Detection probabilities in the presence of speckle noise

Dmitry Savransky (LLNL)

Many applications of adaptive optics (AO) involve binary classification
tasks (i.e., determining whether an object of interest is present in a
given image) or source separation tasks (determining how much of the light
in a given pixel is due to the object versus the background).  Multiple
approaches exist for both classes of tasks, but in order to objectively
assess the performance of a detection/extraction algorithm, it is
necessary to describe the sensitivity of the method, for example via a
receiver operating characteristic (ROC).  To do so, we must model not only
the statistics of the signal and noise in the image, but also the
post-processing applied to the image.  In this talk, I will present
modeling of the detection task in the presence of speckle noise, using
ground-based exoplanet imaging with adaptive optics as a motivating
example.  I will show the effect of the processing model on the
calculation of true and false positive rates, and briefly discuss some
applications of these derivations.