THE APPLICATION OF NON-PARAMETRIC STATISTICS METHODS IN IMAGE CLASSIFIERS BASED ON STRUCTURAL DESCRIPTION COMPONENTS

Abstract

The results of the solution of image classification problems in computer vision systems using structural methods are presented. The models of statistical data estimation during the image relevance definition and classification were developed. A component-by-component presentation of descriptions with the estimation of mathematical expectation and variance values was proposed. The integrated criteria of data relevance based on these estimated values follow and implemented. The mathematical and software data processing and analysis models for the classification based on the structural description were developed, the properties of the application of these models were investigated, and the analysis of image processing efficiency was performed. The efficiency of classification using statistical estimations for the system of components for descriptors has been experimentally confirmed. Based on the research, a conclusion about the effectiveness of classification technologies using non-parametric evaluation for etalon data for the compression of the feature space and computational cost reduction was reached.