Estimation of reconstruction error surfaces of non-stationary Gaussian images

This paper explores the reconstruction of non-stationary Gaussian images. The non-stationary image character is only given in space but not in time. The reconstruction method is specified on the basis of the conditional mean rule when the quantity of samples is limited. A spatial covariance function is proposed to describe the non-stationary nature of Gaussian image. This method provides the surface of the optimal reconstruction error functions in the whole space domain of the image. Furthermore, an easy statistical description of stationary or non-stationary image is directly obtained by this methodology in any spatial region with these properties.