Нечіткий класифікатор зображення у великих динамічних базах даних

In today's rapidly growing visual information environment, the task of efficient image search and classification in large dynamic databases, which are updated daily with tens of thousands of new objects, is becoming especially relevant. Such databases are characterized not only by their significant size but also by a high degree of variability, which requires the development of algorithms capable of quickly and accurately recognizing distorted or modified versions of images in conditions of limited response time. The subject of study is a fuzzy classifier for clustering distorted versions of images in large dynamic databases. The aim of this work is to increase the accuracy of fast searches for distorted versions of images in large dynamic databases, in which the speed of adding information reaches 10-12 thousand images per day. Methods used: mathematical modeling, two-dimensional discrete cosine transform, image processing methods, decision-making methods, fuzzy mathematics. The following results were obtained. A fuzzy classifier for clustering distorted versions of images in large dynamic databases was developed. The experiments demonstrated that clustering distorted versions of images was sufficiently fast and cost-effective in terms of data volume and computational resource requirements. ROC analysis indicated the high quality of the developed fuzzy classifier.