Are 22 megapixel APS-C sensors realistic?

Discussion in 'Digital Photography' started by mexican_equivalent, Aug 31, 2006.

  1. Question for those who are more technically inclined than I am...

    What is the likelihood that Canon will eventually mass produce APS-C
    (ie 1.6x crop) sensors that can handle 16 megapixels? 22 megapixels?

    Or will there come a point where it's no longer realistic to cram so
    much on a cropped sensor, and a full-frame camera would be needed to
    keep up the megapixel wars?
    mexican_equivalent, Aug 31, 2006
    1. Advertisements

  2. Well, in theory you could produce an APS-sized sensor with the sort of
    pixel pitch seen on 10-megapixel point&shoots. You'd probably be talking
    about 100 megapixels or thereabouts, and of course you'd have the same
    sort of pixel-level noise limitations as a small-sensor camera.

    I have a bad feeling that in a decade or so, such a camera will be
    produced, to appease the Marketing Gods, if nothing else.

    Daniel Silevitch, Aug 31, 2006
    1. Advertisements

  3. mexican_equivalent

    Scott W Guest

    It would not surprise me at all in canon at some time upped the pixel
    count to 16mp for a 1.6 crop sensor. There are many who will cry about
    the added noise but in theory the noise going from 8 MP to 16 MP would
    go up by the ssqrt(2) if you down sized you 16MP image to 8 MP you
    would be right back to the same noise level as an 8 MP camera, assume
    good low readout noise.

    People have repeatedly predicted the end of pixel wars. When the
    first 6MP DSLR cameras came out a number of people thought that would
    be enough for anyone.

    But there are two reasons why some people are going to want more
    pixels, some people are going to want to make larger and larger print
    and want those print to be as sharp as possible. The other reason is
    for some people even more important, going from 8MP to 16MP on the same
    size sensor will pretty much eliminate aliasing.

    Scott W, Aug 31, 2006
  4. mexican_equivalent

    Kinon O'Cann Guest

    Yes they are, and I wouldn't mind seeing that as an option. For people
    shooting things like landscapes who don't need ultra-low noise at high ISOs,
    the ability to capture enourmous amounts of detail would be nice.
    Kinon O'Cann, Aug 31, 2006
  5. mexican_equivalent

    Alfred Molon Guest

    22MP on an APS-C sensor would mean 4 micrometer pixels, which in turn
    would mean that you'd have to shoot at max. ISO 100 to have low enough
    noise levels.

    This means that if you need a lot of DOF and have to stop down the lens,
    you will need a tripod quite frequently, considering the low light
    sensitivity and the very high resolution. Such a camera would be a
    "tripod camera".

    However with 4 micrometer pixels the smallest aperture would be around
    F12-13 or so (just guessing, don't nail me down on this), because of the
    diffraction limit. You could stop down further the lens but would then
    lose resolution due to diffraction. F12-13 however should give you
    sufficient DOF on an APS-sized sensor.
    Alfred Molon, Aug 31, 2006
  6. mexican_equivalent

    Jones Guest

    Mmm. But why can't noise levels also be reduced with the latest
    technological enhancements?
    Jones, Aug 31, 2006
  7. mexican_equivalent

    Scott W Guest

    The F30 has just about 4 um pixels, it seem tos do pretty good at ISO

    Scott W, Aug 31, 2006
  8. mexican_equivalent

    Pete D Guest

    So you don't think technology can improve enough to also improve the quality
    at higher ISO's, why the hell not?
    Pete D, Aug 31, 2006
  9. mexican_equivalent

    Pete D Guest

    No reason what so ever! CPU's are a good example, Moores Law was in trouble
    for a while there but just recently there has been some large leaps in
    technology or at least the way it is applied. Some can apply to camera
    technology, why not dual sensors to gather twice as many photos, four
    perhaps, how about LMOS or LCCD?
    Pete D, Aug 31, 2006
  10. There's a limit to those sorts of games. Once you get to the point where
    the electronics are capable of counting individual photons (and we're
    not all that that far away from that point), then the limiting factor
    becomes number of photons/pixel and the uncertainties (i.e. noise)
    inherent in a Poisson-statistics system. Bigger pixels means more
    photons per pixel (assuming the same lens etc), and a better
    signal/noise ratio.

    What could work is a small-pixel sensor that handles high-ISO tasks by
    binning pixels together. Call it the digital equivalent of large-grain
    high-ISO film.

    Daniel Silevitch, Aug 31, 2006
  11. mexican_equivalent

    JohnR66 Guest

    I'd figure they'd stop at 12mp for APS sensors. Let's say (for nice round
    numbers) there was a 6000x4000 24mp APS sensor, 22mm on the long side.
    That's 272 pixels per mm or 136 max possible cycles per mm. The actual
    resolving power of the camera will likely still be less.

