Digital night photography

Discussion in 'Digital Photography' started by Mojtaba, Aug 16, 2004.

  1. Mojtaba

    Mojtaba Guest

    Hi,

    I am very much interested in night photography. By that, I mean long,
    long exposures. Exposures like minutes and longer. I have always taken
    such pictures by manual film cameras and although it should be
    possible to find out the right exposure, I have always tried to guess.
    Consequently the results are very much by chance. From time to time I
    think about going digital and today I begin to wonder how this type of
    photo works in digital photography. I just wonder as most decent digi
    cameras feature an LCD monitors the photographer should be able to
    observe the process of shaping image. That way the photographer can
    see exactly when the exposure time in enough and stop the shutter at
    that time? Does is really work like that? Can someone who have actual
    experience tell me a little about such experiment?

    Regards,

    Mojtaba
     
    Mojtaba, Aug 16, 2004
    #1
    1. Advertisements

  2. Well, you won't be able to see the image as it comes along, but then
    again, you will get to view the final result when it is done on the LCD.
    No waiting for the film to develop and such.

    I've found that digital has made me a better judge at exposure times in
    different situations. Having the instant feedback of the LCD screen
    helps a lot, even though it might not be a "real time" view like you
    were hoping for.
     
    Brian C. Baird, Aug 16, 2004
    #2
    1. Advertisements

  3. Mojtaba

    Alan Meyer Guest

    I haven't found the LCD display to be very helpful when
    shooting at night. It gives you an instantaneous image of
    what it sees right now, not a building up image as light
    accumulates on the sensor. It has to work that way because
    if it allowed light to accumulate to build the image everything
    would be a blur as you move the camera.

    One of the difficulties of taking night photos with film
    is "reciprocity failure". Film responds to changes in light
    in a relatively linear way at the exposures that the film
    is most optimized for - i.e., subsecond exposures. At the
    extreme ends of high and low light, film response is no
    longer linear. So if, for example we can take a particular
    scene with a 10 second exposure, then cut the light in half,
    it might take a lot more than 20 seconds to make an
    equivalent exposure.

    I'd be very curious to know if there are analogous
    effects with digital sensors. That might be an important
    question for you.

    Another issue is "noise". Digital noise can look like
    grain on a digital image. It seems to be my experience
    that noise levels can increase at low light levels. But I
    don't know the theory behind it. This too is something
    you might need to know about if you want to do a lot
    of night work.

    Unfortunately, I don't know the answers to these
    questions, but maybe I've helped add some more
    specific questions for you.

    Perhaps one of our experts can give us the answers.

    Alan
     
    Alan Meyer, Aug 16, 2004
    #3
  4. Mojtaba

    Ian Stirling Guest

    CCDs don't do this.
    1/2 the light * 2 times the exposure = same picture.

    However the dark noise increases in direct proportion to the time too.
    For good CCDs this isn't a problem.

    For the best CCDs (not yet in cameras AIUI) the readout noise is under
    1 photoelectron, so they can with not too much penalty simply add short
    frames, instead of taking one long exposure.
    This has a number of benefits.
     
    Ian Stirling, Aug 16, 2004
    #4
  5. Amateur astronomers are doing just this (adding multiple
    shorter exposures) with DSLRs. The cameras of choice seem
    to be the Canon 10D, 300D and Nikon D70. With the canon,
    people are doing 1 to 6 minute exposures at iso 800.
    Here are some results with a 10D, adding 40 some 1-minute
    exposures:

    http://clarkvision.com/galleries/gallery.astrophoto-1

    This works better than film!

    Roger
     
    Roger N. Clark (change username to rnclark), Aug 17, 2004
    #5
  6. Mojtaba

    Ian Stirling Guest

    Not quite what I meant.

    To expand.
    A photon hits the sensitive part of a CCD, and if lucky (50-90% or so,
    depending on design) generates one electron that is then stored in the CCD
    until it's read out.

    AIUI, the current generation of DSLRs have readout noise well over
    1 electron per pixel.

    For example, if the average noise is 5 photoelectrons per pixel, and
    you add 10 frames, then you have a random noise that will average 50
    counts, but there will be significant numbers of pixels with 20 or
    30 counts different than the average.

    You'r not going to be able to reliably pick out a star that's put a total
    of 20 photoelectrons into one pixel over the many exposures.
    If the dark current is low enough, then one long exposure will let you
    pick it up.

    However, if you have an average readout noise that is much lower, you can
    stack frames without boosting the noise in the same way.

    Even if taking several frames boosts noise, it can be worth it.
    If you are taking a long exposure, and a plane flies over with running
    lights on, or a cloud passes over, or ..., then your exposure is
    ruined.
    If you are taking several shorter exposures, you can simply throw out
    bad ones, and add the rest together.
    Also, you can correct to an extent if your pointing is not perfect,
    by moving each frame slightly before addition.
     
