3 Source detection

For the source detection and analysis, the HIIphot package, a robust and fully automated method developed by D. Thilker was used (Thilker et al. 2000). It was originally designed for determination of positions, fluxes and sizes of HII regions on continuum-subtracted ${\rm H}_{\alpha}$ images in galaxies. To our opinion, this program surpasses other existing algorithms in detecting overlapping sources, especially in crowded regions (e.g., DAOPHOT, Stetson 1987, or ALLFRAME, Stetson 1994 which have been tested for comparison as well). Applied to IRAS and ISO data, we benefit from several aspects of this code:

• the source detection is efficient and reproducible;
• no assumptions on the intrinsic source structure are required;
• the program provides adaptivity to the actual source morphology by making use of an iterative approach for growing sources from an initial guess at the shape and orientation;
• overlapping sources are accurately detected even in crowded fields and in the presence of a substantially inhomogeneous background (which is most important for the two MIR IRAS bands at 12 $\mu$m and 25 $\mu$m);
• the program offers the possibility of interactively selecting appropriate object and sky regions, a feature that helps to avoid a contamination of those fields considered for background estimation by possible source candidates.

The algorithm requires a number of parameters to be set initially which are then optimized by running the code several times for all data sets in the five IR bands (four IRAS bands and one ISO band). Most important are the final S/N value below which a detection is not accepted, and the final gradient in surface brightness in the vicinity of the source where the growth procedure stops. All relevant parameters for the different data sets are given below.

Initially, the program defines a rank list of possible detections, to which one out of six basic morphologies, ranging from Gaussian profiles to rings with different major-to-minor axis ratios, is assigned (for an extensive discussion on the construction of the source candidate list see Thilker et al. 2000). In a next step, so-called "footprint'' areas are constructed by "allowing'' the program to allocate pixel areas of the SMC input image according to the morphology assigned to the source (Fig. 4a) which may contain pixels which are not bright enough to remain in the final boundary of the source after the end of the growth procedure. For this reason, "seed'' regions are constructed by rejecting pixels falling below a certain median surface brightness limit of the initial "footprint'' region (Fig. 4b). In a third step, iterative growth starts: pixels are considered down to a limit equal to the outermost isophote, this limit is reduced by 0.02 dex in every iteration until a certain lower limit is reached (Fig. 4c). The program offers the option of making arbitrary selections for this limit where the surface brightness profile has become "sufficiently'' flat (see Fig. 4d).

The resolution of the IRAS HiRes maps is highly asymmetric, especially for the MIR range at 12 and 25 micron. Basically, this is the result of the rectangular detector mask shapes and the geometry of the scans covering the sky. Consequently, point sources appear elongated with the narrow dimension in the scan direction and the larger dimension determined by the cross-scan-size of the detector. Since we use IRAS HiRes data, it is very difficult to determine an effective resolution, however, the resolution of unenhanced coadded IRAS images of approximately $1'\times 5'$, $1'\times 5'$, $2'\times 5'$, and $4'\times 5'$ for the 12 $\mu$m, 25 $\mu$m, 60 $\mu$m, and 100 $\mu$m data provides us with an impression of which resolution changes may occur in our maps when proceeding towards longer wavelengths.

After the source catalogs were generated for every wavelength and the classification of the detected sources (the classification scheme is explained in detail in the next section), they were correlated with each other, i.e., we tried to identify sources in different catalogs within a certain correlation radius. We decided to choose a value of $r_{\rm corr}=90''$ due to the high spatial resolution of the ISO and IRAS data, though the average values for that radius were slightly larger in former studies (e.g., see Filipovic et al. 1998b who used $r_{\rm corr}=2.5'$ for the comparison between IRAS and radio data). This ensured the detection of all relevant source pairs in different wavelength bands and avoided a too large number of multiple correlations at the same time, a condition which constitutes the corresponding upper limit for $r_{\rm corr}$. As is clearly visible from the resulting tables which are presented in Appendices A-E, we never encountered more than 5 cross-identifications with more than one source in one band.

Figure 4: Illustration of the various stages of the HIIphot procedure for a special area. a) IRAS 100 $\mu$m image of the SMC with "footprint'' boundaries marked in white. Darker colors indicate higher flux densities. b) "Seed'' boundaries marked. c) The extents of the emitting regions are shown after growth to a certain terminal surface brightness. d) Same as in c), but for a state of growth to a terminal surface brightness well below the one used in c). Note that in this last state (to a smaller extent also already in plot c)) the boundaries do not only contain HII regions but also the associated colder dust.