Pylira: deconvolution of images in the presence of Poisson noise

Abstract

All physical and astronomical imaging observations are degraded by the finite angular resolution of the camera and telescope systems. The recovery of the true image is limited by both how well the instrument characteristics are known and by the magnitude of measurement noise. In the case of a high signal to noise ratio data, the image can be sharpened or “deconvolved” robustly by using established standard methods such as the Richardson-Lucy method. However, the situation changes for sparse data and the low signal to noise regime, such as those frequently encountered in X-ray and gamma-ray astronomy, where deconvolution leads inevitably to an amplification of noise and poorly reconstructed images. However, the results in this regime can be improved by making use of physically meaningful prior assumptions and statistically principled modeling techniques. One proposed method is the LIRA algorithm, which requires smoothness of the reconstructed image at multiple scales. In this contribution, we introduce a new python package called Pylira, which exposes the original C implementation of the LIRA algorithm to Python users. We briefly describe the package structure, development setup and show a Chandra as well as Fermi-LAT analysis example