Douglas Hope (Institute for Astronomy, University of Hawaii), Stuart M Jefferies (Institute for Astronomy, University of Hawaii)
Keywords: Imaging
Abstract:
An important problem in imaging is the inverse imaging problem, i.e. given a noisy blurred image of an object and knowledge about the imaging system, obtain a high-fidelity estimate of the object. When the image is formed under isoplanatic conditions, the image can be modeled as a convolution of the object intensity distribution and the point-spread-function (PSF) of the imaging system. In practice, the PSFs are unknown or poorly known, so one must estimate both the object and the PSF simultaneously, a problem known as multi-frame blind deconvolution (MFBD).
An important requirement for solving the MFBD problem is the use of prior information about the object, the instrumentation and the physics of the observing conditions. Examples of such information include a positivity constraint on the object, a constraint on its spatial extent (commonly referred to as support), and any partial knowledge of the PSF, possibly obtained from a wave-front sensor. All such information is pertinent and should be used to constrain the solution space that must be searched by the algorithm.
We propose new constraints for the MFBD restoration of large ensembles of atmospherically degraded imagery. These constraints, defined in the Fourier domain, essentially restrict the number of spatial frequencies in the object and PSF estimates. A natural outcome of applying this spectral support is a robust regularization method that greatly improves the estimates of both the object and PSFs. We present a method for extracting spectral supports from data and show how to implement them in two different cases of MFBD, 1) images where only part of the object falls on the detector and 2) images where the object falls fully on the detector.
%Z Stockham,T.G., Jr., Cannon, T.M., & Ingebretson,R.B., Proc. IEEE, 63, 678, 1975
Date of Conference: September 10-14, 2006
Track: Imaging