Filter designing for image processing based on multidimensional linear extrapolation

This article is proposed to develop a method of the synthesis of the prediction filter based on the multidimensional linear extrapolation to improve the accuracy of the pixel value prediction, as well as to assess the effectiveness of the proposed predictive filters depending on the prediction step and the dimension of training vectors to minimize the prediction error and the entropy of the differential signal. The article contain a solution of the special cases for a number of elements of the vector prediction, statistical dependence of the vector dimension and prediction filter order from the mean square error and the entropy of the differential signal are determined. Optimal prediction step is founded based on the experimental data and all the results are being analyzed.