JPS607580A - Pattern matching device - Google Patents

Abstract

PURPOSE:To perform the pattern matching at a high speed and to eliminate the effect of an error of level variation due to the variation of illumination, etc., by providing a function to avoid the level variation of the signal of a picture having variable densities. CONSTITUTION:The picture output of a camera 1 is stored to a teaching picture memory 11 via a differentiation circuit 2 and an A/D converter 3. In the same way, the output of a measured picture is stored to a memory 12 for pictures to be collated. The pattern matching is carried out between the contents of the memory 11 and the picture to be collated and stored in the memory 12. The contents of the memory 11 and the data of the memory 12 are sent to latches 14 and 15 for each picture element, and the difference between both is obtained by a subtractor 16. Then the absolute of the output of the subtractor 16 is obtained by a circuit 17, and the sum total is obtained by a sum total adder circuit 18. The measured picture is shifted so as to minimize said sum total, and the result of decision is stored to a system memory 13.

Description

DETAILED DESCRIPTION OF THE INVENTION This invention relates to a pattern masoting device used for identification and positioning.

Is there a pattern manoching method for measuring images from photographs, TV cameras, etc.?The conventional method uses a computer and its internal memory to generate grayscale image signals of the target image and the matching pattern image. The difference in shading between the two was compared and identified, but the shading image was due to fluctuations in lighting.

etc., there were cases where incorrect judgments were made due to errors.

The present invention improves the method used by conventional computers in conventional pattern matching devices, adds a function to avoid signal level fluctuations of grayscale images, speeds up pattern matching, and improves lighting, etc. This device is capable of accurately determining the target image and the teaching pattern without being affected by level fluctuation errors due to fluctuations.

First, the conventional method will be explained with reference to FIGS. 1 to 6. For example, in Figure 1, there is an object A on the screen at a certain time,
In order to measure the location of this area after a certain period of time, a small area B that clearly shows the characteristics of the object is memorized. Next, from among the target images to be measured in Figure 2, move the same one as 13 from inside the screen, and the method is to finely shift the small head area B from the inside of the screen to the right and from top to bottom. That's it.

Verification is performed on the portion that overlaps with B. The position with the highest degree of matching as a result of verification is the position to which object A has moved. When the signal is a gray level signal, the conventional matching method is the 8SDA method - 8equentialSimilarity Det as shown in Figures 3 to 6.
A method such as a residual error sequential test method is used.

3 to 6 are shown one-dimensionally (four scanning lines) because of the 1hJ unit. This is the target image and matching pattern I3
This method calculates the absolute value of the difference in density for each pixel, and uses the sum of these absolute values within a small area to make a comparison judgment. If there is a complete match, the difference is zero and the sum within the small area is also zero. It is.

In the 88DA method, when the sum exceeds a certain value in the process of sequentially adding the sum within a small area, it is considered that there is no chance of a match, and the calculation is aborted, and the small area I3 is moved to the next position and the next "-1" is added.
Let's move on to calculation. Figure 3 is original! ! The teaching position of the small area 13 and the jI image are shown in the relationship between the position X of the scanning line and the 14 degrees Y of the image.

Figure 4 shows the measurement target li! FIG. 5 shows the determination data of the density Y when pattern matching is performed using 100 target images.

The sum of the absolute values of the differences within the small region B is represented by the area of the range indicated by 30. Figure 6 shows the small area of B as the target image.1 The sum of the absolute values of the differences between the target image and the taught pattern when moving from left to right over the entire surface is 10 knots, and the density is constant in areas where there is no image. However, at point D, the difference in density shows the minimum value, and at this point, the teaching pattern of verification pattern B has been found.

Next, we will discuss the drawbacks of this method. for example.

If the image becomes darker or brighter due to the ``A'' image imitation to be measured or due to changes in lighting, compared to the image when taught,
The gray level signal will have a DC-like offset. For this reason, the location where the sum of the absolute values of the differences is the minimum is not necessarily the location where the comparison was made, and may instead show a large value, or the sum may take the minimum value at another location, resulting in an erroneous determination. The circumstances surrounding this will be explained with reference to FIGS. 7 to 1O. FIG. 7 shows the relationship between the position and density of the original image, and FIG. 8 shows an example in which the gradation signal has a DC-like offset 32 in the image to be measured. Figure 9 is an example of pattern matching at point 0 and point D on the target screen, and the sum of the absolute values of differences within small area E is 81a at point 0 and 81b at point D.
Each area is shown as a range of area. Figure 10 shows small area E.
The judgment data of the sum of the absolute values of the differences when moving from left to right over the entire surface of the target image is plotted.The area with no image shows a constant density, and the 19 points are Offseno 1-32.
This example shows an example in which a large value is displayed instead of the minimum value due to the influence of

According to the above explanation, in the conventional method, if the level of the I...1 image to be measured changes in the taught image, it may become impossible to make a determination.

