Multi-scale decomposition of astronomical maps -- a constrained diffusion method

Abstract: We propose a new, efficient multi-scale method to decompose a map (or signal in general) into components maps that contain structures of different sizes. In the widely-used wave transform, artifacts containing negative values arise around regions with sharp transitions due to the application of band-limited filters. In our approach, the decomposition is achieved by solving a modified, non-linear version of the diffusion equation. This is inspired by the anisotropic diffusion methods, which establish the link between image filtering and partial differential equations. In our case, the artifact issue is addressed where the positivity of the decomposed images is guaranteed. Our new method is particularly suitable for signals which contain localized, non-linear features, as typical of astronomical observations. It can be used to study the multi-scale structures of astronomical maps quantitatively and should be useful in observation-related tasks such as background removal. We thus propose a new measure called the "scale spectrum", which describes how the image values distribute among different components in the scale space, to describe maps. The method allows for input arrays of an arbitrary number of dimensions, and a python3 implementation of the algorithms is included in the Appendix and available at https://gxli.github.io/Constrained-Diffusion-Decomposition/.