Tricky shot of an old church

Discussion in 'Photography' started by Scott W, Nov 16, 2005.

  1. Scott W

    Scott W Guest

    Well first off your a bit off on your sampling theory, the Nyquist
    limit does not need any given relationship between the signal and the
    sampling phase. This gets people all the time because they will show a
    sample like yours, where you show what looks like a frequency right at
    the Nyquist limit, but in fact it is not. What you showed was a
    rectangular pules of a sine wave, this has side bands some of which go
    past the Nyquist limit. If this seems like nit picking it is not, if
    you shape the sine wave smoothly the side bands will be much closer to
    f0, you still need to shift f0 down just a bit to keep everything below
    the Nyquist limit. It turns out that if you keep all the energy,
    including side band that are cause by modulating the signal, below the
    Nyquit limit you will capture all the information there is.

    BTW This is not just an intellectual interest of mine, this is what I
    do for a living and have been for many years.

    Scott W, Nov 22, 2005
    1. Advertisements

  2. Scott W

    Scott W Guest

    Well yes, sometime for DOF you need to go to a large FN, but this does
    not mean you will no loss detail when you do. You will loose the same
    percentage of detail shooting at F64 as someone would who is shooting
    35mm at F64, but you start out with a lot more.

    The image you show in your test sample is a very soft scan for film at
    3300 ppi. I am surprised at how much grain seems to be visible in a
    3300 scan.

    I am sure you 30 x 40 in print looks great, but if you were to cut a 4
    x 6 piece out would it really compare well with a normal 4 x 6 inch
    print. Large prints look sharp in part because we expect them to look
    soft, since this is the norm. Tonight I had 6 12 x 18 prints made at
    Costco, the people at Costco commented on how sharp the prints were. 4
    of the 6 were just straight out of the 20D and two were from stitched
    photos. Yes the one right out of the 20D looked very sharp, until you
    compared them against the one printed from 20MP file from the stitched

    We can look at this another way, for you to make a 30 x 40 inch print
    means a magnification of 8. If we scale that to 35mm it would be a
    print very close to 8 x 12. Now using a good lens and good film you
    can get a pretty sharp looking 8 x 12 from film. But if you do the
    same shot at f8 and at f64 the print made from the shot at f64 will
    look much softer then the one short at f8. This scales up directly to
    your case, since the film would be the same, the diffraction spot would
    be the same size and he magnification would be the same. In both
    prints the diffraction spot will scale up to 0.8 mm in diameter. And
    if we are printing at 300 ppi the diffraction spot is 9.5 pixels in
    diameter, that has to hurt the sharpness.

    Scott W, Nov 22, 2005
    1. Advertisements

  3. hmmm...I see I should have been more specific and said:

    "That, by the way, is the same reason that the miniature format (aka 35mm)
    'normal' lens image has more depth of field than the medium format 'normal'
    lens image."

    That having been said, no argument from this corner that the shorter lens
    has more depth of field than the longer lens at the same f/stop, regardless
    of film/camera format. At the same f/stop, the 22mm lens (on my 16mm
    camera) has deeper depth of field than the 50mm lens (on my 35mm camera)
    than the 80 lens (on my 120 camera) than the 150mm lens (on my 4x5) than the
    300mm lens (on someone else's 8x10). By the same token, at the same f/stop
    with my 35mm camera the 20mm lens has a greater depth of field than the 50mm
    lens than the 85mm lens than 135 lens than the 200mm lens than the 500mm
    lens. Likewise, at the same f/stop the 150mm lens on a 35mm camera has the
    same depth of field as the 150mm lens on a 120 camera as the 150mm lens on a

    Gosh-be-darned - "MF is the miniature format". I swear I learn something
    new each and every day! And here all these 50 plus years I thought MF is
    medium format! hmmmm....if MF is miniature format, makes you kind of wonder
    what medium format is. After all, MF can't be both miniature format and
    medium format at the same time...can it?
    Lawrence Akutagawa, Nov 22, 2005
  4. Uh, well yes, but it sounds as though you are missing the point.

    On the same format, DOF goes down with the _square_ of the focal length, or
    up with one over the square of the focal length. So if you keep the format
    the same, a 24mm lens has four times the DOF as a 50mm lens.

