2 questions regarding Bayer sensors

Discussion in 'Digital Photography' started by Georgette Preddy, Jun 2, 2004.

  1. By now everyone in the group should understand that rated Megapixels
    for Bayer cameras are strictly monochrome, and that full color MPs for
    a Bayer camera equals the rated monochrome-MPs/4. But that is the
    absolute best a Bayer can ever do, actual performance is always a lot
    lower. Here are two interesting questions, you fill in the answers...

    1) Suppose a Bayer sensor takes a picture of a color mosiac, where all
    the colors of the target image are misaligned with the colors of the
    individual sensor photosites of the Bayer RGGB mosiac sensor. So all
    the red photosites always see non-red features of the target image,
    all the blue sensors see non-blue, and all the green sensors see
    non-green. What will the recorded output image look like?

    2) Now suppose the opposite is true, the target mosiac is perfectly
    aligned with the sensor's color mosiac filter, so red squares fall on
    red sensors, blue squares on blue, etc. What will the recorded output
    look like?
    Georgette Preddy, Jun 2, 2004
    1. Advertisements

  2. Zzzzzzzzzzzzzzzzzzzzzzzzzzzzzz
    Randall Ainsworth, Jun 2, 2004
    1. Advertisements

  3. Georgette Preddy

    JPS Guest

    In message <>,
    No, that is total nonsense, fabricated by you.
    Slightly magenta grey, the same overall color as it would if it were
    recorded on a SD9 sensor, and the centers of the colors fell between
    Slightly magenta grey. The SD9 would get it right if the colors were
    aligned with the pixels, but if the ratio were slightly off from 1:1,
    there would be sections of the image solid magenta-grey, and sections
    with the distinct colors. The overall modulation makes for a very
    screwed up image.
    JPS, Jun 2, 2004
  4. Georgette Preddy

    Paul Howland Guest

    Continuing this hypothetical nonsense, assuming you really did want to
    take a picture of this multicoloured chessboard at such a range that
    each square exactly occupied one pixel on a 6MP Bayer sensor, consider
    then what would happen if you took the same picture with the smaller 3MP
    Foveon sensor.

    Hint: The Foveon wouldn't be able to resolve the different coloured
    squares as they'd be smaler than its poorer resolution. The colours
    would be rendered as some god-awful blend of each individual tile.
    You'd get severe aliasing artifacts due to the repetitive high frequency
    image and lack of AA filter. ie. an unmitigated disaster.

    Moral of the story: Don't take pictures of contrived test targets
    designed to confuse sensors, with either Bayer or Foveon cameras.

    Heh, George, why not add a picture of this interesting test target to
    your fascinating photo gallery
    (http://www.pbase.com/imageprocessing/lenscompare) and prove me wrong!
    Paul Howland, Jun 2, 2004
  5. Georgette Preddy

    Tony Spadaro Guest

    Tony Spadaro, Jun 2, 2004
  6. Georgette Preddy

    Bryce Guest

    You are such a loser.

    Does your mother pay your isp fees?
    Bryce, Jun 2, 2004
  7. the image that would result would look remarkably like a little man that
    lives under a bridge... commonly known as a troll.
    Christopher Muto, Jun 2, 2004
  8. Georgette Preddy

    m Ransley Guest

    Read the independent reviews george,

    the independent reviews george
    the independent reviews george
    the independent reviews george
    the independant reviews george
    the independent reviews george
    the independant reviews george

    Since you are learning disabled george i repeated it a few times. That
    is todays lesson for you george. Then write them out and study them
    george , paying special attention to the
    " conclusion " george, and bring us the results george. Yes, bring us
    the results of all the unbiased updated [ not old ] comparisons george.
    Your next job will be with Yugo, tobaco, mold and asbestos, George.
    m Ransley, Jun 2, 2004
  9. Georgette Preddy

    Don Stauffer Guest

    Bayer works because monochrome (luminance) resolution is by far the most
    important to the human eye. The spatial resolution to color changes is
    far less. Most standard resolution tests use a set of black bars on a
    white background as the main feature, even if color bars are included in
    chart. If we have colored bars on a white background, that still creates
    a considerable luminance contrast.

