Data Processing: Source Detection
Source detection is performed in raw detector coordinates in order to make the difficult task of quality flagging as trivial as possible. Quality issues such as readout bleeding, diffraction spikes and ghost rings are most easily propagated from one source to the next in detector coordinates before rotating, undistorting and translating the image to sky coordinates. The measured detector coordinates of each source are transformed to sky coordinates in a later process (Sec 2.6).
Although the intrinsic background in an XMM-OM image is generally uniform, the background count level is estimated iteratively for each source by rejecting bad pixels stored within the quality map and pixel outliers ±2σ from the median value within a box 100 / δx + 1 on a side, where δx is the pixel binning factor. Once converged, the mean background is calculated from the remaining pixels, which is a better measure than the median when UV background count rates are typically of order unity.
All pixels with values above a background threshold of 2σ are
considered to potentially contain source photons. Neighbouring pixels
which meet this criteria are treated as the same source. Based upon
their extent, each cluster of pixels across the image are flagged as
either point-like (consistent with the XMM-OM point spread function)
or extended. 2-dimensional Gaussian moments are used to calculate the
centroid position, Full-Width Half-Maxmima of the major,
Fmaj, and minor, Fmin, axes (which
should in principle be equal within uncertainties) and the orientation
of the major and minor axes for each source. Sources are considered
point-like if they obey one of the criteria:
- Fmaj − 3δFmaj > Fpsf,maj
- The central pixel contains a count level > 5,000.
Fpsf,maj is the Full-Width Half Maximum of the image-appropriate point spread function, reconstructed from data within the Calibration Access Layer.
The above steps are repeated after all point sources have been removed from the pixel sample, albeit in this iteration all detected source candidates are now flagged as extended. For extended-sources, any pixel already flagged as belonging to a source is rejected from the background calculation and if the number of pixels associated with a single source exceeds 200, then the dimensions of the background box are increased to 120 / δx + 1. Extended source positions, extent and orientation are again calculated using Gaussian moments.
The credibility of each detected source is tested by the following
criteria:
- The centroid of a source must be contained within the cluster of pixels associated with the source.
- The significance of a source must be > 3 in a 5.7 arcsec radius aperture.
- If the significance in a 5.7 arcsec radius aperture is < 10 the source must have a significance > 3 in a 2.8 arcsec radius aperture.
Extended sources are further tested for confusion with scattered light features which can originate from bright off-axis objects, 12.1−13.0 arcmin from the OM boresight, scattering of a chamfer in the detector window housing. These scattered light features are annular rather than centrally concentrated. For sources containing > 1,000/δx pixels, we determine the maximum and minimum pixel extent along both dimensions of the detector array. If nhole > nsrc / 7, where nhole is the number of pixels within the 'hole' at the center of the annular source and nsrc is the total number of pixels contained within the source itself, then the source is rejected as being a scattered light feature. Although arbitrary, trials have proved this criterion to be effective at removing scattered features from the source list.