Guglielmetti, Fabrizia (2010): BackgroundSource separation in astronomical images with Bayesian Probability Theory. Dissertation, LMU München: Fakultät für Physik 

PDF
Guglielmetti_Fabrizia.pdf 24MB 
Abstract
In this work a new method for the detection of faint, both pointlike and extended, astronomical objects based on the integrated treatment of source and background signals is described. This technique is applied to public data obtained by imaging methods of highenergy observational astronomy in the Xray spectral regime. These data are usually employed to address current astrophysical problems, e.g. in the fields of stellar and galaxy evolution and the largescale structure of the universe. The typical problems encountered during the analysis of these data are: spatially varying cosmic background, large variety of source morphologies and intensities, data incompleteness, steep gradients in the data, and few photon counts per pixel. These problems are addressed with the developed technique. Previous methods extensively employed for the analysis of these data are, e.g., the sliding window and the wavelet based techniques. Both methods are known to suffer from: describing large variations in the background, detection of faint and extended sources and sources with complex morphologies. Large systematic errors in object photometry and loss of faint sources may occur with these techniques. The developed algorithm is based on Bayesian probability theory, which is a consistent probabilistic tool to solve an inverse problem for a given state of information. The information is given by a parameterized model for the background and prior information about source intensity distributions quantified by probability distributions. For the background estimation, the image data are not censored. The background rate is described by a twodimensional thinplate spline function. The background model is given by the product of the background rate and the exposure time which accounts for the variations of the integration time. Therefore, the background as well as effects like vignetting, variations of detector quantum efficiency and strong gradients in the exposure time are being handled properly which results in improved detections with respect to previous methods. Source probabilities are provided for individual pixels as well as for correlations of neighboring pixels in a multiresolution analysis. Consequently, the technique is able of detecting pointlike and extended sources and their complex morphologies. Furthermore, images of different spectral bands can be combined probabilistically to further increase the resolution in crowded regions. The developed method characterizes all detected sources in terms of position, number of source counts, and shape including uncertainties. The comparison with previous techniques shows that the developed method allows for an improved determination of background and source parameters. The method is applied to data obtained by the ROSAT and Chandra Xray observatories whereas particularly the detection of faint and extended sources is improved with respect to previous analyses. This lead to the discovery of new galaxy clusters and quasars in the Xray band which are confirmed in the optical regime using additional observational data. The new technique developed in this work is particularly suited to the identification of objects featuring extended emission like clusters of galaxies.
Dokumententyp:  Dissertation (Dissertation, LMU München) 

Keywords:  methods: data analysis, statistical techniques: image processing 
Themengebiete:  500 Naturwissenschaften und Mathematik
500 Naturwissenschaften und Mathematik > 530 Physik 
Fakultäten:  Fakultät für Physik 
Sprache der Dissertation:  Englisch 
Datum der mündlichen Prüfung:  20. November 2010 
1. Berichterstatter/in:  Böhringer, Hans 
URN des Dokumentes:  urn:nbn:de:bvb:19127320 
MD5 Prüfsumme der PDFDatei:  9a6184e1c3ed8a460d16323369864b6e 
Signatur der gedruckten Ausgabe:  0001/UMC 19293 
ID Code:  12732 
Eingestellt am:  15. Mrz. 2011 15:05 
Letzte Änderungen:  27. Okt. 2016 08:37 