Deblending in DESDM E Bertin IAP E Bertin

Deblending in DESDM E. Bertin (IAP) E. Bertin DES Munich meeting 05/2010 1

Deblending • Detecting sub-components • Recovering objects from the sub -components • Forthcoming developments E. Bertin DES Munich meeting 05/2010 2

Deblending How sources are detected in SExtractor • 4 steps: – Sky background modeling and subtraction – Image filtering at the PSF scale (matched filter) – Thresholding and image segmentation – Merging and/or splitting of detections E. Bertin DES Munich meeting 05/2010 3

Deblending Detecting sub-components – Removal of noise peaks based on local constrast ratio • Photo (or DAOPhot): peak detection • IMCat: multiscale peak detection • SExtractor _PSF parameters: multiple PSF fitting with proximity constraints relative pixel value • SExtractor (or COSMOS): Multithresholding x E. Bertin DES Munich meeting 05/2010 4

Deblending IMCAT empircal multiscale approach Kaiser et al. 1995 E. Bertin DES Munich meeting 05/2010 5

Deblending Wavelet analysis • Extend the benefit of filtering from point-sources to very extended objects • Wavelet analysis: a data cube w( x, a) is obtained by correlating the image with the basis functions is localized, isotropic, and has zero mean. • The last difficult (yet unsolved) step is to connect the detections done at each scale to reconstruct the final object (Bijaoui & Rué 1995). • pyramidal median transform is an alternative to wavelet decomposition (Starck et al. 1995) Starck et al. 2000 E. Bertin DES Munich meeting 05/2010 6

Deblending Recovering objects • Sextractor: 1 pixel « belongs » to one object only – Pixels lying close to boundaries are reassociated to an object on a statistical basis (dithering) • Photo: flux fractions reassociated based on fits of « symmetrized » templates E. Bertin DES Munich meeting 05/2010 Lupton 2005 7

Deblending Image segmentation in SExtractor E. Bertin DES Munich meeting 05/2010 8

Deblending Suggested improvements • Drop the assumption: 1 source per pixel – Still looking for a way to do that in the cleanest way – Allow to do multiple source fits? • Try to deblend source blends that show no saddle in their profiles? • Multichannel deblending? • Metrics to measure deblending performance? – Cluster simulation in Sky. Maker E. Bertin DES Munich meeting 05/2010 9