Dynamic scale adaptation algorithm of image etalon functions

Abstract

An algorithm for large-scale adaptation of prototype functions representing image classes is proposed. The algorithm identifies the parameters of nonlinear scale distortions contained in the functions representing the observed image realizations, and then transforms the original prototype function using the previously proposed model. The algorithm works on a class of images of regular phase processes that have the property of quasi-similarity of shape. Two models of large-scale nonlinear transformations are considered: symmetric and asymmetric. The differences between the models and the practical results of their application are given. The algorithm was experimentally tested on the images of prototypes of fragments of speech signals, electrocardiosignals, and engine cylinder pressure detector signals. Examples and experimental data confirming the effectiveness of the algorithm are given. Conclusions are formulated about the possibility of using the algorithm with both models in practical problems.