SU446087A1 - The way to recognize images of objects - Google Patents

Description

(54) METHOD FOR IDENTIFYING IMAGES OF OBJECTS

This invention relates to the field of enanavia; samples and can be used to identify multi-gradation and two-gradation images of objects

in particular for the identification of species

,

; the availability of micro-objects using television technology.

There are known methods for identifying objects, according to which a field is decomposed into a raster, and

raster with an identifiable image, convert the image into a video signal, generate sample signals and compare them with the reference signals.

However, when reading multi-gradation images rotated and deformed relative to the standard, such methods require a large amount of digital calculations and do not provide sufficient reliability of identification.

The aim of the invention is to accelerate the process of identifying these images without significantly complicating the hardware implementation of the method.

This goal is achieved by the fact that after selecting a raster with an identifiable image, the reading raster rotates, the video signals of all rows are summed in a frame direction at each position of the reading raster, and the total video signals are compared with the reference signals.

FIG. 1 shows three objects A,

B, C on the television raster; in fig. 2, a - object A;

I. 6 - video signals of the lines of the object A; c - total video signal

- the image of the object A. In FIG. 3 shows three objects A,

B, C on the TV raster and total video signals are samples of objects A, B, C;

in fig. 4, a — three objects A, B, C on a television raster rotated, 90 relative to objects compared to the raster in FIG. 3; b - total video signal samples of objects A, B, C. In FIG. 5 given device; realizes gedledenny method. The method of obtaining the total signal is as follows. FIG. 1 shows a television raster with its assigned coordinates: X - along the line and y - frame by frame. In the frame there are images of three objects (the images are two-gradation, the background has the brightness O, the object is single) of the figures (A is a triangle, B is the letter T. C is the letter P). Figure A begins on the line and ends on the line, which takes k lines and its polo. The x-axis is defined by the x-chr. All sections of the lines occupied by object A are located inside this segment j. In FIG. 2 shows a section of a raster with figure A and shows the shapes of the video signals from this figure on all lines from y to UGP + -K. Denoting the video signal span at the point with X, located on CTpOKeJKaK, you can record that the total video signal from object A for a point with the coordinate x i is equal to v-.j Thus, in this formula, you can find the total video signal for any point X {, in the interval. The view of the total video signal for FIG. A is shown in FIG. 2b, I sum up the video signals line by line for figure C from line Ur over line ll points V., and for figure B, on frame 1I1 interval about 4 over line y. in the first tier, we get the total video signal samples shown for these figures in FIG. 3. From the configurations of these total signal-images, we can conclude that they have completely characteristic forms allowing to classify these objects. To make the description invariant to blends, rotations and deformations of the image, you need to get several total video signals at different yiviax on the raster gate transmission tube relative to the image projected on its photocathode. FIG. 4, and the original image is shown, but the raster of the transmitting tube is orthogonal to the raster in FIG. 1, and the summed video signals for the objects A, B, C for this case are shown in FIG. 4, b, From a comparison of images ft. 3 and 4, it can be seen that the images of the object A in both cases are different from the images of the objects B and i C, and the images of the objects B and C are chunked, | raster, as in FIG. 4, they do not differ from each other, i.e., these objects can only be classified by total video signals obtained for a raster orthogonal to the raster in FIG. 4, Reduction of the description of an identifiable object by this method; occurs due to the replacement of the three-dimensional video signal of the original image by the two-dimensional total video signal-image. Each total video signal can be described by several signs, such as the number of vertices and their position relative to the beginning and end of the video signal, the steepness of the slopes and the position of the slopes relative to the vertices, etc. Moreover, these signs are conveniently represented in digital form. For multiple, 1 boundary images, the video signals are summed up using the above formula and receive features in the same way. The method can be implemented in the device (Fig. 5), containing a television sensor 1, an image selector of objects: 2, a summation unit 3, a block of video signal recognition 4, a control unit: laziness 5, a characteristic feature block 6, and re-; A smart block 7. Signs of patterns 6 introduce signs of total video signals in digital form, which are determined experimentally for each class of identifiable objects; The television sensor 1 sends the video signal to the image selector of objects 2, the mountain COP, and selects a raster area occupied by an image of an identifiable object. The video signal from the selected section of the raster is fed to block 3, which can be built using a storage electron-storage tube, for example, type LN-17. In this block, the video signals of all lines from the highlighted are summed; the raster section, which is accomplished by turning off the frame scan on the memory tube. At the same time, the recording beam draws the same path for all lines of the frame onto the targets of the memory tube. The selector 2 opens the recording beam at those times when the beam of the transmitting tube reads the image of the object, and thus the video signals with a section of lines superimposed on the same track of the memory tube

divided by the selector 2, i.e., the target is recorded as potential; video relief, the magnitude of which at each point is expressed by the formula Bbt. Summation takes place simultaneously with the image scanning process. Therefore, the device operates in real scale.

time with a television sensor 1. During the return run of the frame p 13 of the transmitting tube, the total video signal received at the target of the fflnar tube tube is read in analog form and is transmitted to the feature block of the total video signal 4, where the features shown earlier are measured. If it is necessary to obtain several summed video signals at different angles of rotation, the control unit 5 performs this operation. The same block provides control over the operation of comparing the attributes of an identifiable object with the features of the standards and making a decision about classifying the object to a particular class, which is provided by the p & 7 block.

This method allows to identify images of objects, arbitrarily located X, h%

framed using a television system with a scanning sweep standard as the sensor. The method can also be implemented with any scanning system: discrete-step, radia, shno-circular, spiral, etc.

As a totalizer, digital totalizers or accumulative analog adders can be used.

: Subject of invention

A method of recognizing images of objects by decomposing an image field into a raster, selecting a raster section with an identifiable image, converting an image into a video signal, forming image signals and comparing them with reference signals, in order to accelerate the recognition process multi-gradation images, rotated and deformed relative to the standard, summed up the video signals of all the lines in the direction, frame by frame; at each position of the rotating paci i pa and compare the total video signals with the reference signals. h

1%

g Xt Xz, X3 X / fX5 X

OX, X

 2

#V

1 Xj X,

x

Object A

Object B

Xj hb

Object C

FIG.