RU2718172C1 - Method of searching objects on digital images - Google Patents

Abstract

FIELD: physics.

SUBSTANCE: invention relates to automation and computer engineering and can be used in designing vision systems for various purposes, as well as for analysis and synthesis of filters when processing digital images. Method of searching and recognizing objects on digital images is based on measurement of energy spectra of input image and reference, adaptive spatial filtration and threshold processing, in which matrices are formed from coefficients of decomposition of energy spectra of input and reference images into two-dimensional Fourier series of cosines, from obtained matrices, a discrete two-dimensional filter is synthesized and adaptive spatial discrete filtration of the input image is carried out, filtration result is compared to a threshold; according to the invention, the input image and the reference image are pre-processed with a detector field, and the aperture of the filter is matched with the reference dimensions.

EFFECT: high probability of correct detection and reduced employment of a computing resource and a memory resource.

1 cl, 1 dwg

Description

The invention relates to automation and computer technology, and can be used in the construction of vision systems for various purposes, as well as for analysis and synthesis of filters in digital image processing.

A known method for searching and recognizing objects in a digital image is based on a step-by-step combination of normalized images of recognizable objects centered and inscribed in the same cell size of the table of recognized objects, and images of patterns centered and inscribed in similar cells in the template table (see patent RU 2234127 C2, 06/05/2002 "The method of computer recognition of objects"). In this method, it is mandatory to have a preliminary normalization procedure (bringing the object in the image to its standard form - zooming, rotation to the desired position, etc.), centering, which leads to an increase in search time and, accordingly, recognition. And only then, line by line from bottom to top or side, alternately combine the rows or columns of the object table with the template table for final recognition.

There are also known methods for searching and recognizing objects in a digital image, based on the selection of contours of the same color on the image with their subsequent presentation in the form of Fourier transform coefficients (see, for example, US Patent 6563959 B1, 05/13/2003, "Perceptual similarity image retrieval method" ) and on the selection of a closed loop that matches the boundaries of the object as much as possible, followed by alternately overlaying the patterns stored in the computer’s memory (see patent RU 2250499 CI, November 17, 2003 “Method for computer recognition of objects”), where, if the contour matches, in advance of known patterns with objects discovered last recorded as recognized. In the first case, the degree of similarity of two objects in different images is determined by a number of factors, but this method is very sensitive to the quality of the input image and takes a lot of time to process it if the image of the object is presented on a complex inhomogeneous background. The disadvantage of the second method is the low probability of correct detection and recognition if the image of the object’s borders is blurred and have low contrast, i.e. there is an effect of the background characteristics on the characteristics of the object, and thus it can be difficult to distinguish the boundaries of only the object without background components, in addition, when the recognizable object is rotated with respect to the identical template, a decision can be made about unrecognition.

As a prototype, a method for searching and recognizing objects in digital images was selected (RF Patent RU 2458397 C1. Method for searching and recognizing objects in digital images, application No. 20111111070/08, 03/23/2011), based on the measurement of the energy spectra of the input image and the adaptive reference spatial filtering and threshold processing, in which the matrices are composed of the coefficients of the expansion of the energy spectra of the input and reference images in two-dimensional Fourier series in cosines, the resulting matrices are synthesized by retny two-dimensional filter that produces a discrete adaptive spatial filtering of the input image and comparing the result with a threshold filtering.

The disadvantage of the prototype is the low probability of correct detection, due to the ambiguity of the choice of filter aperture, as well as the requirement of significant computing resources and memory resource when processing high-resolution images.

The technical result of the proposed method is to increase the probability of correct detection and reduce the occupancy of the computing resource and memory resource.

The specified technical result is achieved by the fact that in the method of searching and recognizing objects in digital images, based on measuring the energy spectra of the input image and the reference, adaptive spatial filtering and threshold processing, in which matrices are formed from the coefficients of the expansion of the energy spectra of the input and reference images in two-dimensional series Fourier on cosines, on the received matrixes synthesize a discrete two-dimensional filter and produce adaptive spatial discrete phi filtering the input image, compare the filtering result with a threshold according to the invention, the input image and the image of the standard are pre-processed with a detector field, and the filter aperture is matched with the dimensions of the standard.

The essence of the proposed method lies in the fact that the input image and the image of the standard are pre-processed by the detector field, and the filter aperture is coordinated with the dimensions of the standard.

Processing by the detector field reduces the dimension of the input image and the image of the standard, which leads to a significant decrease in the aperture of the synthesized discrete two-dimensional filter and at the same time reduce the processing time (reduce the occupancy of the computing resource and memory resource).

