OBJECT IDENTIFICATION AND CLASSIFICATION METHOD Russian patent published in 2019 - IPC G06K9/62 G06F17/30 

FIELD: physics.

SUBSTANCE: invention relates to methods for recognizing flat images of objects by their shape with extracting features of objects based on a contour analysis, with the subsequent processing of the extracted signs on the basis of statistical analysis, and can be used in technical vision systems. In the method, the classifier contains (N+1) predefined classes, the belonging of the object under study to N pre-defined classes for which the training sample is presented is determined in the indicated classifier by assessing the similarity by their physical characteristics, and the belonging of the object under study to the (N+1)^th pre-defined class for which the training set cannot be represented, is determined in the specified classifier by the fact that the object under study does not belong to any of the N pre-defined classes based on the assessment of similarity in their physical characteristics. To perform the physical similarity assessment in the specified classifier of at least one object under study with each of the N pre-defined classes that have a training set, carry out the construction of boundaries for each of the N pre-defined classes by analyzing the density and shape of the distribution of the training sample in the n-dimensional feature space.

EFFECT: technical result of the invention is to improve the accuracy of the classification of objects by providing filtering of noise objects.

1 cl, 4 tbl, 18 dwg