Hype or Fact?

Discussion in 'Digital Cameras' started by Walter E., Oct 28, 2003.

  1. Walter E.

    Walter E. Guest

    I am looking at the fairly inexpensive Fuji A310. ($200 - 300)

    It has 3.1 MP but Fuji proclaims that, by a process called emulation, they
    achieve an effective resolution of 6 MP. Is this too good to be true?

    Also, their new xd memory cards are more expensive, but I wonder what their
    advantage is. Would anyone know?

    Otherwise it looks like a neat little camera.
     
    Walter E., Oct 28, 2003
    #1
    1. Advertisements

  2. Walter E.

    Walter E. Guest

    Walter E., Oct 28, 2003
    #2
    1. Advertisements

  3. Emulation is another term for interpolation. Taking one or more pixels to
    generate another pixel artificially. The best bet is to rely on native
    hardware resolutions.
     
    Joe Thibodeau, Oct 28, 2003
    #3
  4. Walter E.

    mark herring Guest

    YES, it is. See my recent post entitled: "Resolution, Mpixels and
    marketing hype"

    So now they call it "emulation"---before it was interpolation, and several
    other euphemisms. Fuji certainly has some creative marketing, but you cant
    apply ANY post-processing and increas effective resolution

    -Mark
    for private e-mail, say no to "No"
     
    mark herring, Oct 28, 2003
    #4
  5. Walter E.

    JPS Guest

    In message <bnmc2i$gnu$>,
    .... but you can't create a bitmap from a super-CCD with less than 2x the
    pixels, without throwing away detail. In that sense, the 2x inflation
    is justified. It's not as unnecessary and dishonest as a 13.7mp image
    from an SD9; there is no justification for archiving images like that.
    --
     
    JPS, Nov 1, 2003
    #5
  6. Correct. Same for all Bayers, all of which do the same thing Fuji does but
    to a lesser degree.

    Only Foveon produces non-interpolated images.
    Fuji's has a much, much lower optical standard and MP output than the SD-9:

    A 6.2MP Fuji image has 0.78M complete RGB sets = 13% color data per output
    pixel
    A 6.2MP Bayer image has 1.60M complete RGB sets = 25% color data per output
    pixel
    A 13.72MP Foveon image has 3.43M complete RGB sets = 25% color data per
    output pixel

    Ref: http://www.pbase.com/image/22273598
     
    George Preddy, Nov 2, 2003
    #6
  7. Only Foveon produces non-interpolated images.

    Absolutely correct.

    The Bayer sensors _need_ an interpolation process to store pictures formats
    with three-colored pixels (e.g. RGB JPG) of _any_ resolution. (The RAW
    format using single colored entities, not three colored pixels.) Storing
    Bayer Pixels in an YUV format with a lower compression on the brightness
    channel and more compression on the color channels (e.g. possible with JP2)
    would be more appropriate.

    But the Foveon S9 (hardware: three Meg truly three-colored pixels, so the
    RAW format would be a standard (36 Bit ?) TIFF) can store JPG files with
    much more than the three Meg pixels resolution. No information is gained
    here over a three pixel JPG What a nonsense !


    -Michael Schnell, Krefeld, Germany,
    mailto:
     
    Michael Schnell, Nov 2, 2003
    #7
  8. They claim
    "6 M recording pixels, 3.1 M effective pixels" for the S5000 and
    "12.3 M recording pixels, 6.3 M effective pixels" for the S7000

    What should be the interpretation of this ? The D100 or 10D is 6 Meg sensors
    in a Bayer matrix. So I can't believe that the S7000 does 12.3 M sensors for
    half of the price.

    -Michael

    -Michael Schnell, Krefeld, Germany,
    mailto:
     
    Michael Schnell, Nov 2, 2003
    #8
  9. Only Foveon produces non-interpolated images.

    BTW.: Any picture you look at goes through an interpolation process. When
    displaying on a screen or paper you never use the same x/y pixel count the
    RAW format of the camera offers. So it's not important at all how the
    transport format is done. It should preserve as much of the picture taken by
    the sensor as possible. A 3 Meg pixel 36 Bit TIFF file is an appropriate
    format for an S9 Bayer sensor (wasting no bytes on information the sensor
    did dot take in). A 6 Meg pixel 36 Bit format is appropriate for a D100 or
    10D Bayer sensor (wasting 2/3 of the bytes to dummy information, generating
    a far too big file due to the inappropriate file format). All this has not
    much to with the quality of a picture printed with -say- 2 M Pixel or even a
    huge print with -say- 6.5 M pixels, as the post-processing will or will not
    recover the information correctly.

    -Michael

    -Michael Schnell, Krefeld, Germany,
    mailto:
     
    Michael Schnell, Nov 2, 2003
    #9
  10. Walter E.

    Alan Guest

    The problem I see with the Foveon is that the implicit color filters in
    the three layer device are apparently of lower quality than the filters
    built into Bayer masks. Looking at the images, they are apparently unable
    to adjust the spectral response of the layers as well as the individual
    filters can in a Bayer mask.

    I suspect that they have difficulty getting overlapping color response
    curves that match the human eyes, since when a sensor layer is sensitive
    to a color it must absorb light of that frequency, so none of that color
    can be passed on to the layers below.


