The Human Eye: 120 Megapixel Monochrome, 6 Megapixel Color

Discussion in 'Digital Photography' started by Brian C. Baird, Jun 15, 2004.

  1. Brian C. Baird

    Frank ess Guest

    That may be correct, and so is this: he's a needy individual whose "any
    kind of attention is better than no attention at all" attitude is
    unhealthy for him and for those who feed him.

    Frank ess
    Frank ess, Jun 17, 2004
    1. Advertisements

  2. Brian C. Baird

    Dave Haynie Guest

    Well, I'm sure he's alluding to the fact that the eye contains about
    6.4M cones, but 110M-125M rods. And to boot, only about 640K of those
    are blue sensitive cones, while the remaining GR cones are also
    feeding the luma processor in your visual cortex. So basically, Bayer
    is dramatically more balanced, in its color processing, than you eye.

    Some of the more recent digital sensors, though, may be taking a queue
    from this. My not-so-new Canon Pro90 sports a CMYG sensor, which is
    more interesting for an RBG interpolation, albeit requiring more CPU
    power. Some new, single-chip camcorders are incorporating "W" (eg,
    white, eg, clear) sensors in their sensor lattice (such as JVC's
    Prosumer HD-DV camcorders).
    Well, yes and no. The blue cones (chroma-only cones) have a peak
    response at about 420nm. The green cones (chroma and luma) at about
    534nm, and the red (also chroma and luma) at 564nm (rods feed luma
    processing only, sensing blue-green at a peak of about 498nm). But
    that's peak sensitivity.

    If you want to get technical, the blue cones peak between violet
    (400nm) and indigo (445nm). But they're still responsive out to blue
    (475nm) and into green (510nm). The green cones peak is strongly
    green, sure enough, but they're responsive down to indigo, and well
    into red (640nm). The red cones peak is actually on the green side of
    the generally accepted center yellow (670nm), but they're sensitive
    back to indigo as well (though less so than the green cones), and all
    the way up to near-infrared (which, of course, IS the reason for that

    There's a reson for this: as well as doing "the color thing", the
    green and red cones do much of our bright-light luma sensing as well
    (the response of the rods is limited in bright light; they saturate in
    normal daylight).
    Yup; the eye has three separate processes in visions: luma, the
    red-green oppositional process, and the blue-yellow oppositional
    process (where yellow, as mentioned, is the red less the green).
    Detail is processed in the luma sense, then basically augmented with
    the color, at a dramatically lower resolution.

    And yeah, this was being investigated 100 years ago or so. When color
    TV came along, the current analog kludge, preserving full
    compatibility with the old transmissions, was deemed acceptable. And
    sure, you can hate NTSC or PAL color, especialy once you spend lots of
    time looking at digital video. What's amazing isn't that, but the fact
    that such horrible color, made worse on VHS, is generally considered
    acceptable to humans.

    And of course, digital, which gives you dramatically better color
    (much of the impact of DVDs and HDTV over broadcast is the color
    vibrance. Which, of course, is 4:1 subsampled, even now :)

    Dave Haynie | Chief Toady, Frog Pond Media Consulting
    | Take Back Freedom! Bush no more in 2004!
    "Deathbed Vigil" now on DVD! See
    Dave Haynie, Jun 17, 2004
    1. Advertisements

  3. Brian C. Baird

    Dave Haynie Guest

    But 18 million single color pixels are better than six million fully
    RGB sensors. Foveon can't win.

    And not only that, it can't even break even, and that's not the
    concept itself. If they had a technology that was reasonably close, in
    technology cost, to the CMOS Bayer sensors, without any major color
    issues that couldn't be addressed in the camera's DSP, the Foveon
    approach would probably dominate at some point, simply because "better
    for the same or slightly more" beats "not better".

    The problem is that this isn't a 6Mp bayer vs. 6Mp Foveon, it's more
    like a 6-8Mp Bayer vs. a 3Mp Foveon. And, I'm told, that sensor takes
    up 60 million transistors. That's half the size of a Pentium 4, one of
    the most bloated chips known to man. How do they scale that to 6Mp and
    beyond, and still deliver something that's going to be manufacturable
    and run from batteries.

