I have a scientific application where I have a very bright image - basically a laser beam cross-section - and want to detect very small variations in brightness (around 0.01% or less) within it.

Under normal conditions with a linear sensor, if I set the exposure so that the maximum brightness was just below the saturation level and 'zero' represented total darkness, then I would need at least a 14-bit sensor to even see the sorts of variations I'm looking for (e.g. with a 14-bit sensor the theoretical maximum resolution in brightness is one part in 16,384 or 0.006%). To make a meaningful measurement of the variations I'd need at least a 16-bit sensor with a resolution of one part in 65536, or 0.0015%. In practice, however, I can't find a 16-bit sensor on the market at reasonable cost (I think there may be MFDBs with this resolution, but they have much greater spatial resolutions than I need and are prohibitively expensive).

In reality, however, I'm not at all interested in the region below 99.99% of maximum brightness, so what I'd really like to do is to use a standard camera's 8- or 10-bit range to cover the region from 99.99% to 100% of maximum brightness, in other words simply resetting the baseline so that anything below 99.99% of maximum appeared dark. Maybe I'm being thick-headed, but I can't think of a way of achieving this. Simply reducing the exposure time may make 99.99% to be 'dark' and 100% to be 'barely visible', but I'm not sure how I would then expand the range so that 100% became 'near saturation' again: just increasing the gain may work, but I'm sceptical because of the increase in noise this would entail.

With this many photons in play, there must be a way of doing it but I can't seem to figure it out. Does anyone have any suggestions?

  • 1
    Not to rain on your parade further, but have you thought about photon shot noise? Even if you have the ADC resolution to measure the variation you're looking for, I would think shot noise would be a problem. (Averaging over many pixels would help, if you don't need the full spatial resolution. Or over several exposures.) – coneslayer Apr 16 '13 at 0:58
  • 1
    I don't think this is really about photography at all. Detecting extremely small variations in laser beam brightness isn't about making pictures or anything related to it. – Please Read My Profile Apr 16 '13 at 17:05
  • Light, lens, image sensor, camera, image processing - sounds like photography to me. – Esa Paulasto Apr 16 '13 at 17:24
  • 1
    @EsaPaulasto "How to photograph the Aurora Borealis?" is about photography. The desired result is a photograph--something to be looked at and appreciated. Not a measurement. (My background is observational astronomy, and there's a clear difference between collecting data for measurement, and astrophotography as an aesthetic endeavor. Different goals, different methods, different results.) – coneslayer Apr 16 '13 at 20:34
  • 1
    "Experimental designs and results" and "Experimental technology used in physics or astronomy" are on-topic and physics.SE, BTW. – coneslayer Apr 16 '13 at 20:54

You can use curves to expand the dynamic range and make a very bright part of the image very dark and make a very very bright part still bright, but that isn't going to make more bits used on the very bright portion, just expose whatever detail is caught more clearly.

Even if you were somehow able to get the image processing to look only at the 99.99% to 100% intensities, noise is going to kill you. Unless the sensitivity and accuracy of the sensor is FAR FAR higher than the bit depth of the output normally allows, you are going to have all noise and no signal since the sensor likely can not detect such small variations reliably.

You actually want the opposite of high dynamic range, you want a very detailed but very VERY small dynamic range and the only real option for that is a purpose built sensor that is designed for detecting such minute differences accurately.

| improve this answer | |
  • You're right that what I really need is low (but controllable) dynamic range, rather than high! – Eos Pengwern Apr 16 '13 at 14:26
  • This is a good idea as suggested. In effect, attenuate the unnecessary radiation and distort the range of the peak. By the way, what is the energy distribution? This would be great for a "top hat" distribution but the tiny peak area of a gaussian one would have to be at greater magnification. Sounds like fun, I wish I was there. – Stan Sep 5 '13 at 2:30

Not the answer you're looking for? Browse other questions tagged or ask your own question.