A one-parametric reduced filter for image restoration

The linear algebraic restoration filters for discrete linear degradation models with additive noise are n/sup 2//spl times/n/sup 2/ matrices for n/spl times/n images in general. In this paper, a new reduced filter is derived which is realized by products of n/spl times/n matrices and for which the regularization parameter is easy to obtain. Although being computationally more economic, its restoring power has proved to be somewhat better than that of the two filters, with different dimensions, which are compared here. Each of the filters are regularizations of the circuit of the Gauss-Markov theorem, for example in the case of circulant matrices or white noise.