The present invention has been made to eliminate such problems, and is a pattern matching device that optically captures an object by scanning scanning lines, converts the grayscale image into a digital signal, and identifies it using a computer. A differentiation circuit that differentiates pixel signals of a grayscale image by scanning, an analog/digital converter that digitizes the 811 output of the differentiation circuit, and a teaching image memory that stores signals output from the analog/digital converter of the teaching image.

? , U A to-be-matched image memory that stores the analog/digital converter output signal of the matching image, a taught stroke number memory, and a to-be-matched lI! ! 11 Equipped with a subtracter that takes the difference between the signals stored in the elephant memory, a circuit that takes the absolute value of the difference caused by the subtracter, and an adder circuit that takes the sum of the absolute values in the small area taught. The present invention relates to a pattern matching device characterized in that the absolute value of the difference is taken for each small region of signals of shading and lightness lL'jj and compared.

Next, the content of the present invention will be explained in detail according to FIGS. 11 to 17.

FIG. 11 is a diagram showing the overall configuration of the present invention, and 1 is a part J containing strokes of a television camera or the like. 2 is a differentiation circuit that differentiates the analog stroke number signal. 3 is an analog decile converter (A/D converter) that converts this to digital.
It is. 35 Differential circuit 2. The A/D converter 3 is an A/I) converter 3. as shown in FIG. It may be changed to 36 consisting of the differential circuit 4. 4 in FIG. 12 is a difference circuit which takes the difference between adjacent digital data. 5 is a data bus of this device, and 6 is a signal line transmitted by an address bus. 7 is a generator that generates a synchronization signal at "tt L" for camera 1. 8 is a clock pulse generator for synchronizing the timing of the camera and the entire system. IO is the clock pulse generator of this device. 11 is a processor that controls the living body. 11 is a small region side 1 of I3 as shown in FIG.
Teaching to memorize the differentiation of elephants! ! ! This is an 11-elephant memory. 12 is.

The total σ ratio shown in FIG.
1α joint picture 1 interval memory. 9 is a memo according to the scanning signal of camera 1! This is a 71-res generating unit that generates a storage address when storing image data in Jll, +2. The memory 13 is a system memory for 10 crumbs, buffers, etc. for the professional staff 10. 14.1
5 is a day clutch.

Each teaching memo') 11. The image data from the illuminated Kongo 1 image memory 2 is temporarily held. I6 is a subtracter that calculates the difference between the two, and 17 is a circuit that calculates the absolute value.

Reference numeral 18 denotes a total sum addition circuit which adds the 17 absolute A-valued data to the 13 small areas and calculates the total sum.

] 9 is an output that outputs the determination result of whether the small area of 13 is located in the target image of [2] in the matched image memory;
It is 1.

20. Start up this device on the console and open the small teaching area ■3
It is a man-machine interface for configuring.

Next, the first point of the present invention in the above configuration will be explained with reference to FIGS. 11 to 17.

13 to 17 show the relationship between the position X and the density Y of one image, and FIG. 13 is the original image at the time of teaching.

FIG. 14 is a measurement target image on which a direct current off-cella 1-32 is shown. FIG. 15 shows the differentiated waveform of FIG. 13.

Although FIG. 15 is drawn in one dimension in FIG. 9, this may be a one-dimensional differentiation along the screen scanning line, or if necessary, since the nine screens are two-dimensional, it may be a two-dimensional differentiation. FIG. 16 is a differential waveform of the measurement target image in FIG. 14. If we teach the microhead MB to this differentiated waveform, we can search for the same region as this B in FIG. 16.

As in the conventional method, the sum of the absolute values of the differences becomes the minimum after moving the small region B, as shown in FIG. 17, and there is no misjudgment.

The relationship between the method described above and the overall configuration shown in FIG. 1I is as follows. In FIG. 11, the camera 1 includes a synchronizer generator 7, a chronopulse generator 8. The number of strokes is scanned from left to right, and the output of the number of strokes is differentiated by the differential circuit 2 or the differential circuit 4 in FIG.
/ D Converk 3 will be digitized. The digitized data is stored in each position of the teaching image memory 1 with an address to be stored designated by the address generating section 9. The small area 13 of the pattern to be determined is Konso/l/2
Specified from 0. On the other hand, as for the input of the measurement image, the image is scanned in the same procedure and stored in differentiated form in the reference image/shadow memory 12 designated by the 71-res generation section 9.

