D600 High ISO in DX

Started Nov 23, 2012 | Questions thread
Joofa Senior Member • Posts: 2,655
Re: On random processes

Leo360 wrote:

The absence of good information on the priors is unsettling. For example, conditional mean estimator (at least the way I derived it) depends upon an unknown mean electron count <n_e> (or equivalently mean photon count <n_ph>). "Naive estimator" solves the problem by plugging in the observed value of n_e in the palce of the mean. The price of this "ansatz" is a diverging variance if QE goes to zero. Can we do better? For instance:

  1. Can the output of the metering sensor be used to extract some hint on the priors for the incoming photon flux?
  2. Or alternatively, would it be better to do a joint estimation of the photon counts across adjacent pixels?

I think it is possible to do better. The prior knowledge in this case is the mean value of photons.

If the sensor is uniformly illuminated with a constant intensity then the mean number of electrons accumulated can be worked out. With the knowledge of QE (say provided by the manufucturer) one should be able to derive the mean number of photons.

For a real image or photograph, which is not just a simple uniform illumination, say a picture of a cat, the situation is tricky. Because, the Poisson statistics are changing spatially - a single image is just a random sample from a non-stationary Poisson process. I have some personal thoughts on how to proceed here, though, I have to ascertain first that they are a good approximation before I proceed with them.

-- hide signature --
Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow