Класифікація зображень з використанням мережі Хемінга

Abstract

In this work, a methodology for image classification is proposed, which combines BRISK extraction of binary descriptors, medoid-based aggregation, and a Hamming network with lateral inhibition. The keypoint descriptors are sorted by significance, enabling the formation of a single class prototype based on a majority rule. The activation of the Hamming network is calculated via the dot product, while lateral inhibition ensures the competitive selection of the most relevant neurons. Descriptor voting enhances the stability of the system when processing noisy data.