    Now, obviously, the image sensors aren't running at P4 speeds, nothing
    even close. But they're approaching those transistor counts, and those
    counts aren't unlimited without eventually shrinking the feature size.
    And Foven's supposedly at 130nm already (the best process National
    Semi has). The problem is, can they really put 120M or 180M
    transistors on a 130nm die? Sure, there's no good reason to shrink the
    die itself; if anything, you want to grow it. But heat and yield
    issues become critical. If you shrink to 90nm, you'll start to have
    Intel style leakage problems.. and leakage is per transistor,
    independent of clocks or use. National's not even in the game on
    solving the 90nm node yields (even Intel's behind, which is unlike
    Intel; they usually beat everyone else in process if not in design).
    No, it's never, ever going to happen. And I think today, Foveon's big
    problem isn't taking over the world ("They're Pinky and the
    Brain..."), but rather, getting out of that cage. The one in which
    Sigma, a company who may or many not be in the DLR market next year,
    is their only serious customer. And for that, they need more pixels
    and better qualtiy pixels. And to stop lying about the pixel count;
    they will alienate people in the know, without their ever giving
    Foveon-based chips a second chance. Maybe not an issue if your focus,
    as theirs is now, is entirely consumer oriented. But if they put that
    chip (or, for a little while anyway, the next one, if they get to
    6Mpixel soon) in a pro-class camera, the company's reputation and
    relative levels of monkey business will be an issue.

    Dave Haynie | Chief Toady, Frog Pond Media Consulting
    | Take Back Freedom! Bush no more in 2004!
    "Deathbed Vigil" now on DVD! See
    Dave Haynie, Jun 17, 2004
  4. Brian C. Baird

    Dave Haynie Guest

    No, it doesn't. Cones are interstitial, as in a Bayer sensor, not
    stacked, as in a Foveon senor. And as menitoned, even if you ignore
    the fact that less than 5% of the light sensors in the eye are
    involved with color vision in any way at all, the mix in the eye is
    even worse than in the Bayer sensors, with only 10% of the color
    sensors devoted to blue, for example.
    Or, maybe, Foveon was named after the Fovea. Ya think?

    However, the sensor in question is called the X3, not the Foveon.
    Foveon itself began life making unmasked and Bayer sensors. Their
    first projects were simply making CMOS sensors, then large CMOS
    sensors (I believe they had the first 16Mpixel CMOS sensor) that were
    accetable replacements for the CCDs most folks used. They and Canon
    successed, basically at the same time, on the quality front (CMOS
    sensors powered cheap webcams and Vivitar digital P&S before this, but
    nothing serious).

    But Foveon has always had a problem -- customers. Most major digicam
    companies making at least some of their own, including Kodak, Sony,
    Canon, Nikon, and Fujifilm. Not sure about Olympus, but that's about
    65-70% of the market, anyway. Then there's Sanyo, which makes about
    50% of all of the CCDs in the world, and they're strong in the digicam
    market (certainly they fill in the low end). Sony and Kodak have also
    sold CCDs on the open market.

    So who's the customer? As we've seen, only (so far) fourth tier
    companies. And that's also a technology issue -- the money for the
    next design comes form today's market.
    Again, no, they're entirely unrelated. The company name was in place
    long before there was an X3 sensor, or even the first X3 patent filed.
    And the X3 doesn't work anything like the eye (neither good nor bad,
    but a correct observation). Bayer sensors are closer, even to the
    extent that they're using dye for color filtering.

    Dave Haynie | Chief Toady, Frog Pond Media Consulting
    | Take Back Freedom! Bush no more in 2004!
    "Deathbed Vigil" now on DVD! See
    Dave Haynie, Jun 17, 2004
  5. Brian C. Baird

    Dave Haynie Guest

    Not surprising at all. But the specific lack of a road-paving animal
    made the wheeled animal a liability, not a competitive advantage. A
    wheeled vehicle (or animal) can't get over a barrier larger than a
    fraction of the wheel's diameter. Thus all the work in robotics on
    Dave Haynie | Chief Toady, Frog Pond Media Consulting
    | Take Back Freedom! Bush no more in 2004!
    "Deathbed Vigil" now on DVD! See
    Dave Haynie, Jun 17, 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.