    So far, so good, right?

    But if you change the format, you have to enlarge the negative more, so you
    lose half the increase in DOF you were expecting, and DOF goes up linearly
    with the focal length. So 6x7 has twice the DOF (not four times the DOF) as
    No. This is wrong, because you've left out the enlargement to get to the

    When you compare DOF, you have to keep the print size and viewing distance
    the same. (You really have to keep the angle of view the same as well, so
    comparing DOF of the same focal length on different formats really doesn't
    make sense since the images are completely different.)

    But if you insist on the comparison, then since the enlargement to the print
    is larger for 35mm, the DOF of a 150mm lens on 35mm is much smaller than it
    is on a 4x5 camera.

    David J. Littleboy
    Tokyo, Japan
    David J. Littleboy, Nov 22, 2005
  5. Scott W

    Scott W Guest

    David is right on this and there is a very easy way to visualize it.
    Think of projecting a point from the film plan towards the object. This
    will be a cone converging at the focal plane in space, how fast the
    image losses focus is dependent on how steep the angle of the cone is.
    Now take the case of two cameras each with the same f number and the
    same field of view, the smaller camera will have a smaller exit
    aperture and so its cone will be at a smaller angle and therefor more
    depth of field. It should be noted that the resolving power of the
    system is better for larger angles. This is the immutable fact, there
    is a trade off between resolution and DOF in an imaging system and this
    trade off is independent of the camera format used. If you want to
    increase resolution in the object space it has to come at the expense
    of DOF. A system that only resolves on the order of an inch can have a
    DOF that is huge, but a system that resolves down to say a half a
    micron can not have more then a a couple of microns DOF.

    Since the goal of a LF camera is in part to capture more detail in the
    object plane then a MF or 35mm camera it has to have less DOF, if it is
    gong to caputure this finer detail.

    Scott W, Nov 22, 2005
  6. With all due respect, no I am not off. Yes, this is a common confusion.
    But what must be considered in cases like this is that the sampling
    is limited, in fact to as little as one cycle. Consider the following:
    a sine wave. Sample at 0 and 180 degrees. Nyquist theorem says sampling
    must be more than twice the frequency. If you had infinite sampling,
    you can get recover the sine wave by sampling every 180.000000......1 degrees.
    But images are not infinite sampling, so sampling must be much higher.

    The effects on the sampling page show the same basic results with a
    sine wave pattern.

    (I do this stuff for a living too.)

    Roger N. Clark (change username to rnclark), Nov 23, 2005
  7. Yes, I agree.
    Perhaps you haven't had a drum scan before. I drum scan (in my experience)
    has about 50% more resolution than the same ppi scan from a consumer
    While this image is at f/45 for large format and f/11 for 35mm,
    it illustrates the same point you are citing above:
    In theory, diffraction limited optics should give the
    exact same resolution, but obviously, it does not.

    Now print it at about 412 ppi and it will be equivalent to a 30 x 40
    inch print. How sharp is the 4x5 part of the image?

    The full image is here:

    The testing page is here:

    This will give you an idea of 4x5 drum scan sharpness at f/45.
    And f/64 is only about 40% lower resolution.

    If you ever visit Colorado, I can direct you to a gallery where my
    f/64 flower image is on display, and you can see for yourself how
    sharp a large print from 4x5 is. It is interesting to watch
    visitors approach such a print, stop take in the view, say "wow!"
    then approach until they are a foot or less away
    looking at the detail. I've seen it again and again.

    The Rayleigh resolution criterion for f/64, green light gives
    26 lines per mm. An 8x enlargement gives 3.2 lines/mm. Not
    super sharp, but not bad either, and enough to be very impressive
    sharpness in a large print.

    Roger N. Clark (change username to rnclark), Nov 23, 2005
  8. Scott W

    Scott W Guest

    The images may not bee infinite but they are very large compared to the
    sampling rate.

    Let's take a case, say a row of fence post where there is a post ever
    other pixel, so the post would be right at the Nyquist limit. And lets
    say that there are no frequencies on the film past the Nyquist limit,
    that due to the combined effects of the optics and the film that by the
    time we hit one half the sampling rate that is no energy left, in this
    case the fence posts would not be visible on the film, regardless of
    how we samples.