    Ian Stirling, Aug 17, 2004
    #6
  7. There are two main sources of noise other than photon noise to worry
    about in obtaining long exposures: readout noise and dark current or
    dark noise. Readout noise is the uncertainty in the amount of charge in
    each pixel as it is being read out. This is typically a fixed amount,
    and clearly it is best to have a signal (total number of electrons in a
    pixel) that is far greater than the readout noise of the CCD. Thus, if
    one takes many short exposures, there is the danger that on each of them
    many pixels have noise that is dominated by readout noise rather than
    photon noise. Therefor it it is best to expose long enough so that the
    readout noise is of no consequence. I don't know what the readout noise
    is on consumer digital cameras, but I uspect it must be tens of
    electrons per pixel. A high grade astronomical CCD can have a readout
    noise of less than 2 electrons per pixel.

    The dark current will ultimately be the killer in long exposures. At
    room temperature a CCD is constantly generating large signals all on its
    on, and as the exposure goes on, this signal can completely fill up
    every pixel on the CCD. How long this will take depends on the
    particlar CCD, but exposures of several minutes must be the maximum for
    most cameras. The escape from this is to cool the CCD, and astronomers
    cool CCDs to near liquid nitrogen temperatures. Then dark currents as
    low as one electron per pixel per hour can be obtained. In these long
    exposures, however, another source of "noise"is highly prominent:
    signals generated by cosmic rays. Astronomers generally take more than
    one exposure and have automated cosmic ray recognition and removal software.

    So, for long exposure with a consumer digital camera, you have to trade
    off longer exposures with lower effects of readout noise, but higher
    dark current, vs. many shorter exposures with higher readout noise and
    lower dark current. It's simplest to use trial and error to figure out
    what is best unless you know the readout noise and dark current values
    your camera has. Then, given the brightness of a scene, you could
    calculate the optimum combination of number of exposures and expposure
    lengths to get the deisred result.

    CCDs do not suffer from reciprocity failure, and silicon is much more
    sensitive to light than film. Recent DSLR cameras have taken increasing
    advantage of the high native sensitivity of CCDs, though the "ISO"
    rating of the best astronomical CCDs is greater than 35,000.
     
    Joseph Miller, Aug 18, 2004
    #7
  8. I have quantified the signal to noise per exposure at:

    http://clarkvision.com/astro/canon-10d-signal-to-noise

    With multiple exposures the readout noise gets averaged
    as square root number of frames, wheres the signal builds
    as the number of frames. You can produce clean images of
    intensities less than the readout noise by adding
    multiple frames. It is being done every day by
    amateur astronomers. The differencees between cooled CCDs
    and DSLRs has become very small with this technique.

    Roger
     
    Roger N. Clark (change username to rnclark), Aug 19, 2004
    #8
  9. Mojtaba

    Ian Stirling Guest

    Fascinating page.

    Have you put a number on the absolute values of the noise?
    For example point it at a star, defocussed, with a known brightness, and
    measure the total signal, along with the lens details to get a ratio
    of pixel value to incoming photons.

    I'd do it, if someone'd donate a 10d :)
     
    Ian Stirling, Aug 19, 2004
    #9
  10. Exactly as you have stated is not complete, unless you you do a noise
    frame subtarction. In each frame the signal has noise as well, so in
    each frame, the noise is composed of both the readout noise and the
    photon noise. Imagine that on average you detected 10 photons per pixel
    in an exposureso that the photon noise is the square root of 10 in a
    pixel and the readout noise is 20. On average you will have a signal of
    30 electrons. Now add together 100 frames. The total will be about 3000
    electrons with an uncertainty of 55 electrons. A accurate measurement
    indeed! But roughly 2000 electrons in that signal were contributed by
    readout noise, while only 1000 were contributed by photons. That's a
    pretty low contrast image. But wait! You can measure or estimate the
    contribution of the readout noise to an accuracy of 45 electrons or much
    better if it's constant and you you take the time to measure it very
    carefully. That means you can SUBTRACT the readout noise from the final
    summed image and really win. The point is, by summing many frames, you
    measure both the signal and the readout noise to high accuracy. The
    photon noise and the readout noise both go down as the square root of
    the number of frames. Now many CCDs that amateurs use have such constant
    readout noise per pixel that they just need to do a simple subtraction
    of some average value, and they are done. Exactly the same thing is
    done with sky background subtraction, but in this case you have to
    measure the sky background very carefully while observing. I have been
    able to observe and obtain good measurements of objects that were only a
    few percent of the brightness of the sky itself by this method.
     
    Joseph Miller, Aug 19, 2004
    #10
  11. Yes, dark frames are also done and subtracted. There are "hot"
    pixels, those with higher dark current, so the dark subtraction
    step is important, as you suggest. If you look at my astrophoto
    page, you will see the dark subtraction mentioned.

    Roger
     
    Roger N. Clark (change username to rnclark), Aug 20, 2004
    #11
    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.