Abstract

We describe a new method to achieve point-spread function (PSF) subtractions for high-contrast imaging using principal component analysis that is applicable to both point sources or extended objects (disks). Assuming a library of reference PSFs, a Karhunen–Loève transform of these references is used to create an orthogonal basis of eigenimages on which the science target is projected. For detection this approach provides comparable suppression to the Locally Optimized Combination of Images (LOCI) algorithm, albeit with increased robustness to the algorithm parameters and speed enhancement. For characterization of detected sources, the method enables forward modeling of astrophysical sources. This alleviates the biases in the astrometry and photometry of discovered faint sources, which are usually associated with LOCI-based PSF subtractions schemes. We illustrate the algorithm performance using archival Hubble Space Telescope images, but the approach may also be considered for ground-based data acquired with angular differential imaging or integral-field spectrographs.