This paper addresses the problem of disturbance re jection for faint Poisson point-like measurements. The aberration function is approximated with a finite Zernike based expansion. We use nonlinear observers to estimate the aberrations and a linear quadratic regulator to reject the aberrations. Kalman filtering is compared with the phase diversity maximum a posteriori estimation. The approach presented here is beneficial for instance for 2-photon observations, single molecule observa tions or natural/laser guide star observations in astronomy. Index Terms-Aberration retrieval, Phase Diversity, LQR, Optimal Control, Kalman Filtering, Maximum a-posteriori I. IN TRODUCTION Literature presents us with many methods to estimate the point spread function (PSF) of an optical system [4], [12].However, in this paper we present an approach that can be used to control the PSF to a predefined shape. An example of such a shape could be a diffraction limited spot for astronomical observations [15] or a PSF with a small astigmatism aberra tion for single molecule localization [8]. Other examples are found in fluorescence scanning microscopy, where a diffraction limited PSF must be restored [2], [13].Our method employs the extended Kalman filter (EKF) [16] and the second order extended Kalman filter (EKF2) [6] as the observers, which are then used to control the system using a linear quadratic regulator (LQR) [19] and compares the results with the ones obtained from the same algorithm that uses the maximum a-posteriori (MAP) [9] estimator as the observer. A maximum likelihood solution has earlier been derived in [14], but the MAP estimator is closer to our proposed EKF approach. We decrease the computational complexity of the EKF [18] by approximating our system's PSF with a quadratic form. The classical approximation used to obtain this quadratic form is the Born approximation [4]. We use a more accurate approximation, that is the second order Taylor expansion of the PSF [10]. This has also the advantage of allowing us to use larger diversities than the classical approach. This assumption leads to a quadratic measurement equation. One important motivation to use the EKF is that we can use a pixel-wise recursion through the image instead of a vector recursion, and start processing the high intensity pixels until improvements are no longer made.We consider an optical system that images a biological sample. A microscope scans the optical sample point by pointThe authors are with the