METHOD OF RECOGNIZING AND CLASSIFYING OBJECTS ON IMAGE Russian patent published in 2025 - IPC G06T7/00 G06F18/23 G06T7/11 

FIELD: digital image processing.

SUBSTANCE: method of recognizing and classifying objects on an image involves binarization of an initial image, selecting separate objects, analysing morphological features of objects and clustering. Coordinates of isolated objects are determined using a pre-trained convolutional neural network, forming an array of radius vectors from the geometric centre of the object to points of its contour, applying fast discrete Fourier transform to the array of radius vectors and obtaining values of spectral components, values of spectral components are used as morphological features of an object; the values of amplitudes of spectral components are clustered by the K-Means method with a given number of classes. If the number of classes on the image is unknown in advance, a range is set, the average value of the metric of the silhouette coefficient of objects is calculated for each value of the number of classes from the given range, the number of classes is determined, and the optimum number of clusters is selected.

EFFECT: possibility of classifying objects on an image that vary in shape, including when the number of classes is not known in advance.

1 cl, 6 dwg