turning into indeterminate. THIS IS ALIASING CAUSING THE WRONG IMAGE.

    The Canon EOS-1Ds is known to have a relatively weak antialiasing filter.
    Luminous Landscape's review includes a picture where the 1Ds got the wrong
    image.

    <http://www.luminous-landscape.com/reviews/cameras/1ds/1ds-field-5.shtml>

    What you'll see is that the the way that the 1Ds reproduces this
    wall appears to be quite a bit "crisper" on screen. On large prints,
    frankly, it clearly appears to show higher resolution than medium
    format.

    This turns out to not be so. After some very careful last-minute
    scrutiny (just to be sure) we realized (it was Thomas who first
    saw it) that the apparent higher resolution of the 1Ds was in fact
    caused by a form of artifacting. Somehow the frequency of the line
    pattern on the building was "beating" with that of the chip, and
    so what we were seeing was a sort of doubling of the texture width
    with this particular subject.

    This is not to say that bayer mosaics are good while Fovean is bad.
    Demosaicing a bayer is not trivial and artifact free. Bayer sensors
    also have a harsh resolution vs. aliasing tradeoff, since an antialias
    filter strong enough to significantly reduce aliasing in each of the
    color bands would also reduce the luminance resolution significantly.

    Sam
     
    Samuel Paik, Nov 1, 2003
    #7
  8.  
    George Preddy, Nov 2, 2003
    #8
  9. Mike Engles

    Todd Walker Guest

    USM with 300%? You must LOVE halos.

    ________________________
    Todd Walker
    http://www.toddwalker.net
    Canon 10D
    http://www.toddwalker.net/canon10d
    ________________________
     
    Todd Walker, Nov 2, 2003
    #9
  10. Mike Engles

    Flycaster Guest

    It totally depends on the image and the output device. 200-300% with a
    radius of 0 .3 - 0.5 pixels is the *recommended* amount for enlargements
    going to Lightjets; check out the Calypso printing FAQ. Even for an inkjet,
    if this is applied to the luminance channel (edit-fade-luminance), this
    would be appropriate *for the right image.* The radius is really the key.
    IMO, anything over approx. 1.5 is too much and will artifact like crazy
    unless you are taking a picture of, say, a pumpkin.

    Last, he didn't even mention the other mitigating setting of threshhold, or
    the fact that USM can be applied selectively, and in many, many different
    ways.
     
    Flycaster, Nov 2, 2003
    #10
  11. They're probably tired of people blaming image blur and excessive color
    interpolation artifacts on their extreme high end printing process. The
    truth is, even a non-interpolated Light Jet will produce a blurry, spatially
    and color interpolated image if the image itself is blurry and spatially
    plus color-interpolated in the first place, and all Bayers are.

    The SD-9 and SD-10 are the only DSLRs in existence capable of producing a
    non-interpolated color photograph from start to finish. So
    non-interpolated Light Jets make perfect sense for Sigma prints if you want
    rough medium format parity, and little sense for other DSLRs for which color
    interpolation limits quality to 35mm parity when using the double-digit
    (though that is 4X spatially plus color interpolated) megapixel cameras.
     
    George Preddy, Nov 2, 2003
    #11
  12. Mike Engles

    Flycaster Guest

    The CS Lightjet 5000 USM values I quoted are for all images, regardless of
    capture device, or film size. (And if you think your Sigma can "roughly"
    match a 4000 dpi drum scan of a MF transparency - I just cannot believe you
    said "medium format parity" - you are wrong. Calypso and the other high-end
    service bureaus I use do very little Lightjet work that originates as a
    digital image, and that's because most of their clients still use film...for
    a reason. Boy, give it a break.
     
    Flycaster, Nov 2, 2003
    #12
  13. Mike Engles

    JPS Guest

    In message <>,
    I hate halos.

    There are no halos at 300% when the radius is tiny. Halos happen at
    bigger radii.
    --
     
    JPS, Nov 2, 2003
    #13
  14. Foveon rates the SD-9/10 at roughly medium format film parity (56mm) when
    enlarged less than 40 inches on the long side. It is pretty well
    established that color-interpolated DSLRs can only achieve rough 35mm parity
    at the highest end (professional DLSRs only, meaning roughly 10MP and
    higher).
    The reason is that Light Jets don't interpolate color, that is their main
    advantage. So obviously it makes no sense at all to print an
    already-color-interpolated digital image using one, if you want to reap the
    full quality advantage of true 36-bit color. The SD-9/10 is currently the
    only digital option for true non-interpolated 68 billion color phtotographs.
     
