Space-variant recursive restoration of noisy images

A new approach to the modeling and filtering problems of nonhomogeneous images is proposed. The main assumption is that the whole image can be modeled by a collection of open subregions where the 2-D signal is regular enough to be well described by a smooth two-dimensional Gaussian process. The boundary points between two adjacent subregions are the image edges which represent sharp discontinuities in the signal distribution. A non-stationary state-space representation with structural information about image edges is then obtained. This allows the model to vary, adapting to the presence of spatial discontinuities. The reported simulation results show high filter performances.<
>