    NTSC color TV is based upon this fact. The bandwidth of the chrominance
    signal (color info) is order of magnitude less than luminance (B&W
    Don Stauffer, Jun 2, 2004
  10. On 1 Jun 2004 22:27:24 -0700, (Georgette
    Preddy) wrote:

    The spacial resolution of the SD10 is c3.4megs (2,268~1,512), the
    spacial resolution of the 300D is c6.3megs. (3072x2048) so on pure
    spacial count the 300D has more resolution, but has to use clever
    interpolation to decide colours of any given point. With the SD10 you
    have to up-scale the image (asuming you want the same pixel
    dimentions) at which point you are using a very similar algorithm to
    compare adjacent pixels to guess the inbetween colours and de-pixolate
    (not sure the actual term) to remove "steps"

    Both cameras have to do some formulaic interpolation to decide
    colours, hues, etc from the AD conversion.

    Interestingly enough, the crop factor of the SD10 is 1.7, v the 300D
    of 1.6... sugesting that the physical dimentions of the sensor plate
    are close, and that either the sensor sites on the SD10 are larger, or
    more widely spaced.

    Both cameras deal with trade offs, both cameras interpolate to guess
    missing details where the sensors (or should that be sencells?) dont
    fully get covered by a given item.

    Heck, films been doing that for years but it works as a chemical
    reaction to the light, each "grain" reacts to the average colour of
    the light falling on it within the grains colour type (I may be
    slightly wrong there... I'm not 100% sure if colour film is made up of
    different colour/spectrum sensitive grains, or all the grains are the
    same but change according to the colour of the light)

    I personally dont see a lot of the "noise" and strange things that
    others seem to find when looking at SD10 photos, I think the only
    thing I do see is they tend to look over sharp and a little over
    saturated, but then likewise I dont see the fuzzyness thats supposed
    to be present in bayermasks.

    Likewise I dont find the noise issue to be overly bad in either camera
    from the images I've seen, compared with film at similar ISO's both
    cameras seem to show far less noise/grain than film.

    Something very interesting about photo-mag reviews now is that the
    size of the prints/crops combined with the printing techniques means
    that the examples they use to show "problems" are very hard to
    actually see the noted problems, perhaps its just my eyes are not as
    good as they used to be, lol.

    On a web site you can have a small crop and blow up each pixel to
    larger than its 1 pixel size to show the resultant noise, but most
    printing solutions do not increase the physical dimentions of a pixel.

    I first really saw this in "real life," so to speak, when some ISO1600
    photos I'd taken in quite poorly lit rock venue were printed. On the
    monitor they looked quite bad, especially as not only had they been
    ISO1600, but they had also been taken with as fast a shutter I could
    get away with to try and freeze motion, so were quite under exposed as
    well, so with pushing some of them almost 3 stops using the ISO
    compensation; as you'd expect, the results looked hugely noisy. But
    when I printed the first one at A4 (roughly 8/12 inches) the noise
    while present seemed to reduce significently I was viewing at 1-1 on
    the monitor tho.

    (12" width monitor, 1600 pixels = 133pixels per inch viewable, 12"
    paper at 3072 pixels = 256 pixels per inch, so 1-1 is almost 200%
    magnification on screen)

    This leads into the old argument about pixels divided by print size,
    etc... most large scale print works dont enlarge prints pixel for
    pixel (actually thats a pain for me, as I'm wanting to produce some
    large pixilated prints) instead they re-sample or RIP prior or during
    the printing process and do lots of other nifty things, but I digress.

    I actually feel sorry for Sigma having GP post on here... the almost
    psycopathic ranting that is in here is probably going to make most
    people decide not to buy the camera no matter how good it is. I really
    dont see what GP hopes to gain, unless its some sad notoriety, it
    definatly does no favours for Sigma having him post here.