With adaptive filtering, the filter aperture is usually not determined, and adaptability is that the impulse response is synthesized based on the measured coefficients of the expansion of the energy spectra of the input image in a two-dimensional Fourier series in cosines. The number of such coefficients is not set in advance, which leads to a decrease in the detection efficiency, and, as a consequence, to an increase in the probability of false alarms, and in many cases to a decrease in the probability of correct detection. In the proposed method, the number of coefficients of the expansion of the energy spectra of the input image in two-dimensional Fourier series in cosines is determined by the size of the image of the standard. At the same time, maximum processing efficiency and a potentially high probability of correct detection with a minimum probability of false alarms are achieved.

In the proposed method for searching for objects in digital images, in contrast to the prototype:

pre-processing the input image and the image of the standard with a detection field, the dimensions of the detectors of which are consistent with the dimensions of the standard. Image processing by the detector field as shown in [Ponomarev, A.V. Detector fields / Ponomarev A.V., Bogoslovsky A.V., Zhigulina I.V. M .: Radio engineering, No. 7, 2018, p. 129-136. DOI 10.18127 / j00338486-201807-23], leads to the formation of a contour image of a reduced dimension;

the aperture of a discrete two-dimensional spatial filter is determined by the size of the standard [see Bogoslovsky, A.V. Processing multidimensional signals. Prince 1. Linear multidimensional discrete signal processing. Methods of analysis and synthesis: monograph / A.V. Bogoslovsky, E.A. Theological, I.V. Zhigulina, V.A. Yakovlev M .: Radio engineering, 2013.168 s.].

The method can be implemented, for example, using a device whose structural diagram is shown in the figure. On the structural diagram are indicated:

1 - image sensor;

2 - detector shaper field;

3 - unit for measuring the components of the energy spectrum;

4 - calculator;

5 - discrete two-dimensional filter;

6 - a bank of images of standards;

7 - detector field detector;

8 - unit for measuring the components of the energy spectrum;

9 - threshold device;

10 is a display device.

Shapers of the detector field 2 and 7 are designed to obtain a contour image of reduced dimension and can be implemented on adders and microcontrollers.

The operation of the device is similar to the operation of the prototype device. The difference is that after the formation of the input image using the image sensor 1:

the image video signal is fed to the first input of the detector field detector 2, where it is processed by the detector field, the dimensions of the detectors of which are consistent with the dimensions of the standard coming from the second output of the image bank of the standards 6, as a result of which a contour image of reduced dimension is formed;

the video signal of the contour image of a reduced dimension is fed to the first input of the unit for measuring the components of the energy spectrum 3, the second input of which receives data on the size of the standard. In block 3, the components of the energy spectrum of the video signal of the input image are measured, the number of which is determined by the size of the standard;

from the first output of the image bank of the standards 6 to the input of the detector of the detector field 7 receives the video signal of the image of the standard, where it is processed by the detector field;

from the output of the detector field of the detector field 7, the contour image of the standard is fed to the input of the unit for measuring the components of the energy spectrum 8, where the components of the energy spectrum of the video signal of the contour image of the standard are measured;

from the output of the unit for measuring the components of the energy spectrum 8 to the second input of the calculator 4, the components of the energy spectrum of the video signal of the contour image of the standard are received, from the output of the unit for measuring the components of the energy spectrum 3 to the first input of the calculator 4 are the components of the energy spectrum of the video signal of the reduced image. In the calculator 4, the impulse response samples of the discrete spatial filter are determined.

from the output of the calculator 4, the samples of the impulse response of the discrete spatial filter are received to the second input of the discrete two-dimensional filter 5, where the contour image of a reduced dimension is filtered, coming from the output of the detector of detector field 2 to the first input of the discrete two-dimensional filter 5.

Application of the proposed method for searching for objects in digital images will reduce the amount of processed data by several orders of magnitude and increase the likelihood of correct detection of objects in digital images that have a complex inhomogeneous background.

The analysis of the prior art by the applicants, including a search by patent and scientific and technical sources of information and identification of sources containing information about analogues of the invention, allowed us to establish that the applicants did not find analogues that are characterized by features identical to all the essential features of the invention. Therefore, the claimed invention "Method of searching for objects in digital images" meets the criterion of "novelty."

Claims (1)

A method of searching for objects in digital images, based on measuring the energy spectra of the input image and the reference, adaptive spatial filtering and threshold processing, in which matrices are formed from the coefficients of the expansion of the energy spectra of the input and reference images in two-dimensional Fourier series in cosines, a discrete two-dimensional is synthesized from the resulting matrices filter and produce adaptive spatial discrete filtering of the input image, compare the filtering result with m, characterized in that the input image and the image of the standard are pre-processed with a detector field, and the filter aperture is matched with the dimensions of the standard.

Images