No 4 (2024)

The tasks of text image normalization arise simultaneously in several modules of the automatic document image recognition system. The paper presents a solution for two classical tasks of geometric normalization of a digital text image: compensation for the global document skew angle and slant elimination for its textual fragments. For both tasks, which differ in the type of geometric distortions, the solution is based on a single method of image analysis of the fast Hough transform. This method is specified and two algorithms for solving these problems are proposed, and they are tested: for the task of slant normalization – on a variety of both known dataset and on a specially collected and published dataset of Cyrillic fragments K_RUS, for the task of document skew normalization – on the popular DISEC dataset. It is shown that a distinctive feature of the proposed method is high speed with the ability to process a large range of angles, and the method itself can be successfully applied in systems for automatic processing of document images.

Graph models occupy an important place in tasks related to information security, including the construction of models and methods for managing the continuous operation of systems and system recovery, countering denials of service. Finite dynamic systems of complete graphs orientations are considered. States of a dynamic system are all possible orientations of a given complete graph, and evolutionary function transforms the given complete graph orientation by reversing all arcs that enter into sinks and there are no other differences between the given and the next digraphs. In this paper, the algorithm to calculate indices of system states is proposed. Namely, the index of the state is equal to 0 if it does not have a sink or its indegrees vector (a vector whose components are the degrees of entry of all its vertices located in descending order) contains all possible degrees of entry, otherwise its index is equal to the power of the largest set of consecutive degrees of entry, starting with the maximum possible degree, which is a subvector of its indegrees vector. As a consequence, the states with non-zero index belong to a basin with an attractor of length 1, whose generator state has a source and no sink. The maximal index of the states in the system is found. The corresponding tables are given for complete graphs with the number of vertices from 1 to 8 inclusive.

In the process of creating information systems at the enterprise level, we always have to solve the same problem to design a subschema, in the tables of which the lists of objects will be placed. Ideally, there should be one object table in the database, which will be used in all associations in which the objects participate. However, a wide range of requirements for the structural organization of objects lists, such as compactness, modifiability, semantic unambiguous identifiability, and a number of others, force the objects table to be partitioned into many local tables to accommodate in the latter objects with similar properties and identical appointments. The paper calls into question the dogma of compactness, which leads to an avalanche-like increase in the number of object’s tables, and as an alternative, the concept of controlled sparsity is put forward. Following this concept allows database creators, at the discretion of the user, to combine into one object type the set of local compact types. We proposed the structural framework, which includes two interrelated blocks: the block of objects’ types and metablock. The later contains meta-types, meta-relationships, and meta-constraints that collectively provide and guarantee the correctness of objects’ data placed in the database.

The article describes a solution to the problem of automating the process of segmentation of pulmonary nodules on computed tomography scans to expand the functionality of the previously developed module for determining the size and volume of pulmonary nodules. The main focus of the article is on comparing the accuracy of the models with the ResU-Net, Attention U-Net and Dense U-Net architectures when training on computed tomography images from the LIDC-IDRI dataset in their original form and using two proposed three-channel approaches to their preprocessing. For the three architectures considered, the DSC and IoU values in the ranges 0.8570–0.8735 and 0.7545–0.7881 were achieved. The best metric values were demonstrated by models trained on three-channel images with averaging. In such images, the first channel is represented by a scan in its original form, the second by an averaged scan, and the third by a scan to which anisotropic diffuse filtration is applied. The obtained results allow us to conclude that the use of preprocessing methods is promising for improving the accuracy of segmentation. The article also describes the training of the lung lobes segmentation model using data from the TotalSegmentator dataset. The input data of the modified software module are computed tomography scans, and its output data are processed images and a structured report (DICOM SR). This report, in addition to data on the size and volume of pulmonary nodules, contains information on the lobes in which the detected nodules are located.

The research is devoted to the development of an algorithm for estimating the convergence of stochastic Pareto optimization. The relevance of the work is due to the need to reduce the computational costs that arise with large multi-criteria calculations, where it is necessary to take into account many conflicting criteria to find optimal solutions. One of the problems in this context is finding a compromise between the accuracy of the Pareto front and the resources needed to calculate it. In multicriteria optimization, it is important to evaluate convergence in order to avoid an excessive number of iterations, which may be ineffective in terms of improving the result. The problem lies in finding the optimal number of iterations, at which the Pareto front reaches sufficient accuracy, and further iterations do not lead to a significant improvement in the quality of solutions. The aim of the study is to develop an algorithm that allows us to evaluate the convergence of the Pareto front and determine when it is possible to complete the optimization process without losing the quality of solutions. The results can be useful for specialists involved in multi-criteria optimization tasks and the development of algorithms based on stochastic conditions.

The paper presents an approach to the development of a new digital twin type. It offers to use closed-loop identifiers for generating point identification models based on associative knowledge. Procedures for calculating control actions in the conditions of possible abrupt changes of process operation modes are described.

The paper deals with the following essentially combinatorial problem: Given N points in ${\mathrm{ℝ}}^n$, compose the ellipsoid of minimum volume containing exactly N – k points where k is much less than N. Six algorithms for an approximate solution of this problem are proposed; they are based on certain heuristic considerations. Under various assumptions on the mechanism of generating the points and their amount, the comparative efficiency of the algorithms was conducted and the results of numerical experiments were presented.