Celldetect identifies point sources in the presence of background. It can do so by three methods: estimating background locally, accepting a background map, or accepting a background value. The first approach is to examine the counts in a candidate source region and estimate the number of the counts in that region that are due to background. The candidate region is called the ``detect cell'' (Figure 1). The size of the cell is chosen to be some large fraction of the Point Spread Function (PSF) size. It is surrounded by a ``background frame'' of approximately equal area. For an isolated point source centered in the detect cell, the cell would contain source and background counts, while the frame would have only background counts. Using the counts in the frame, an estimate is computed for the number of counts in the cell which are due to background. These are subtracted from the cell counts. If the remaining detect cell counts are significantly higher than the estimated background counts, a source is detected.
When a background map or value is supplied to the tool, the background is not estimated and celldetect does not use a background frame.
Regardless of method, the cell is started in one corner of the dataset. It then ``slides'' repeatedly by one third its size, with a new source test made at each position. At the far end of the dataset, the cell slides down by one step and is repositioned horizontally, on the starting side.
This whole approach is essentially the ``Local Detect'' procedure described in DePonte & Primini (1993). The following sections outline advances built on that technique.
Modern X-ray detectors are large, and a spatially complete dataset can overwhelm the memory of a typical user's computer. However, the PSF usually grows with off-axis angle, and full resolution is not needed throughout the dataset. Celldetect starts analysis with a unblocked ``window'' at the center (aim point) of the dataset, then blocking down successively larger regions of the data into the same window (Figure 2). The analysis at each successive pass excludes the spatial extent of the previous pass. In each pass, the window is scanned at each of the cell sizes appropriate for that window. For a given instrument, the size of the window must be chosen large enough so that the blocked pixels still oversample the PSF size in later passes.
When the background frame extends over an area of low exposure (relative to the detect cell), the background counts will be artificially reduced (Figure 3). This will make the detect counts appear artificially significant. The consequence is that celldetect is prone to declaring false sources near detector edges and other regions of low exposure. To assist the user, celldetect provides an ``exposure ratio'' calculation for each detection. This is a ratio of the average exposure (in either seconds or effective collecting area) for pixels in the background frame to the average exposure for pixels in the detect cell. Under-illuminated backgrounds will cause the ratio to be less than unity. The source list can then be filtered on this value to exclude questionable detections.
The background statistics in celldetect change as the cell size increases. This means that a fixed value of the SNR threshold is not appropriate for best source detection. A SNR varying with cell size (and hence off-axis angle) gives better results (Figure 4). A varying SNR is under consideration for a future release of celldetect.
This project is supported by the Chandra X-ray Center under NASA contract NAS8-39073.
Harnden, F. R., Fabricant, D. G., Harris, D. E., & Schwartz, J. 1984, SAO Report No. 393
DePonte, J. & Primini, F. 1993, in ASP Conf. Ser., Vol. 52, Astronomical Data Analysis Software and Systems II, ed. R. J. Hanisch, R. J. V. Brissenden, & J. Barnes (San Francisco: ASP), 425
Dobrzycki, A., Jessop, H., Calderwood, T., & Harris, D. E. 2000, AAS/High Energy Astrophysics Division, 32, 2708