Detecting Planets from Direct-imaging Observations Using Common Spatial Pattern Filtering

Abstract

We present the application of common spatial pattern (CSP) filtering to the detection of point sources in high-contrast astronomical images. The data are preprocessed in two different ways: one copy of the data set keeps the point source in the same spatial location over many images through time, while a second copy of the data set moves the point source azimuthally through the image as in angular differential imaging. The differences between these two data sets are exploited via CSP filtering, highlighting the point source. We develop a forward model for this new process, which is then used to predict the result of applying this method to data with a point source at a particular spatial location. We validate the forward model numerically and use the prediction as template in a matched filter with the actual CSP modes. We present results for multiple sets of observational data and show that the new CSP forward model matched filter (FMMF) can compare favorably to the state-of-the-art Karhunen–Loève image processing FMMF reduction in terms of signal-to-noise ratio and unfavorably in other metrics.