STUDY OF STATISTICAL PROPERTIES OF THE BLOCK SUPPLY MODEL FOR A NUMBER OF DECORATORS OF KEY POINTS OF IMAGES

Context. Effective classification solutions in modern computer vision systems require an in-depth study of the nature of the processed data. The cluster representation for the basic system of structural features as a set of descriptors of key image points helps to reduce dimensionality and significantly simplify data analysis tools. The main tool is a statistical study of these descriptions as part of a cluster presentation, which reflects the generalized properties of a visual object. The implementation of the tree apparatus is based on a statistical analysis of data components to make a decision on assigning a visual object to the corresponding class. The construction of trees is based on indicators of informativeness of data that provide the logical processing process when dividing in tree branches. Having a single probabilistic nature, these indicators measure and evaluate information that is significantly different in content. It is important to study both the general properties of these criteria in the classification problem and the assessment of their individual characteristics. Objective. The solution of the problem of classifying visual objects according to the cluster representation of data for the structural description of the image using the apparatus of decision trees. Method. A method for classifying images based on a cluster representation of data using the apparatus of decision trees and tools of information theory is proposed. Results. The efficiency and effectiveness of the classification method is confirmed by applying the tree apparatus to the cluster representation of the structural image description data. Using examples of various informational content criteria for real experimental image data, the effectiveness of the created trees is estimated. The features of the introduction of various criteria for information content in the construction of a decision tree are analyzed comparatively. Conclusions. The application of the considered informational criteria in various ways sets the sequence for introducing independent variables in the classification tree, which are quantitative indicators of the cluster representation of the image description. The calculations show that the Shannon entropy and the Gini coefficient are quite powerful informational criteria that provide practical construction of a classification decision tree. The similarity of the joint informational function of the root node for different criteria confirms the objectivity of the study, and their difference reflects the individual nature of sensitivity to the analyzed data. The scientific novelty of the study is the improvement and statistical justification of the procedures for making classification decisions for cluster presentation data of image descriptions based on the introduction of tree models. The practical significance of the work is to confirm the effectiveness of the implementation of the tree apparatus for classifying data using examples of images in computer vision systems.