1 Introduction

X-ray astronomy has entered a new era now that Chandra and XMM-Newton are in orbit. Their high sensitivities and unprecedented image qualities bear great promises but also pose new challenges. In this paper, we outline problems of object detection in X-ray images that were not previously encountered. In doing so, we compare the performances of various detection techniques on simulated XMM-Newton test images, incorporating the main instrumental characteristics.

The X-ray observations consist of counting incoming photons one by one, recording their time of arrival, position and energy. Later, the event list is used to create images for a given pixel scale and energy band. Various X-ray telescope effects complicate this simple picture - the point spread function (PSF) and the telescope effective area (the vignetting effect), both dependent on the off-axis angle and incoming photon energy; detector effects like quantum efficiency variations, different zones not exposed to X-ray photons; environmental and background effects like solar flares and particle background. Even for relatively large exposures, the X-ray images could contain very few photons, and some sources could contain only a few tens of photons spread over a large area. Consequently, it is important for the source detection and characterization procedures to be able to cope with these difficulties.

As an example, the same hypothetical input situation is shown schematically in Fig. 1 for ROSAT, XMM-Newton and Chandra. XMM-Newton's rather large PSF, coupled with its higher sensitivity, leads to the detection of more objects but also to blending and source confusion, which become severe for long exposures depending on the energy band. Confusion problems in the hard band above 5 keV are less important, given the smaller number of objects and smaller count rate of energetic photons. Thus, we concentrate our analysis mainly on source detection problems for the more complicated case of the XMM-Newton energy bands below 5 keV.

Figure 1: Typical representation of objects seen by ROSAT-HRI, XMM-Newton-EPIC and Chandra-HRMA. The objects are represented by $\delta$-functions and folded by the corresponding instrumental PSF and efficiency. Full-width-at-half-maximum (FWHM) of the on-axis PSFs are $1\hbox{$.\!\!^{\prime\prime}$ }7$ for ROSAT-HRI, $0\hbox{$.\!\!^{\prime\prime}$ }5$ for Chandra-HRMA and $6\hbox {$^{\prime \prime }$ }$ for XMM-Newton-EPIC. The dotted line represents schematically the detection limit

Each X-ray mission provides data analysis packages - EXSAS for ROSAT  (Zimmermann et al. 1994), CIAO for Chandra (Dobrzycki et al. 1999) and XMM-Newton Science Analysis System (XMM-SAS). They include procedures for source detection, and in this paper we estimate and compare their performances on simulated images using various types of objects. These procedures make use of techniques such as Maximum Likelihood (ML), Wavelet Transformation (WT), Voronoi Tessellation and Percolation (VTP).

In Sect. 2 we describe the X-ray image simulations. A short presentation of the detection procedures is given in Sect. 3. Tests using only point sources are presented in Sect. 4, and extended sources in Sect. 5. We have analyzed realistic simulations of a shallow and a deep extragalactic field with only point sources in Sect. 6 and with extended objects in Sect. 7 for an exposure of 10 ks. Finally, we investigate the problems of confusion and completeness in two energy bands - [0.5-2] and [2-10] keV for two exposures - 10 ks and 100 ks (Sect. 8). Section 9 presents the conclusions. (H₀=50 kms^-1/Mpc, h=0.5, q₀=0.5 and $\Omega_0=1$ are used).