RU2003174C1 - Method of pattern recognition - Google Patents

Description

relative to the center of the image with a constant angular pitch,

In FIG. 1 shows a reference image (Fig. 1, a) and the corresponding reference signal (Fig. 1, b); in FIG. 2 - a recognizable image (Fig. 2, a), its corresponding analyzed signal (Fig. 2, b) and the analyzed signal used for recognition (Fig. 2, c); in FIG. 3 is a structural diagram of a device.

As an example, for the reference image, an image of a flat figure of an arbitrary configuration with a concave part (Fig. 1, a) is selected. On the contour of this image, for ease of understanding, a small number of menu items (N 12) located relative to the center with a constant angular

about SPO

in increments of N- -30 °. In the proposed method, the value is much smaller, since the smaller the step, the greater the recognition accuracy. The reference periodic signal corresponding to the bypass of all selected contour elements (bypass time) is a sequence of discrete samples whose amplitudes are proportional to the distances from the center of the image to the corresponding distinguished elements of the image contour (Fig. 1, b).

For ease of understanding of the essence of the proposed method, the recognized (current) image is selected in the same configuration, the same scale as the reference image, and coincides with it in angular position (Fig. 2a). For the same purpose, the same number of elements is allocated on the contour of the recognized image. In the case of using, for example, television processing of a recognizable image, the selection of elements on its circuit is carried out by progressive scanning of the image (Fig. 2a). The analyzed signal obtained in this case is a sequence of discrete samples, the amplitudes of which are proportional to the distances from the center of the image to the elements of its contour located relative to the center of the image with a variable angular pitch (Fig. 2, b).

To convert this signal into a sequence of discrete samples corresponding to the contour elements of the recognized image located relative to its center with a constant angular step, the signal is interpolated (the interpolated signal is shown in

0

FIG. 2b by dashed lines) and from the interpolated signal, values corresponding to angles that are multiples of

 : () in the interval of the period ° 0-360. The analyzed signal thus obtained (Fig. 2c) is used later for recognition.

To ensure the independence of the results of the recognition of the image O from its scale, the reference and analyzed signals are normalized. For this purpose, the average values of the signals are determined and the amplitudes of all the bscounts of the analyzed signal are changed with the same proportionality coefficient until the average values of the analyzed and reference signals coincide.

Then, reference and analyzed signals are compared, for example, based on a difference algorithm, for which the difference modules of the corresponding signal samples are determined (for the same

value of the angle $ p & re where n sequentially vary from 1 to N), the modules are summed up and the resulting amount is compared with a predetermined threshold. If the sum exceeds the threshold, the analyzed signal is shifted by one step, and again, the sum value is determined. This operation can generally be performed N times. As soon as the sum is equal to the threshold (or less than the threshold), it is concluded that the recognized figure corresponds to the reference one and the recognition process is completed.

In the considered example, the number of selected elements on the contours of the reference and recognized images was chosen the same:. If it is necessary to recognize images with gaps, the number of selected elements on the contour of the recognized image M

JJ may be less than the number of elements in the outline of the reference image N (). However, the signal was interpolated based on an analysis of its values for neighboring contour elements, it is possible to determine the possible signal values for missing elements, i.e., to compensate for the loss of information about the recognized image as a result of gaps. The accuracy of such an interpolation is the higher, the smaller the angular step of the reference sequence of discrete samples Lu. Thus, the generated analyzed signal is independent of affine transformations and characterizes the features of the shape of the recognized figure

The proposed method can be implemented, for example, using a device (Fig. 3), which contains a television sensor 1, connected in series through the first input of the analog-to-digital OBOGO-loreobrazovatel (ADC) 2, the first input of the interface 3 With the input of the microcomputer 4, Second the input of the ADC 2 is connected to the second output of the television sensor 1, the third output of which is connected to the second input of the interface 3. The operation of the microcomputer 4 is performed in accordance with the developed algorithm.

The device operates as follows, the optical image of the recognized figure using the television sensor 1 is converted into a video signal, which is fed to the first input of the ADC 2, its second (gate) input receives clock pulses (from the second output of the television sensor) with a frequency determined by the accuracy converting the image into a digital signal, ultimately determined by the accuracy of recognition, Further

Claims (1)

The formula is invented by them. A METHOD FOR RECOGNIZING IMAGES, including generating a reference periodic signal with a period T equal to the time taken by the loop 30 of the reference image to be followed by a tracking device, converting the recognized image into an analyzed signal with a period T, comparing the analyzed signal with the reference, and deciding on the results of comparison, characterized in that that, in order to increase recognition accuracy, a reference signal is formed 35 as a sequence of discrete from 40 accounts, the temporary position of which corresponds to the moments of passage of the previously selected elements when the loop of the reference image is traced by a tracking device, and the amplitudes of the samples are proportional to the distances from the center of the reference image to the selected elements of its contour, highlight the center fifty fifty 5 0 5 0 the received codes of image brightness values are recorded via interface 3 in the RAM of microcomputer 4. The moment the recording starts in RAM is determined by the frame clock arriving at the second input of interface 3 from the third output of television sensor 1. Under the control of the bootloader located in the ROM of the microcomputer 4, the kernel of the operating system, for example, the OS-DVK, is loaded into the RAM of the machine from an external storage medium. Next, under the control of the operating system, the user program is downloaded. Under the control of the program, an array of information about the corresponding reference figure and the threshold value is loaded into the machine. Recognition results are displayed on the screen. (56) Copyright certificate of the USSR No. 446087, cl. G 06 K 9/00, 1970. USSR copyright certificate Mz 746610, cl. G 06 K 9/00, 1977. of the recognizable image and arbitrary elements on its circuit, the analyzed signal is generated when the circuit of the recognizable image is bypassed, the tracking device, like the reference signal, determines the average values of the reference and analyzed signals, change the amplitude of all samples of the analyzed signal with one and the same proportionality coefficient until the average values of the analyzed and reference signals coincide, the total difference between the amplitudes of the samples of the reference signal with the interpolated at the same time instants of the samples and the analyzed signal, the analyzed signal is shifted in time with respect to 1 reference signal to minimize the total difference in the amplitudes of their samples, the minimum value of the total difference in the amplitudes of the samples with a given threshold is compared and a decision is made based on the results of this comparison. 90 ° fSO ° Yu Phi & 1 muv; tf b) - so0 160 ° gto ° 5bo r Then 1 2 I I Figure 3 #