Стиснення опису у структурних методах класифікації зображень

Abstract

The research focuses on methods for reducing the volume of analyzed data in image classification methods within computer vision systems. The goal is to develop approaches for reducing the dimensionality of the vector description of the reference base using metric granulation, which lowers computational costs and speeds up the classification process while maintaining a sufficient level of accuracy. Applied methods: keypoint descriptors, metric data granulation, classification and image processing theory, and software modeling. Obtained results: a formalism for granulated representation was developed , and experimental modeling using five-level granulation was carried out, which reduced time costs significantly while maintaining high classification accuracy. In a comparative aspect, methods for reducing the volume of vector descriptions based on data rejection were studied, and the influence of the granulation level on classification accuracy and time was investigated. The practical significance lies in increasing the performance of structural image classification methods by implementing granulation, which provides much faster data processing without significant loss of classification effectiveness.