METHOD AND DEVICE OF CHARACTERISTIC EXTRACTION Russian patent published in 2017 - IPC G06K9/46 

FIELD: physics.

SUBSTANCE: device realizes a method for extracting characteristics comprising the following stages: segmenting the image into a plurality of units, wherein each unit includes a plurality of cells, converting each cell from the spatial domain to a frequency domain, and extracting a characteristic of histogram of oriented gradient (HOG) of the image in the frequency domain. The stage of extracting the characteristic of histogram of oriented gradient (HOG) of the image in the frequency domain comprises the following sub-stages: calculating the magnitude of the gradient and the gradient direction of each cell in the frequency domain to obtain the descriptor of each cell, calculating the corresponding descriptor of each unit in the frequency domain to obtain the HOG characteristic of each unit, and calculating the HOG characteristic of each image unit in the frequency domain to obtain the HOG characteristic of the image. The stage of calculating the HOG characteristic of each image unit to obtain the HOG characteristic of the image comprises the sub-stages of adjusting the HOG characteristic of each unit in the image from the original L*1-dimensional vector to the M*N matrix, wherein each unit includes M*N pixels, and L=M*N, and the HOG image characteristic is obtained based on the corrected HOG characteristic of each unit and the corresponding position of each unit in the image.

EFFECT: providing the possibility of increasing the frequency and accuracy of detection in the recognition of structures.

11 cl, 11 dwg, 1 tbl