DIGITAL vs. FILM (Round 2)

Discussion in 'Digital Photography' started by Annika1980, Nov 22, 2003.

  1. This is a crucial point, I think. At the risk of sounding like a
    doubting Thomas (sorry zuuum), would other members in this NG (in
    particular those who know what they are talking about...) agree with the
    statement that scanners (eg. the Nikon 8000) do not interpolate colour?
    Does the Canon 1ds camera interpolate colour? To me, interpolation might
    be a good thing but only in extremely small doses, like arsenic when it
    is used as a medicine. However, these small doses seem to be often if
    not invariably grossly exceeded in practice. Would others agree with the
    latter statement as well?
    nobody nowhere, Nov 30, 2003
    1. Advertisements

  2. Scanners do not interpolate color. The Nikon 8000 has four LED lights (RGB +
    IR) and makes 4 exposures on a linear monochrome CCD sensor with the film in
    the same place, so unlike scanners that have to move the film between each
    RGB exposure, RGB and IR are guaranteed to be in exactly the same place. On
    the other hand, scanned images are grossly soft at their native resolution,
    so the actual optical resolution you get is 1/2 to 1/4 their nominal

    Here's one of the sharpest, most noise-free scans you will ever see. The
    image on the left; the image on the right is with the scanner defocussed.
    (Velvia 100F, Mamiya 645, 35mm lens, Nikon 8000)

    If you compare the same number of pixles at the same magnification on the
    screen, it's soft fuzzy garbage compared to what people get from Bayer dSLRs
    every day. Since that's a crop from a 6100 x 8800 pixel (53 MP) image, it'll
    still make a nice print. (IMHO, those 53MP are worth 18MP of 10D pixels. But
    note that this requires 2.5 times as much film area as 35mm provides.)
    This unhappiness and unease with "interpolation" is completely misplaced.
    Bayer sensor cameras manage to pack as much luminance information
    (resolution) into their pixels as it's possible to without intoducing
    artifacts, and as much color information as your eye can see.

    Although possible, it's quite difficult to get as good images out of 35mm
    film as people get from their dSLRs every day. Unless you use films of
    Velvia 100F quality, 35mm is much worse than digital. Color negative films
    are hopeless in 35mm.

    David J. Littleboy
    Tokyo, Japan
    David J. Littleboy, Nov 30, 2003
    1. Advertisements

  3. Annika1980

    Rafe B. Guest

    Phil Lipincott rates the LS-8000 at around 2900 dpi optical
    based on USAF test targets. And mind you, he's not trying
    to make the Nikon look good.

    Yeah, you've said that many times, but I've sold a whole
    lot of prints based on scanned 35 mm negatives.

    rafe b.
    Rafe B., Nov 30, 2003
  4. Thank you David (including for the last time, when you put me right on
    the Megapixels/megabytes slight :)-)) confusion), and your reference to
    the very helpful articles on colour management. As usual you have been
    very informative and helpful. However, I am now waiting to hear from the
    prosecution (where is my wig?)... :)

    nobody nowhere, Nov 30, 2003
  5. Annika1980

    zbzbzb Guest

    Although possible, it's quite difficult to get as good images out of 35mm
    Hmm, I always thought Reala was quite awesome.
    zbzbzb, Nov 30, 2003
  6. In 4000 dpi Nikon 8000 scans, it's a lot noisier than Provia 100F. It can
    make nice A4s from 645, but the grain gets real scratchy real quick. (Maybe
    that's why it's called Reala?) It's quite sharp, so I find myself wanting to
    make enlargements larger than the sharp scratchy ugly grain structure
    allows, and getting frustrated and irritated.

    David J. Littleboy
    Tokyo, Japan
    David J. Littleboy, Nov 30, 2003
  7. The conclusion is therefore that the best results from the Nikon 8000
    are from medium format, where the scanner still wins. Just as well,
    because I prefer MF anyway. But I am a little concerned when you say or
    imply that even MF is only good for an A4 print size, if it wants to
    beat 35mm in the dslr v. good scanner contest! I wish I could devote
    more time to this, to reach my own conclusions...

    nobody nowhere, Nov 30, 2003
  8. All the scanners I'm familiar with use multiple light sources,
    filters, passes, in some combination, to directly measure each primary
    at each pixel.

    I'm sure that the RGB numbers they report aren't the raw readings from
    the sensors, though; they're modified in accordance with the control
    panel settings, profile selected, etc. That's not what we mean by
    interpolation in this discussion, right?

    Also, if you're scanning at anything other than their optical
    resolution, they then interpolate that data up or down to produce the
    actual results reported.
    And intensity, yes; it's an ordinary Bayer-pattern design.
    This is a theory vs. practice debate. Theory, lightly examined, seems
    to say that Bayer-pattern cameras can't produce the quality of results
    that they do, in practice, produce. Closer examination of the theory
    starts to show where the earlier understanding was wrong -- in
    particular, the higher sensitivity of the human eye to B&W detail over
    color detail.

