Classification of Images of Visual Objects Based on Statistical Relevance Measures of Their Structural Descriptions

Abstract

A problem of classification of visual object images in the space of attributes of descriptors of special points is solved with representation of the description in cluster form and using of statistical measures to calculate relevance of descriptions. The analysis of specific application feature of statistical and metric classifiers in determining the level of relevance of structural descriptions is performed. Comparison of the characteristics of relevance measures on the calculated examples is performed. The Kullback-Leibler divergence is proposed to use as a universal and effective measure for the classification problem. The effectiveness of the proposed approach for application image dataset is shown.