    Now let make it so that the fence post spacing is just a bit larger, to
    lower there frequency into the pass band of the optics and film filter
    if you will. Since I have only 10 fence post I know I am going to get
    strong side band out to at least 10% of the fence post frequency, so I
    have to have the fence post space at least every 2.2 pixels, to make
    out much at all of them.

    In electronic system we can have filters that have a very sharp cut
    off, so these kind of issues are important, in images recorded on film
    the cut off is always slow so that by the time you have part of an
    image that getting close to the Nyquist limit it amplitude is so low
    that you would not see it in any event.

    There is another aspect of this that I think you are really trying to
    say. There is a difference between the sampling rate needed to capture
    all the information in a signal and the sample rate needed to reproduce
    that signal. People confuse that samples, which capture the signal,
    with the regeneration of the signal, these are two different things.
    One of the things that I like to do is show people a waveform where I
    am sampling at twice the Nyquist limit. I then separate the even and
    odd samples, so now I have two sampling of the signal but with a phase
    shift between them. Because I have sampled at twice the Nyquist limit
    ether the even or odd sample should be enough to capture the data. The
    even and odd samples will look very different for each other, but when
    I reconstruct the waveform with each both the reconstructed waveforms
    are exactly the same. Now if I want to simply use my samples to
    represent the waveform, where I am not going to generate the waveform
    by summing all the sine and cosine waves but rather just plop the
    samples down one after another, I might want to sample at a much
    higher rate, but this is not a limitation on the Nyquist theory. It is
    rather a consequence of using the samples as the waveform instead of
    reconstructing the waveform.

    Scott W, Nov 23, 2005
  9. Scott W

    Scott W Guest

    I think that is the main point, what looks sharp in a large print is
    different then was looks sharp in a small print. But clearly you would
    have had a sharper print if the shot was taken at f/32, but of course
    with some areas much softer due to DOF limits. People shooting 35mm
    cringe at shooting F/64, but they don't loss any more detail on a
    percentage basis then what you do when you shoot at f/64, again
    assuming you are using a good lens. But then the person shooting 35mm
    will only get a print that is maybe8 x 12 inches, so it will be viewed
    closer then a print from a 4 x 5.

    At 26 lines per mm, and this is where the lines completely disappear,
    the most useful pixels that you can have would be 35 MP. And even at
    that these would be somewhat soft pixels. Whereas a 35 MP is very good
    a 4 x 5 camera should be able to produce a image with a bit over 100 MP


    Scott W, Nov 23, 2005
  10. Scott W

    Lorem Ipsum Guest

    This whole thread has made me a hard-core convert to LF! I'm moving on, not
    looking back, not doing any more silly math, no bench racing, just making
    real pictures.
    Lorem Ipsum, Nov 23, 2005
  11. Scott W

    Scott W Guest

    I thought you were already shooting LF.

    I have always said the LF is a great format. This summer we spent a
    lot of time traveling around the country and went to a lot of national
    parks. Often there would be posters for sale in the visitor centers,
    these would look pretty good when you first walked in but when you
    would get at all close it was clear that they were shoot with 35mm. At
    the time I really wondered why anyone would go to the trouble of
    printing up larger posters and starting with such a low res photo, you
    had to wonder why there were not using a LF camera.

    As for the math being silly remember that it was someone who did not
    think math was silly who designed the lenses that you use, so that you
    can just go out and take photos.

    As for me I would be out taking photos, but it is a very cloudy day
    here today..

    Scott W, Nov 23, 2005
  12. Yes, but depending on the phase of those samples (10*2.2 -22 pixels),
    you may or may not pick up something that is at all recognizable.
    Adjust the phase just right and you can maximize the response.
    If you want to ensure against phase errors in finite sampling
    situations, you must increase the sampling rate higher than Nyquist.
    We are on to something here, but Nyquist refers to infinite
    sampling. Finite sampling requiring higher than Nyquist does
    not violate the theorem, it is simply a different situation
    not covered by the theorem. There are two basic effects in photography:
    digitizing film, and digital cameras with discrete pixels.
    Both are finite sampling. One can go down to something as
    simple as two closely spaced grass blades. How many samples do
    you need to record the grass blades correctly 1) on film,
    or 2) in a digital camera? Consider the grass blades are
    straight and sticking up against blue sky at an angle.
    The digital image should show no phase inversions and
    consistent separation, but can be pixelated indicating
    they are just detected.