    But, as always in science, when theory and experiment conflict, we
    know what wins!
    David Dyer-Bennet, Nov 30, 2003
  9. Annika1980

    Rafe B. Guest

    Bear in mind that David L is... shall we say... a bit more
    sensitive to film grain than most of us.

    Personally, film grain isn't one of my biggest peeves --
    even on 35 mm Reala scans printed at 13x19".

    I've captured thousands of images on Tri-X pushed
    to obscene speeds, so frankly the grain of Reala is
    a non-issue.

    I've admired the photos of Cartier-Bresson and
    Robert Capa since I first delved into photography
    as a kid. To me, grain isn't the root of all evil.

    That's not to say I haven't lusted after the sheer
    perfection that comes from shooting large-format
    chromes. We all draw our limits somewhere or

    rafe b.
    Rafe B., Nov 30, 2003
  10. Annika1980

    Rafe B. Guest

    As an aside: CIS scanners and Nikon film scanners use
    multiple (R, G, B) LED light sources and a monochrome
    (grayscale) sensor.

    rafe b.
    Rafe B., Nov 30, 2003
  11. These scanners can produce a full-colour image without Bayer
    demosaicing, if that's what you mean. On the other hand, you're
    unlikely to ever see the raw data from the scanner. It always undergoes
    a sequence of image processing operations before you see it. If you are
    using automatic dust removal software or you've selected an output
    resolution different from the hardware's native resolution, some
    interpolation is going on.
    There are different kinds of interpolation, and you're not
    distinguishing between them. The algorithms used for Bayer demosaicing
    are a form of interpolation, but not identical to the interpolation used
    for resizing images for example. Demosaicing algorithms necessarily
    accompany the use of a Bayer-type mosaic sensor for colour, but you have
    to compare that as a package against the other alternatives.
    How can anyone agree with something so free of meaning? What kind of
    interpolation do you think is overused? Where? Examples?

    How do you know? Have you compared to a film scanner with a Bayer
    sensor? If not, are you just going on faith that no interpolation is
    better than interpolation?

    Anyway, film scanners have a huge advantage over a general-purpose
    camera: the subject is stationary. Thus, you can use a linear array
    that slowly scans the subject rather than an area array. It's easy to
    get linear arrays with 6000 or 8000 elements, and they're relatively
    cheap, unlike area arrays 6000 or 8000 pixels wide. There are scanning
    cameras that use the same principle, but they're only useful for still
    life or landscape photography - things that don't move. (There are also
    scanners that move the film past a stationary light source and sensor,
    but the limitations are the same).

    To acquire colour, the scanner might use a B&W sensor plus a light
    source that cycles through RGB, or a filter wheel in the optical system,
    or a trilinear array that has 3 rows of sensors with RGB filters in
    front of the rows. All of these give full-colour sensing, but the
    colour images are acquired at different times for each pixel, again a
    method that works only with static images.

    A film scanner has an additional advantage: the subject only *has* 3
    colours of dye in it so you don't have to worry about spectral response
    very much.

    But for a general-purpose still camera, you have to capture the entire
    field of view, in colour, in a single exposure. Thus, all of the
    image acquisition methods used in any scanner I have ever seen or read
    about do not work, and you have to pick something that does work. There
    are only a few alternatives, and so far *for a given cost and size* the
    Bayer sensor seems to work the best. A 3-CCD sensor using prism block
    beamsplitter gets large and heavy and expensive as resolution increases,
    and it places limitations on the lens design too. The Foveon sensor has
    promise, if they can ever solve noise and colour reproduction problems.
    What other choices are there?

    If you're willing to shoot film and scan it afterwards, you have a
    variety of technologies to choose from - but you don't have a digital
    camera. If you want a digital camera, Bayer mosaic sensors are
    currently the sensor of choice for building them.

    Dave Martindale, Nov 30, 2003
  12. Arbitrarily continuous lines, in particular straight lines, which give
    the picture an artificial dimension, not pleasant to the eye -
    presumably, here the camera "interpolates" a few pixels, assuming that
    they should be there anyway (admittedly this is not interpolation of
    colour, which was the subject of the post to which I replied). Or parts
    of a picture which have too many geometrical patterns in them, again,
    looking artificial to the eye, as a repetition of a pattern which was
    not there in reality in such a regular or near-perfect form. Colour:
    (occasionally) artificially strong uniform colours, without subtlety in
    them, again, presumably because the algorithm thinks that this is, say,
    "blue" and "interpolates" (according to my understanding of this term)
    more blue pixels, "for good measure". I have to confess that I have
    never used a canon 1ds or one of the more expensive dslrs, however, the
    few pictures taken with these cameras as reproduced on the Internet have
    not alleviated my concerns in this respect (rightly or wrongly).

