Класифікація зображень за матрицею відстаней для хеш-кодів даних

Abstract

The work is devoted to improving image classifiers using a distance matrixin the hash code space obtained by applying random hashing to 488-dimensional AKAZE descriptors. The proposed method uses an internal Hamming distance matrix, on the basis of which a histogram (distribution) is constructed. The use of distribution as a feature makes it possible to successfully classify images and ensures high accuracy.