Distance matrix for a set of structural description components as a tool for image classifier creating

Abstract

The subject of the paper is the methods of image classification in computer vision systems. The goal is the further development of structural classification methods in terms of introducing a system of classification features based on the values of the distance matrix for multidimensional description components. Applied methods: AKAZE keypoint detector, set theory and vector spaces methods, metric models for determining relevance for a set of multidimensional vectors, theory of data distribution formation, elements of probability theory, software modeling. Results: modifications of the image classification method based on the implementation of the formalism of distance matrices for a set of description components have been developed, integration models for the formation of classification features and actions on sets of vectors based on the distance matrix have been proposed, metric features of a set of multidimensional vectors as classification features have been established. The effectiveness of the developed modifications of the classifier depends on the choice of a subset and the number of descriptors in the description, a measure for comparing descriptions. Based on the introduction of the distance matrix, it was possible to form built-in features in the form of one-dimensional data distributions and reduce computational costs while ensuring the effectiveness of classification on the training data set. The practical significance of the work is the formation of classification models based on the distance matrix, confirming the performance of the proposed modifications using image examples, and creating a software application that applies the proposed classifiers in computer vision.