Filter design for the detection of compact sources based on the Neyman-Pearson detector

Abstract

This paper considers the problem of compact source detection on a Gaussian background. We present a one-dimensional treatment (though a generalization to two or more dimensions is possible). Two relevant aspects of this problem are considered: the design of the detector and the filtering of the data. Our detection scheme is based on local maxima and it takes into account not only the amplitude but also the curvature of the maxima. A Neyman–Pearson test is used to define the region of acceptance, which is given by a sufficient linear detector that is independent of the amplitude distribution of the sources. We study how detection can be enhanced by means of linear filters with a scaling parameter, and compare some filters that have been proposed in the literature [the Mexican hat wavelet, the matched filter (MF) and the scale-adaptive filter (SAF)]. We also introduce a new filter, which depends on two free parameters (the biparametric scale-adaptive filter, BSAF). The value of these two parameters can be determined, given the a priori probability density function of the amplitudes of the sources, such that the filter optimizes the performance of the detector in the sense that it gives the maximum number of real detections once it has fixed the number density of spurious sources. The new filter includes as particular cases the standard MF and the SAF. As a result of its design, the BSAF outperforms these filters. The combination of a detection scheme that includes information on the curvature and a flexible filter that incorporates two free parameters (one of them a scaling parameter) improves significantly the number of detections in some interesting cases. In particular, for the case of weak sources embedded in white noise, the improvement with respect to the standard MF is of the order of 40 per cent. Finally, an estimation of the amplitude of the source (most probable value) is introduced and it is proven that such an estimator is unbiased and has maximum efficiency. We perform numerical simulations to test these theoretical ideas in a practical example and conclude that the results of the simulations agree with the analytical results.