    What categories of interpolation are known in the digital photography
    field? Would you please list them briefly, for the benefit of this NG?

    nobody nowhere, Dec 1, 2003
  13. SNIP
    I believe the lynch mob just set fire to it ;-)

    Bart van der Wolf, Dec 1, 2003
  14. SNIP
    Phil is trying to make only 'his' Aztek scanners look good ;-)
    Using the USAF target makes little sense in a serious test, because that
    target relies on alignment of the bars with the sensor array.

    Bart van der Wolf, Dec 1, 2003
  15. Can you point to an example. Frankly, I don't know what you are
    referring to.

    It's also worth pointing out that the sort of interpolation you're
    talking about is very different from the standard kind of interpolation
    used in resizing images. Interpolation used in resizing is a
    well-defined calculation using coefficients (weights) that are carefully
    calculated based on the position and spacing of the output pixels with
    respect to the input pixels, but they *do not depend in any way on the
    actual content of the image*. The algorithm performs exactly the same
    sequence of calculations in the same order with the same weighting
    factors no matter what the image contains.

    Essentially, standard interpolation algorithms assume that the input
    image is a series of point samples of some adequately-smooth continuous
    function (well, 3 functions for RGB), and they fit a continuous
    mathematical function through the data points. Then they can calculate
    the RGB intensity at any point within the image, including points that
    were not input pixel locations. Different interpolation algorithms
    use different interpolating functions that trade off smoothness and
    preservation of fine detail against cost. But they're all fitting some
    sort of smooth function to data points.

    In the description above, you seem to believe that the computer actually
    looks for edges or patterns and "fills in" detail based on these
    guesses. That's not what standard resizing algorithms so. What you
    describe might be called an "adaptive" method. There are also programs
    like Genuine Fractals that effectively guess how to create your image by
    using a sequence of primitive patterns and operations, and that lets
    your image be recalculated at a higher resolution - but how well it
    represents the original depends on how well the first step worked.
    The sort of interpolation I described above is the basis of image
    resampling, which in turn is used to resize images (both larger and
    smaller, with slight changes in method), to warp images to correct
    perspective distortion or lens barrel/pincushion distortion. It's also
    necessary for panorama photography, both in simply warping images so
    that they align along their edges, and also in remapping the image
    geometry from rectilinear to cylindrical or spherical perspective.

    And then there is Bayer demosaicing, which probably uses interpolation
    combined with some adaptive image processing. The camera manufacturers
    don't seem to say exactly what they do internally.

    You might also call Genuine Fractals an interpolation application,
    though I think it's better to call that extrapolation rather than

    I've probably forgotten some, too.

    Dave Martindale, Dec 1, 2003
  16. Annika1980

    andrew29 Guest

    This article is simplistic nonsense: it doesn't consider noise, and it
    doesn't consider contrast. Although Velvia can capture 22 Mpixels,
    they're not very good pixels.

    andrew29, Dec 3, 2003
  17. Annika1980

    gsum Guest

    As a user of both digital and film, I really can't
    understand why film advocates persist in pedalling
    this rubbish. 35mm Velvia/Provia 100 captures
    less information than my D100 - about 4 to 5 mpixels.

    gsum, Dec 3, 2003
  18. Annika1980

    zbzbzb Guest

    As a user of both digital and film, I really can't
    Have you scanned Velvia/Provia with a top rated film scanner or a commercial
    type Imacon and compared?
    zbzbzb, Dec 3, 2003
  19. Annika1980

    gsum Guest

    I've scanned with a scanner that clearly shows film
    grain and defects due to cassete scratches etc. and
    have had 35mm film professionally scanned in
    the past.
    I concluded that the limit for Provia 100 is about
    1800 ppi, beyond which no further information
    can be extracted due to grain i.e. the film was
    the limiting factor rather than scanner.
    Another view of this is to consider the maximum
    size that a 35mm Provia 100 slide can be wet
    printed before grain dominates. I find that
    I can get to 18x10 inches from the D100 with good
    quality but film prints from 35mm look too grainy
    at that size.
    I usually work with 6x4.5 and 6x9cm MF. 6x9
    gives 20 mpixel resolution at 1600 ppi with
    little visible grain. At the moment MF certainly beats
    digital but both meduims have strengths and
    weaknesses and I can't see the point of exagerated
    claims for either.

    gsum, Dec 3, 2003
  20. Annika1980

    zbzbzb Guest

    I've scanned with a scanner that clearly shows film

    I'm sorry but did you mean a 1800 *dpi* scan setting on the scanner? You lost
    me there.
    zbzbzb, Dec 3, 2003
    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.