    Alan
     
    Alan, Nov 2, 2003
    #10
  11. Exactly right. Bayer images downsampled to 25% size will lose no color
    information and per-pixel accurace will greatly improve, though luminance
    will still be only estimated. However downsampling will greatly improve
    overall accuracy in the luminance channels too.
    Exactly right. The Foveon's 13.72MP interpolated output results in 0
    optical resolution gains, just like Bayer interpolation or any type of
    interpolation, it only yeilds a procedurally smoother image. This can be of
    some benefit if you are printing extrememly large sizes. But one thing most
    don't realize is that Foveon does the same for its lower resolution modes.
    For example, a "Medium Resolution" Foveon shot outputs a 6MP image, which is
    both color interpolated and smoothed to the same degree as any Bayer 6MP
    image.

    On a related note, most don't make the connection since it isn't necessarily
    intuitive, that when comparing continuous drive modes and buffer sizes that
    you need to compare the 6MP Bayers DSLRs to the SD-9/10 in Foveon's "Medium
    Resolution" (which can then be similarly interpolated to a 6MP-output file
    size), not Hi Resolution mode, which is instead directly comparable to a
    13.7MP Bayer DSLR.
     
    George Preddy, Nov 2, 2003
    #11
  12. Walter E.

    Ray Fischer Guest

    George is lying. The interpolation does NOT discard any sensor
    information.
     
    Ray Fischer, Nov 2, 2003
    #12
  13. Exactly right. The only thing one needs to note is that you are getting
    3.43MPs of true non-interpolated color information with the Foveon, while
    with Bayer's 75% (double green means 75% waste, not 67%) waste you only get
    an image that contains 25% of the rated MPs, for full color pictures.

    So in your example above, the Foveon is yeilding a true 3.43MPs, while "6MP"
    Bayers only yield 1.5MPs.

    The exception being a B&W target, where Bayers perform much better since any
    of their 1/3rd color sensors can sense black (i.e. no color) accurately, and
    thus color interpolation has no penalty when no color is intrinsic to the
    image. Well, more accurately, it is the Foveon's lack of an interpolation
    process that hurts it in the special case of no color, since 3 primary
    component sensors are combined to form 1 "color" (still black) anyway,
    dumbly.
     
    George Preddy, Nov 2, 2003
    #13
  14. Walter E.

    Ray Fischer Guest

    Which results in staircasing.
    No, idiot, it does not mean anything of the sort. It means that it
    collects more green information.
     
    Ray Fischer, Nov 2, 2003
    #14
  15. It doesn't "discard" it, it uses it to almost no benefit. Each RGGB set
    results in a single pixel color (output as 4 interpolated pixels all with
    colors centered on the one color), but instead of simple R+B+G color
    building you are left with a sensor wide average process of R+B+2(G/2),
    which obviously gives a tiny benefit in per pixel green accuracy, but that
    comes at the expense of 25% of the sensor count which is effectively
    orphaned. The tradeoff is obvious a very, very poor one, but when you have
    4 scalable sensor positions per full color and only 3 primary colors, you
    have no choice but to duplicate a color.
     
    George Preddy, Nov 2, 2003
    #15
  16. Optical perfection = stairstepping using any finite 2DF grid
    Why trash your own credibility? Weird.
    Then tosses it during its interpolation process, to avoid an overly green
    image. Obviously. R+2(G/2)+B is a very poor tradeoff for a 25% sensor
    loss.
     
    George Preddy, Nov 2, 2003
    #16
  17. Walter E.

    Ray Fischer Guest

    George is lying.
    Each sensor produces a single pixel. Each RGGB set is used to produce
    four pixels by interpolating colors.
     
    Ray Fischer, Nov 2, 2003
    #17
  18. Walter E.

    Ray Fischer Guest

    George has his head up his ass and doesn't know the first thing about
    sampling.

    Colors can vary with nearly infinite frequency (don't quibble, folks)
    over he space of a single pixel. One approach is to average the light
    over the entire pixel area. That produces the best results since it
    isn't subject to high-frequency changes. The worst approach is to
    point sample the color, resulting in inaccurate sampling.

    Think of a computer monitor. Sampling a point of light will produce
    only red, green, or blue. Never any other color because all of the
    points on a cimputer monitor or one of those colors. Sample a small
    area and combine the output from several phospors and you get what the
    eye sees.
    So you're childish as well as a dishonest asshole.
    You're lying.
     
    Ray Fischer, Nov 2, 2003
    #18
  19. So in your example above, the Foveon is yeilding a true 3.43MPs, while "6MP"
    I did not say this and it can't be concluded from what I said. It's
    definitely wrong.

    -Michael Schnell, Krefeld, Germany,
    mailto:
     
    Michael Schnell, Nov 2, 2003
    #19
  20. Check the archives of this group using Google. We've discussed this
    several times before.

    The basic summary is that a standard square-grid sensor actually has 40%
    higher resolution in the diagonal directions than it does for horizontal
    or vertical detail.

    Fuji rotates the sampling grid in the sensor so the pixels are aligned
    on 45 degree diagonal lines. This gives 40% more resolution for
    vertical and horizontal lines with the same number of pixels, but of
    course the diagonal resolution is now reduced to what the H/V was
    before. There's no additional information; just the orientation of best
    resolution has changed.

    However, Fuji has to output a standard image file with row/column grid
    for use in standard software. In order to maintain the extra H/V detail
    captured by the rotated sensor, the image has to be interpolated onto a
    standard row/column grid without reducing pixel horizontal or vertical
    spacing. That's done by interpolating one new pixel halfway between
    each existing pixel in each row (or column, depending on how you look at
    it). The result is a standard square grid, but with twice as many
    pixels as the sensor itself has. That's how you get a 12 MP image from
    a 6 MP sensor. You can also ask the camera for a 6 MP output image, but
    it doesn't retain the extra H/V resolution that the sensor captured.

    Dave
     
    Dave Martindale, 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.