    George Preddy, Nov 2, 2003
    #14
  15. Mike Engles

    Flycaster Guest

    phtotographs.

    If you actually believe that marketing BS that Foveon promulgates, what can
    I say? A forty inch enlargement of *any* digital camera, or digital back
    (including Sinar's latest) will not match film. Not yet, no way, no how.
    The "color interpolation" is NOT the issue, it's the total data. Run the
    pixel count numbers, not the bit depth...you're off on yet another
    nonsensical tangent.

    There are so many obvious errors, or misunderstandings, in what you've said
    above that I'm stunned. You *could* learn about this stuff, if you wanted
    to, but I'm not at all sure you are interested in anything but your
    misinformed opinions and regurgitations. I'd be glad to give you some
    reference books and some phone numbers if you change your mind.
     
    Flycaster, Nov 2, 2003
    #15
  16. Using color interpolated Bayer you are 100% correct.
    Send them to Foveon instead, they are the ones parading the SD-9 wall sized
    emulsion photographs around.
     
    George Preddy, Nov 2, 2003
    #16
  17. Mike Engles

    Tilman Kolk Guest

    I´m looking for some good (book) references on the techniques of
    digital imaging (interpolation, color, white balance, film comparison ....).
    ´cause I am new in the digital shooting area.

    Any suggestions for reading ?

    Thanks, Tilman.
     
    Tilman Kolk, Nov 2, 2003
    #17
  18. Mike Engles

    Flycaster Guest

    Sure, be happy to help, but most of the reference books that I use (almost
    daily) are Photoshop oriented. Since my background is film and most of my
    product used to be MF and LF landscape images, with the advent of high end
    digital wet-printers accompanied by the decline in traditional cibachrome, I
    really had little choice but to immerse myself inside the computer end of
    the process. If you really get bitten by the bug, that's probably where you
    will end up. ;)

    Digital imaging (with a Photoshop emphasis):

    "Real World Photoshop" - Blatner and Fraser; (superb color management
    section, and more than you'll ever want to know about Photoshop, written in
    layman's terms)

    "Photoshop for Photographers" - Martin Evening; (not as in-depth as "RWP",
    but still a classic reference book.

    "Photoshop Artistry: Mastering the digital image" - Haynes and Crumpler; an
    *excellent* starter book, complete in its own right, with a great series of
    tutorials.

    "Mastering Digital Printing: The Photographer's and Artist's Guide to
    High-Quality Digital Output" - Harold Johnson; the title is fairly
    self-explanatory. Our Texan friend, George, would not like what Harold, one
    the acknowledged masters of this field, has to say.

    "Professional Photoshop" - Dan Margulis; another classic reference manual,
    especially if you ever want to understand RGB-CMYK workflows.

    "Photoshop Color Correction" - Michael Kieran: somewhat like Margulis' work,
    but without the emphasis on CMYK.

    Last, in case you have no intention of going to Photoshop, but rather would
    like to stay with something less expensive and yet still good, here are two
    books for Photoshop Elements:

    "Photoshop Elements 2 Solutions" - Mikkel Aaland

    "The Hidden Power of Photoshop Elements" - Richard Lynch
     
    Flycaster, Nov 2, 2003
    #18
  19. Mike Engles

    Flycaster Guest

    I *have* seen 24x30's made from a number of prosumer digital cameras,
    including Nikon (D-1 and D100), Canon (G-3, 10D, and 1Ds), *and* Sigma (SD9)
    while up at CameraWorld in Portland. Compared to what I'm used to seeing
    from drum-scanned MF film output, the 1Ds output was (at best) just OK, and
    the rest looked like CRAP.

    You cannot print an image from *any* of them at that size without high
    multiple interpolation. Do the math: at full printer resolution and 8 bit
    color, a 24x30 requires a file size of 200MB. Surely you can understand the
    problem here - or has this crusade you are on negated your ability to add,
    subtract, multiply, and divide?

    Do you work for Foveon, Sigma, or for *any* company that might directly or
    indirectly benefit from your efforts here?
     
    Flycaster, Nov 3, 2003
    #19
  20. Probably ink.

    I said photograph.
    No.

    I hope some others can benefit from getting non-commercialized information
    though. Canon is the worst, giving pre-release perks to only pros who give
    them favorable reviews. Perhaps not coincidentally, Canon also provides by
    far the worst DSLR values. Well built (except the 300D, which is a
    below-prosumer class build quality embarrasment), but with very poor
    resolution and per dollar spent.
     
    George Preddy, Nov 3, 2003
    #20
    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.