    Roger N. Clark (change username to rnclark), Nov 23, 2005
  13. But have you printed the image? How does it look, and how
    does the 35mm versus 4x5 portions look?

    Roger N. Clark (change username to rnclark), Nov 23, 2005
  14. Scott W

    Scott W Guest

    I would be delighted to print out a section, could you post something
    like a 825 x 825 crop from a shot that was done at f/64? I have enough
    35 scans around to compare it to. a 825 x 825 crop would give me a
    print that is 2 inches square, this should be large enough to look at
    detail. The grass section is hard to use because it is hard to tell
    what is grain and what is grass.
    An area with some of the dead branches would be the best area for
    looking at detail I am thinking.

    An image from a f/45 shot would do but the softness would be most
    visable from a f/64 shot.

    Scott W, Nov 23, 2005
  15. Scott W

    Neil Gould Guest

    How practical is this solution for photography?
    Now, there's an understatement. In audio, it's 192 kHz and counting to
    capture a <20kHz signal "accurately", or almost 5x Nyquist.

    Neil Gould, Nov 23, 2005
  16. Scott W

    Scott W Guest

    You only need 40 kHz to capture a 20 kHz signal, but then where do you
    find a 20 kHz signal that does not have some energy past 20 kHz. It is
    far easier to sample at a very high rate and then digitally filter out
    those frequencies above 20 kHz. On the output end it is a different
    story, you can't just put out the samples that you took, it does not
    work that way.


    Scott W, Nov 23, 2005
  17. Scott W

    Lorem Ipsum Guest


    Lecture all you like about math. So I should be thankfull of mathematicians?
    Should I remind you of the people who built the Interenet? Get over it.
    Lorem Ipsum, Nov 24, 2005
  18. Scott W

    Neil Gould Guest

    That's not a problem for many synthesizers. A few of the mixes that I've
    run into lately put a lot of energy into the high end, as well. I think
    they may be over-compensating for their loss of hearing as they age. A
    different question may be, how many people's typical audio systems can
    reproduce that kind of signal?
    Perhaps that's why many people are recording at 96 & 192 kHz? ;-)

    As Roger pointed out, you can capture some of the information Nyquist, but
    if you want to represent the actual waveform (presuming it's not a square
    wave), the sample rate needs to be higher or phase alignment will limit
    you. How much higher is a matter of debate.

    Neil Gould, Nov 24, 2005
  19. Scott W

    Scott W Guest

    Well let's put some numbers on it, here is a file of 256 numbers, that
    I say are over sampled by a factor of 8. I can take every 8th sample
    and reconstruct the wave form, and I don't care about what phase I take
    (there are 8 to choose from).

    Now that is not to say that ever 8th sample can be used as an output,
    rather with every 8th sample I have enough data to reconstruct the
    waveform generate the missing 7/8 of the samples.

    If you look at what the data looks like when you take ever 8th sample
    it looks pretty bad, but all the needed information is there.

    If you have some numbers you would rather use I would be delighted to
    look at them, a sample number that is a power of 2 makes this easier

    Scott W, Nov 24, 2005
  20. Scott W

    Neil Gould Guest

    I don't disagree about this, and in fact says the same thing that I wrote,
    earlier. In the case of 192 kHz, the signal is only roughly 5x
    oversampled, and I personally think even that is overkill to provide
    sufficient data to reconstruct a reasonable facsimile of audio content at
    20 kHz. As you mentioned, there won't usually be a lot of it with any
    amplitude, and further, the input data from most things (other than a
    synthesizer) is likely to be an undesirable sound, anyway. Still, this is
    quite different than a claim that there is sufficient data in 2x
    oversampling (Nyquist) to accurately reconstruct a waveform; there isn't
    for most real-world waveforms.

    Neil Gould, Nov 24, 2005
    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.