2011A&A...530A.133M

Large-scale astronomical surveys from ground-based as well as space-borne facilities have always posed significant challenges concerning the problem of automatic extraction and flux estimate of sources. The recent explosion of surveys in the mid-and far infrared, as well as in the sub-millimeter, brings an increase to the complexity of the source extraction and photometry task because of the extraordinary level of foreground/background due to the thermal emission of cosmic cold dust. The maximum complexity is likely reached in star-forming regions and on the Galactic plane, where the emission from cold dust is dominant. We present a new method for detecting and measuring compact sources in conditions of intense and highly variable fore/background. While all the most commonly used packages carry out the source detection over the signal image, our proposed method builds a ``curvature'' image from the measured image by double-differentiation in four different directions. In this way point-like and resolved, yet relatively compact, objects are easily revealed, while the slower varying fore/background is greatly diminished. Candidate sources are then identified by looking for pixels where the curvature exceeds a given threshold in absolute terms, and the methodology allows us to easily pinpoint breakpoints in the source brightness profile and then derive reliable guesses for the sources' extent. Identified peaks are fit with 2D elliptical Gaussians plus an underlying planar inclined plateau, with mild constraints on size and orientation. Mutually contaminating sources are fit with multiple Gaussians simultaneously using flexible constraints. We ran our method on simulated large-scale fields with 1000 sources of different peak flux overlaid on a realistic realization of diffuse background. We find detection rates in excess of 90% for sources with peak fluxes above the 3σ signal noise limit, and for about 80% of the sources the recovered peak fluxes are within 30% of their input values.