    Oh well, time for another kill file addition :-S

    Jonathan Wilson, Jun 2, 2004
  11. (Georgette Preddy) wrote in
    3) Now suppose the SD9 is aligned so that a pattern of tiny dots
    all fall in between the sensors. How many dots will you then capture?

    4) Now suppose a big grizzly bear stands behind the camera. How
    many of the grizzlys fearsome teeth will be captured before the
    photographer is eaten?

    5) Now, furthermore, assume that you are standing on the moon and
    you want to take a photo of Paris. Will you or will you not be
    able to do that? Will it be easier with a Foven camera?

    Roland Karlsson, Jun 2, 2004
  12. A well designed Bayer camera has a perfect anti alias filter, so in both
    cases the picture will be a blur, i.e. something that cannot be resolved,
    which is exactly the correct result. In practice, no such thing as a
    perfect anti alias filter exist, so you will get some not so nice
    behaviour from your highly unlikely situations.

    Now, you can make the same kind of examples for cameras without an anti
    alias filter. The only difference being that those examples are not
    higly unlikely but rather highly likely. One typical example is
    photographing a thin telephone wire that goes diagonally. What will
    you see? With a filter you will se a blurred smooth line, without you
    will see some pixel patterns that jumps here and there, disappears etc,
    etc. Which do you like better?

    Another typical example is a tiny dot. With an anti alias filter it will
    be a blurred spot. Without, it can be anything. It can be captured by
    zero to 4 pixels, and each case it will look different. Imagine a regular
    pattern of such dots. With an anti alias filter it will be a regular pattern
    of blurred spots. Without, it will be a patten of different sharp but
    strange things that sometimes are missing. Whats your preference?

    Roland Karlsson, Jun 2, 2004
  13. The answer to this last one is a resounding yes. It's perfectly
    obvious that Preddy can do pretty much anything once he's got his
    grubby little paws on a camera with a Foveon sensor in it...

    Whether he'd post the pictures online for everyone else to see is
    another matter entirely...
    Paul Fedorenko, Jun 2, 2004
  14. The Foveon sensor would resolve any square that fell on a particula
    sensor utterly perfectly, you may not se every one but you'd see 3.43M
    of them pristinely with total color perfection. The Bayer shows the
    entire image as pure gray in both cases--nothing at all is resolved.
    Georgette Preddy, Jun 3, 2004
  15. Georgette Preddy

    JPS Guest

    In message <>,
    Nothing *should* be resolved that wouldn't also be resolved if the
    alignment shifted by 1/2 pixel in both directions. That's where the
    Sigma SD9 goes wrong.

    It wouldn't be grey on a Bayer; it would be grey-green if you used 2
    greens to every red and blue, or grey-magenta if you used the
    complementary color for each CFA element.
    JPS, Jun 3, 2004
  16. Bunk, luminance is color. Bayer's senses 1/3rd of a luminance value
    at each photosite, before it throws 25% of the total sensor count away
    due to no RB complimentary info.
    Georgette Preddy, Jun 3, 2004
  17. The Canon 10D's (and/or any Bayer 6MP-monochrome camera) full color
    spatial resolution is 1.5MPs.
    Monochrome. Color is 1.5MPs.
    Interpolation is never, ever used by pro cameras.
    Correct, which is why the 13.72MP Foveon is completely noise free at
    ISO 100 while Canon iamges are awash in noise at ISO 100. See...

    Foveon never, ever interpolates (guesses) the color of a pixel, except
    at its 13.72MP setting. But even at it's 13.72MP output, Foveon
    photographs are much more accurate than Bayer digtial interpolations,
    since Foveon starts from a photographic baseline (100% optical) rather
    than digitally interpolated baseline (25% optical).
    Georgette Preddy, Jun 3, 2004
  18. Georgette Preddy

    Bryce Guest

    What the hell would you know about pro cameras?
    Bryce, Jun 3, 2004
  19. Georgette Preddy, Jun 3, 2004
  20. Georgette Preddy

    BillyBob Guest

    If the Foveon is so good why is Sigma the only company to ship such a
    product to date? The number speak for themselves -
    BillyBob, Jun 3, 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.