LIRA, the Low-counts Image Restoration and Analysis Package: a Teaching Version via "R"

Connors, Alanna

In low-count discrete photon imaging systems, such as in high energy astrophysics, the spatial distribution of a very few (or no!) photons per pixel can indeed carry important information about the shape of interesting emission. Our Low-counts Image Restoration and Analysis (LIRA) package was designed to: `deconvolve' unknown sky components; give a fully Poisson `goodness-of-fit' for any best-fit model; and quantify uncertainties on the existence and shape of unknown sky components. LIRA does this without resorting to chi-square or rebinning, which can lose high-resolution spatial information. However, since it combines a Poisson-specific multi-scale model for the sky with a full instrument response, within a full (Bayesian) probablility framework, sampled via MCMC --- running it thoughtfully requires understanding the underlying model and the employed computational methods.

To this end, we have created and are releasing a `teaching' version of LIRA. This is implemented in `R', and will be available through the standard website, cran.r-project.org. The accompanying tutorial and R-scripts step through all the basic analysis steps, from simple multi-scale representation and deconvolution; to model-testing; setting quantitative limits; and even simple ways of incorporating systematic uncertainties in the instrument response.

In this talk we will show movies of the imaging process in action with the aim of providing participants with a better understanding of the underlying structure and methods.

Return to oral presentation list