Parameter identification for a two-dimensional image fieId†
Tohru Katayama, YASUO FUJII · International Journal of Systems Science · 1978
This paper presents a new approach to image enhancement based on a maximum likelihood identification method. It is assumed that the images are corrupted by a white gaussian noise field. A two-dimensional extension of the classical ARMA model is developed as a mathematical model for the image fields. Since the maximum likelihood identification leads to a parametric optimization problem, Davidon's algorithm is applied for numerical solutions. The advantage of the present method is that the enhanced images based on the predicted estimates are directly obtained from the noise-corruptod images, so that the autocovariance function of the original image is not required. To improve the quality of the enhanced images, a filtering algorithm is also derived. Digital simulation studies are carried out for various artificial images to show the feasibility of this approach.