Pattern matching is taught! This is done between the contents of the I3 area stored in the image memory 11 and the matched image stored in the matched image memory 12. The data in the teaching image memory 11 and the data to be compared (I!L whale meko memory 12-data are sent pixel by pixel to latches 14 and 15, respectively, and the data is temporarily held.
Subtractor] 6 to calculate the difference between the two.

Next, the circuit I7 converts the output of the subtracter 16 into an absolute value.

The output of the circuit 17 is sent to a summation adding circuit 18 in order to calculate the summation over the entire B area.

When the total sum calculation at a specific position on the screen of the B area is completed, the B area is moved to the next position on the screen and the verification is repeated.

The summation or the result of the summation calculation obtained for each movement of all 13 areas, and the determination result of the decision to discontinue the summation calculation, are stored in the system memory 13.

The position where area B is set when the minimum value is taken among all the determination results is the verification-several points.

This minus number point output is output from the output port 19 to the host computer or display section.

The processor 10 is in charge of the above operations, and the non-quench generator 7. ri l: 17 ri generation part 8. Address generation section 9
performs control to import the number of strokes data into the teaching image memo IJII and the memory 12 to be compared.

Compared to the conventional method, the pattern matching device according to the method of the present invention described above differentiates the stroke number signal and matches the high frequency component (AC component) even if the target circle to be measured changes due to fluctuations in the illumination 1, etc. As a result, highly accurate judgments can now be made without being affected by fluctuations in the level signal.

[Brief explanation of the drawing]

FIG. 1 is a state explanatory diagram of an object A and a small area 130 image serving as a teaching pattern, and FIG. 2 is a state explanatory diagram of a state in which object A is scanned with a teaching pattern and compared and judged. Figure 3 is an explanatory diagram of the relationship between the position of the original image and small area B and the density of the grayscale image, Figure 4 is an explanatory diagram of the relationship between the position of the measurement image and the density, and Figure 5 is an illustration of the relationship between the position of the measured image and the density of the taught pattern. An explanatory diagram showing the absolute value of the difference in density within small area B, Fig. 6 is an explanatory diagram plotting the absolute value of the difference in density between the target image and teaching pattern 7 at all positions, and Fig. 7 is the original image. Figure 8 is an illustration of the density and position when the image to be measured has a DC offset. Figure 9 is the example of Figure 8 with 0 points and 0 points. An explanatory diagram of a state in which pattern matching is performed. Fig. 10 is an explanatory diagram of the state in which the absolute value of the difference in stitching depth is taken over the entire surface of Fig. 9 and the density is adjusted. Fig. 11 is an overall configuration diagram of the present invention. Fig. 12 is a differential circuit, A/ 13 to 17 are an example of pattern matching according to the method of the present invention, and FIG. 13 is an original image, and FIG.
The figure shows a measurement target circular image with a DC offset, Figure 15 shows a waveform obtained by dividing the waveform in Fig. 13, Fig. 16 shows a waveform obtained by differentiating the waveform in Fig. 14, and Fig. 17 shows the waveform shown in Fig. 15. An explanatory diagram of the density state when pattern matching is performed from the waveforms of FIG. 16 is shown. 1: Force 15, 2: e minute Fa, 3: A/D converter, 10: Procenoza, 11: Teaching image memory. 12: Matched image memo IJ, +6: Subtractor, I7: Absolute value calculation circuit, ]8: Total sum addition circuit.

Claims (1)

[Scope of Claim] A pattern matching device that optically captures an object using a scanning line, converts the grayscale image into a digital signal, and identifies it using a computer, which includes a differentiation circuit for differentiating pixel signals of a grayscale image obtained by scanning the scanline, and a 9m segment. An analog/digital converter that digitalizes the output of the circuit, a teaching image memory that stores the analog/digital converter output signal of the teaching image, and a matching image memory that stores the analog/digital converter output signal of the matching image. , a subtracter that takes the difference between the signals stored in the teaching image memory and the matched image memory, and a circuit that takes the absolute value of the difference value caused by the subtracter. What is claimed is: 1. A pattern matching device comprising: an adding circuit for calculating the sum of absolute values for each small region taught;