In this paper we present a restoration technique aimed at correcting image degradations by consideration of human visual criteria. A neural network model with an adaptive constraint factor is used. By considering local statistical information about regions within an image, the value of constraint factor can be selected which produces an optimal trade-off between noise suppression and edge preservation in each statistically homogeneous region. In addition a novel image error measure is presented which takes into account the statistical matching of homogeneous regions and its effect on human visual appraisal of image quality.