Consider an image characterized by its intensity distribution , corresponding to the observation of an object through an optical system. If the imaging system is linear and shift-invariant, the relation between the object and the image in the same coordinate frame is a convolution:
is the point spread function (PSF) of the imaging system, and is an additive noise. In Fourier space we have:
We want to determine knowing and . This inverse problem has led to a large amount of work, the main difficulties being the existence of: (i) a cut-off frequency of the PSF, and (ii) an intensity noise (see for example [6]).
Equation is always an ill-posed problem. This means that there is not a unique least-squares solution of minimal norm and a regularization is necessary.
The best restoration algorithms are generally iterative [24]. Van Cittert [41] proposed the following iteration:
where is a converging parameter generally taken as 1. In this equation, the object distribution is modified by adding a term proportional to the residual. But this algorithm diverges when we have noise [12]. Another iterative algorithm is provided by the minimization of the norm [21] and leads to:
where .
Tikhonov's regularization [40] consists of minimizing the term:
where H corresponds to a high-pass filter. This criterion contains two terms; the first one, , expresses fidelity to the data and the second one, , smoothness of the restored image. is the regularization parameter and represents the trade-off between fidelity to the data and the restored image smoothness. Finding the optimal value necessitates use of numeric techniques such as Cross-Validation [15] [14].
This method works well, but it is relatively long and produces smoothed images. This second point can be a real problem when we seek compact structures as is the case in astronomical imaging.
An iterative approach for computing maximum likelihood estimates may be used. The Lucy method [,,] uses such an iterative approach:
and
where is the conjugate of the PSF.