    Isn't this out of reach of optics that cover this much image area? (aside
    from a few exotic lenses) I find the resolution of my 6mp Rebel pushes most
    of my lenses unless I stop down.

    JohnR66, Aug 31, 2006
  12. The reason is photons per pixel. Think of buckets in a rain storm.
    Larger buckets collect more rain drops. Larger pixels collect
    more photons. ALL good cameras are already photon noise limited
    for the current quantum efficiencies. Quantum efficiencies could
    be raised by about 3x, so there could be some reduction in pixel
    size without loss (1.7x reduction in pixel pitch).
    As pixel size goes down and maximum photons drops, so
    does dynamic range.

    Some references:
    Procedures for Evaluating Digital Camera
    Sensor Noise, Dynamic Range, and Full Well Capacities;
    Canon 1D Mark II Analysis
    (see table 1).

    The Signal-to-Noise of Digital Camera images
    and Comparison to Film
    (see table 3).

    Digital Cameras: Does Pixel Size Matter?
    Factors in Choosing a Digital Camera

    Roger N. Clark (change username to rnclark), Sep 1, 2006
  13. mexican_equivalent

    Scott W Guest

    There is another way of looking at this. Lets make two big
    assumptions that admittedly not all that good, one the read out noise
    can be made much lower then it currently is and two that if we
    quadruple the number of pixels in the same area that their collection
    area will drop by a factor of 4 and their well depth will drop by a
    factor of 4. In this case then we could combine the values from 4
    pixels and get the same number of photons detected with the same well
    depth as if it as one larger pixel. Yes there will be more noise if we
    don't combine pixels but only because we are looking at higher
    spatial frequencies. In effect with the larger pixels we are filtering
    out the higher frequencies and therefor reducing the noise bandwidth.

    Now the weak part of the argument is that we can keep the same
    efficiency in terms of active area of the pixels vs dead area and that
    the read noise can be made lower. But as you point out the read noise
    does not dominate currently and increasing the pixels by a factor of
    four and then summing would only increase the readout noise by factor
    of 2, assuming the same readout noise for the smaller pixels as the
    larger one.

    If we take this a step further an idealized sensor would have 100%
    quantum efficiency and record the precise location where each photon
    was detected. Then the trade of between noise and resolution could be
    done in post processing. This is of course not a realistic sensor for
    a digital camera but it can be used as a limit to what could be done if
    no practical consideration needed to be taken into account.

    Scott W, Sep 1, 2006
  14. SNIP
    Other scenarios can be considered as well. How about a sensor array
    that can sample at half pixel positions by e.g. piezo controlled
    repositioning. With adequate sampling density, this might even allow
    to reduce the strength of the AA-filter a bit. Thus we can combine the
    larger sensel area with higher resolution and less aliasing tendency.
    That also would allow to employ techniques known as 'Drizzling',
    allowing even higher resolution in postprocessing.
    Bart van der Wolf, Sep 1, 2006
  15. You are all forgetting that photons are finite. For a typical
    daytime scene, you want an exposure time of a small fraction
    of a second, you only get so many photons/second per square micron
    delivered to the focal plane. It is not a large number.
    Somewhere I have those calculations, probably back a year or so
    ago in a response to Ilya, who was advocating really tiny

    For example, in a 1D mark II with 8.2 micron pixels, 67 sq microns,
    you get 80,000 electrons (photons) per pixel, or a maximum of 1195
    photons per square micron. Increase quantum efficiency to 100%
    (about a factor of 4), and you get 4780 photons/sq micron. But that
    is max signal. The typical 18% gray level would have only 4780*.18
    = 860 photons /sq micron: a signal to noise ratio of 29 for 1-micron
    pixel. It would be an OK (not great) image, and worse than current small
    pixel P&S cameras. Then of course, as your light levels fall and
    you boost iso (the above is for iso ~80), you would have very few photons,
    e.g. at ISO 800, you would get only 86 photons on the gray card.
    Note that the dynamic range also decreases as the photon count goes down.
    All this also assumes that the full well electron problem gets solved.
    Currently, full wells max out at about 2000 electrons/square micron, and
    it isn't clear that you could have much of a well at all with 1-micron
    pixels. There must be a "wall" to hold the electrons (it is a potential
    well in a CCD).

    Back to the original question, are 22 megapixel APS-C cameras
    possible? Yes, but performance would be low. That is my fear,
    that the megapixel wars will degrade image quality. Image quality
    in DSLRs is so high because the pixels hold 50,000 electrons and
    above. In my opinion, 50,000 electrons should be a threshold
    that should not be crossed. For current sensors, that means
    pixels in the 6 to 8 micron range being the sweet spot.
    The some DSLR cameras are pushing that limit down, and that is bad
    in my opinion.

