Last modified: 21 February 2024


The goal of the source validation pipeline is three-fold:

  1. Reconcile detections from wavdetect and mkvtbkg that form the list of compact sources.
  2. Validate extended convex hulls by estimating their likelihood.
  3. Define source bundles that are later passed to the MLE pipeline.

Reconcile wavdetect and mkvtbkg detections

As detailed in the description of the mkvtbkg algorithm, apart from creating background maps, the mkvtbkg tool is an alternative for compact source detection that outperforms wavdetect in regions of high background (e.g. for compact sources sitting on top of extended sources). During source validation, we assess whether sources are sitting on an extended emission region, and if that is the case, the mkvtbkg detections are given priority over the wavdetect detections, and recorded.

Validate extended convex hull sources

In a similar way as we do for compact sources in MLE, we estimate likelihoods of the convex hulls detections, parametrized using Gamma functions, which are the conjugate priors to Poisson distributions. The likelihood is evaluated using the position of the flux-weighted centroid and the region area.

Each convex hull source is flagged as true, marginal or false, according to the likelihood estimation. In this step, convex hulls are rejected if most of their counts are contained within compact sources that are inside the hull.

Define source bundles

When sources are close together, with their PSFs potentially overlapping, we must solve for their fluxes simultaneously. A bundle is a group of sources which contain overlaps with each other but are isolated from any other sources. We assign sources to bundles during the source validation pipeline.

All candidate sources are assigned to bundles (a bundle may be trivial and contain only one source). A blocking factor is defined for the bundle, such that the mean PSF full width half max is in the range 4 to 8 blocked pixels. This allows us to detect large, off-axis sources more efficiently.

The bundles are then passed to the MLE pipeline, and sources within a bundle are processed simultaneously.