/lacosmicx

A fast implementation of the LA Cosmic algorithm

Primary LanguageC

lacosmicx

Name : Lacosmicx

Author : Curtis McCully

Date : October 2014

Optimized implementation of the LA Cosmic algorithm

Lacosmicx is designed to detect cosmic rays in images (numpy arrays), based on Pieter van Dokkum's L.A.Cosmic algorithm.

Much of this was originally adapted from cosmics.py written by Malte Tewes. I have ported all of the slow functions to Cython/C, and optimized where I can. This is designed to be as fast as possible so some of the readability has been sacrificed, specifically in the C code.

L.A.Cosmic = LAplacian Cosmic ray detection

If you use this code, please consider adding this repository address in a footnote: https://github.com/cmccully/lacosmicx

Please cite the original paper which can be found at: http://www.astro.yale.edu/dokkum/lacosmic/

van Dokkum 2001, PASP, 113, 789, 1420 (article : http://adsabs.harvard.edu/abs/2001PASP..113.1420V)

This code requires Cython, preferably version >= 0.21.

Parallelization is achieved using OpenMP. This code should compile (although the Cython files may have issues) using a compiler that does not support OMP, e.g. clang.

Notes

Differences from original LACosmic:

  • Automatic recognition of saturated stars, including their trails. This avoids treating such stars as large cosmic rays.

  • I have tried to optimize all of the code as much as possible while maintaining the integrity of the algorithm. One of the key speedups is to use a separable median filter instead of the true median filter. While these are not identical, they produce comparable results and the separable version is much faster.

  • This implementation is much faster than the Python by as much as a factor of 17 depending on the given parameters, even without running multiple threads. With multiple threads, this can be increased easily by another factor of 2. This implementation is much faster than the original IRAF version (orders of magnitude).

  • The arrays always must be C-contiguous, thus all loops are y outer, x inner. Note that this follows the Pyfits convention.