    Roger N. Clark (change username to rnclark), Sep 1, 2006
  16. There is the photon noise limit.

    One very obvious way to improve performance is to do what
    good video cameras do: don't use Bayer color filters but
    instead use three CCD sensors with a prism beam splitter.

    This of course implies either a relay lens or VERY different
    taking lens design. Expensive!

    Doug McDonald
    Doug McDonald, Sep 1, 2006
  17. mexican_equivalent

    ASAAR Guest

    That's from your perspective, which prefers the highest possible
    quality. But 22 megapixel APS-C cameras would not have poor
    performance or low image quality, despite having what some would
    call "insanely small" pixels. Consider a non-DSLR, Canon's A620.
    It has a 7.2mp 1/1.8" sensor, and according to dpreview:
    Dpreview also shows several different APS-C sensor sizes, but the
    default used in its comparison table shows 23.7mm x 15.7mm. The
    A620's sensor is 7.16mm x 5.319mm. The ration of the sensor areas
    is therefore 372 / 38.2, or about 9.75 to 1. This indicates that if
    the A620's sensor was enlarged to APS-C size, it would contain
    approx. 7.2 * 9.75, or a bit more than 70 megapixels, and give
    results just as good as the A620.

    While it's true that the A620 won't match the image quality of a
    6mp or 8mp DSLR, if it used an APS-C size sensor, and reduced the
    number of pixels by 1/3, making each much larger, it would produce a
    much greater image quality with reduced noise, and if not comparable
    to current DSLRs, it certainly would not be very far behind. The
    point being that while this might not be the ideal camera for those
    looking for cameras producing the highest possible quality images, a
    22mp APS-C sensor would probably make for a very nice sensor in an
    entry level DSLR or an excellent sensor in some future P&S. Its
    performance would only be described as "low" by those who see almost
    anything below the highest DSLR performance as being too low. Most
    photographers would see it as a leap forward. There's one
    additional reason why such a camera would be, as Martha might say,
    "a good thing". It would hasten the development of mid-level full
    frame DSLRs having sensors with 22mp and above.

    What I think you don't give enough importance to is that the
    photographic community is made up of people with vastly different
    photographic needs, and people such as you make up only a very small
    part of it. If manufacturers catered exclusively to your
    demographic, you wouldn't see many sensors with pixels smaller than
    your 6 to 8 micron range sweet spot. In your opinion that would be
    good. But from the perspective of manufacturers losing sales, that
    would be bad, so it's unlikely to play out as you'd like it to.
    ASAAR, Sep 1, 2006
  18. "Roger N. Clark (change username to rnclark)" <>
    wrote in message SNIP
    No, I'm not forgetting that, but I do realise that a double exposure
    is needed, so more light is needed to reduce subject motion. Afterall
    camera backs like the PhaseOne 45 already allow such an option, but
    that is obviously for stationary subjects in a studio setting.

    A somewhat related article I saw today also shows that there is still
    room for creative solutions:
    although that also brings a temporal component into the equation for
    motion reduction.

    SNIP good summary
    That's the reason that I'm a bit sceptical towards a 22 Mp full frame
    sensor, unless we also see a quantum leap in the Digic III
    capabilities. The number of recorded photons is a strong limitation
    for what can be done without adaptive noise reduction software.
    Dynamic range and noise are not getting better with fewer photons.
    Bart van der Wolf, Sep 1, 2006
  19. Actually, it can be done after-the-fact. I asked about this some
    time ago, and was given some pointers which led to a software solution.
    It's free, Open Source, multi-platform (i.e. Windows, linux, *BSD, etc)
    software called ImageMagick. The home page is
    I have a Panasonic FZ5 which is (in)famous for noise, so this isn't
    merely theoretical on my part. I contributed instructions for binning at and
    an example of its effects.

    I took a batch of test pictures 640x480, 1280x1024, and 2048x1536 at
    ISO 80, 100, 200, and 400. Then I binned down the 1280x1024 pictures to
    640x480 and the 2048x1536 pictures to 682x512. After that, I took a
    50x50 pixel crop of blue sky from each image. Then I lined up the
    results in a table, and used standard HTML to get your browser to
    magnify the fragments. If your screen is 1024x768, see...
    If your screen is 1280x1024 or larger, see...

    The exampl shows that there is a definite improvement from binning.

    The digital photo section is still a work in progress, but I hope my
    contributionss can help out people the way that people on this group
    have helped me out.
    Walter Dnes (delete the 'z' to get my real address, Sep 1, 2006
  20. mexican_equivalent

    Bill Funk Guest

    You are asking people to look into their crystal balls, and tell you
    what the future holds.
    My balls aren't crystal (one of them is a Magic 8 Ball).
    The real crystal balls are currently not working, because of the
    sunspot cycle.
    Bill Funk, Sep 1, 2006
    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.