REDUCING COMPUTATIONAL COSTS BY COMPRESSING THE STRUCTURAL DESCRIPTION IN IMAGE CLASSIFICATION METHODS
Volodymyr Gorokhovatskyi, Yurii Chmutov, Iryna Tvoroshenko, Oleg Kobylin · Advanced Information Systems · 2025
The research of the article is focused on ways to reduce the amount of analyzed data when applying image classification methods in computer vision systems. The aim of this work is to develop approaches to reduce the dimensionality of the vector description of the etalon base using metric granulation, which reduces computational costs and speeds up the classification process while maintaining a sufficient level of accuracy. Methods used: keypoint descriptors, metric data granulation apparatus, image classification and processing theory, data structures, software modeling. Results: the formalism of granular representation was developed; experimental modeling was carried out using five-level granulation, which reduced the time spent tenfold while maintaining high classification accuracy. In the comparative aspect, we studied ways to reduce the volume of vector descriptions based on data discarding, and researched the effect of the granularity level on the accuracy and classification time. The practical significance of the work is to improve the performance of image classification structural methods by implementing granularity and data discarding schemes, which provides much faster data processing without significant loss of classification performance.