NEW LINEAR SPATIALLY INVARIANT IMAGE RESTORATION FILTER WITH NOISE SPECTRAL ESTIMATION.

A new linear spatially invariant filter is derived. It requires the 'rare information' about the observation noise, i. e. , its Fourier transform magnitude. This filter is essentially of closed form but the recursive method can be also utilized to get the stable solution at the given spatial frequency points if necessary. The compensation method of degradation of the restored image due to the error of the estimated noise Fourier transform magnitude is shown. The new filter is the deterministic version of a minimum mean square error (MMSE) filter under the same constraint condition. Experimental results are shown and compared to that of other standard filters.