Micro-pixel accuracy centroid displacement estimation and detector calibration

Abstract

Precise centroid estimation plays a critical role in accurate astrometry using telescope images. Conventional centroid estimation fits a template point spread function (PSF) to the image data. Because the PSF is typically not known to high accuracy due to wavefront aberrations and uncertainties in optical system, a simple Gaussian function is commonly used. PSF knowledge error leads to systematic errors in the conventional centroid estimation. In this paper, we present an accurate centroid estimation algorithm by reconstructing the PSF from well sampled (above Nyquist frequency) pixelated images. In the limit of an ideal focal plane array whose pixels have identical response function (no inter-pixel variation), this method can estimate centroid displacement between two 32$\times$32 images to sub-micropixel accuracy. Inter-pixel response variations exist in real detectors, {\it e.g.}~CCDs, which we can calibrate by measuring the pixel response of each pixel in Fourier space. The Fourier transforms of the inter-pixel variations of pixel response functions can be conveniently expressed in terms of powers of spatial wave numbers using their Taylor series expansions. Calibrating up to the third order terms in this expansion, we show that our centroid displacement estimation is accurate to a few micro-pixels using simulated data. This algorithm is applicable to the new proposed mission concept Nearest Earth Astrometry Telescope (NEAT) to achieve mirco-arcsecond accuracy in relative astrometry for detecting terrestrial exoplanets. This technology is also applicable to high precision photometry missions.