JPS6246387A - Method of pattern matching - Google Patents

Abstract

PURPOSE:To reduce collation processing at an enormous number of collation reference points and to speed up pattern matching process by calculating area of an image pattern included in the window frame of a standard pattern. CONSTITUTION:The case where a black cruciform pattern 72 represented by '0' exists in a white background area 71 represented by '1' and the case where a white cruciform pattern 74 exists in a black background area 73 are shown in the figure. In either case, the area of the image pattern included in the window frame of a standard pattern at each collation reference point is calculated before collation process. When the value of the area is smaller than a smaller reference value out of two predetermined reference values or larger than a larger reference value, the collation process is not made. In this way, the area in which collation reference points that require collation process can be narrowed from the whole screen to a small area 75 near the outline of the cruciform pattern, and thus, the pattern matching process can be speeded up.

Description

[Detailed Description of the Invention] □ Industrial Application Field This invention can be obtained from, for example, an imaging device that is necessary when mounting IC components on a circuit board.
6 □ / < p -7 4 ni, k, - < p -7゜ 1 □ Used in two-dimensional position detection method
The present invention relates to a pattern matching method.
□Conventional technology “IF, O-' i-y?-/f y fK
L--tB-.

Broadly divided,
11 A method in which one frame of an image pattern obtained from an imaging device is stored in an image pattern storage device such as a semiconductor memory, and then compared with a standard pattern.

2. Matching with the standard pattern using a device that generates the coordinates of the position being scanned by the imaging device and an image pattern storage device such as a semiconductor memory that has the minimum capacity necessary for matching with the standard pattern. A method that performs processing while importing image patterns.

It can be classified into two types.

Therefore, regarding these two pattern matching methods,
Let's consider three items: processing speed, processing scalability, and cost.

First, regarding the processing speed, the first method performs the matching process after capturing one frame of the image pattern into the storage device, which generally requires processing time and is not suitable for real-time processing. The time required to import one frame of the image pattern into the storage device is Ti, and the amount of time taken into the storage device is 1.
Let Tp be the time required to match the image pattern for one frame with the standard pattern, then the time T required for pattern matching by the first method is calculated by the sum of Ti and Tp. =7・+T・・・・・・・・・・・・・・・
・・・・・・・・・・・・[1) ml 1
It is expressed as p. The second method performs the matching process in the time required to capture the image pattern while all the image patterns are lost, so the processing time is short and suitable for real-time processing. The time "m2" required for pattern matching by the second method is equal to Ti (expressed as follows).

Tm2=Ti・・・・・・・・・・・・・・・
(2) Next, we will discuss the extensibility of processing. In the first method, the image pattern obtained from the imaging device is stored in the storage device, so the operator can freely set the weighting in processing such as filtering, which is preprocessing for matching with the standard pattern. Alternatively, after the matching process is completed, it is possible to selectively extract a specific area of the image pattern again and perform the confirmation process based on the result. Also, when performing verification processing,
It is also possible to scan the alignment point between the image pattern and the standard pattern (hereinafter referred to as a matching reference point), which serves as a reference for matching, in a direction perpendicular to the scanning direction of the imaging device. Second
Although it is possible to perform pre-processing such as filtering and confirmation processing after matching processing in the above method, the pre-processing is fixed and operator weights during filtering cannot be easily changed, and the Confirmation processing cannot be performed on the same screen where the verification results are obtained.

Finally, we will discuss the cost of realizing a pattern matching device using these two methods.

The first method requires more storage elements such as semiconductor IC memory than the second method in order to store the image pattern for one frame. This method requires more logic ICs than the first method because it compares the data with a standard pattern. In the past, semiconductor IC memories were expensive and it was difficult to realize the first method at a low cost, but in recent years the degree of integration of semiconductor IC memories has increased rapidly, and the price per unit storage capacity has decreased. The price of logic ICs for realizing the second method is also becoming cheaper. therefore
Currently, it is possible to realize a putting device at a lower cost by adopting the first method.
1 Among the above two pattern matching methods,
1 The first method is widely used in the field of remote sensing, and the second method is widely used in the field of IC component assembly.

It's coming. In addition, in the first method, in order to increase the speed,
Matching reference points Kl<Glr are set every few pixels on the image pattern.
>I, -=? tfJ'lf, f'kffl, -=, &
ni-ttLI; j.

From the matching reference points with a high degree of matching obtained from the matching results,
Recognition through detailed matching using search methods
If you search for one point, you'll get ``1 point/: i 7?-)
f 7 f method 7 “ 1 Often adopted 0 11゜Problem that the invention seeks to solve
1. The present invention relates to a pattern matching method for performing two-dimensional position detection tV real time using the first method described above. The first method is unsuitable for high-speed processing.
The reason why this method was intentionally adopted is because, as mentioned above, the pattern matching process has a high degree of freedom and can be performed flexibly, and it can be realized at a lower cost than the second method. Further, the present invention is based on the premise of incorporating a high-speed processing method called the hierarchical pattern matching method using the above-mentioned coarse matching and detailed matching.

However, with the above configuration, it is difficult to detect the position of the IC component within the processing time required when mounting the IC component on a circuit reference, for example. The reason for this is that even if matching reference points are set roughly every few pixels on the image pattern, matching processing at these matching reference points alone requires several times the processing time Ti including the number of images. be.

In view of the above-mentioned problems, the present invention has been devised so that, prior to matching an image pattern with a standard pattern at matching reference points roughly provided on the image pattern, each pixel of the image pattern included within the window frame of the standard pattern is Calculate the sum of brightness,
The present invention provides a pattern matching method that speeds up processing by determining whether or not to perform matching processing between an image pattern and a standard pattern based on the calculated value.

Means for Solving the Problems In order to solve all of the above problems, the present invention digitizes the video signal obtained from the imaging means and stores the image pattern obtained as numerical information of the brightness corresponding to each pixel. In a pattern matching process, the image pattern is stored in a storage device and compared pixel by pixel with a standard pattern stored in the storage device using the same method as the image pattern, and the position where the degree of matching is maximized is determined. Before performing pixel-by-pixel matching between the pattern and the standard pattern, calculate the sum of brightness for each pixel of the image pattern included within the window frame of the standard pattern, and calculate the sum of brightness for each pixel of the image pattern included in the window frame of the standard pattern, and calculate the sum of brightness for each pixel of the image pattern included in the window frame of the standard pattern. If the smaller one of the two reference values is smaller, or if the larger one is also larger, the above image pattern and the standard pattern are not compared pixel by pixel, and the above luminance is The image pattern and the standard pattern are compared pixel by pixel only when the total sum is greater than or equal to the lower reference value and less than or equal to the larger reference value.

Function First, for the sake of simplicity, the brightness values of the image pattern and the standard pattern are assumed to be binary values of 0'' or 1'', and since the standard pattern is a characteristic part extracted from the image pattern, the standard pattern is It is assumed that "0" and "1" are mixed in a pattern at a certain rate or more. It doesn't mean you lose your sexuality.

In addition, the calculation of the sum of brightness for each pixel of the image pattern included within the window frame of the standard pattern, which is a feature of the present invention, is carried out by
If the image pattern is binary as in the above assumption, it is nothing but counting the number of "1"s included in the image pattern. In other words, this is the image pattern included within the window frame of the standard pattern. This is equivalent to calculating the area of the “1” region of .

Therefore, from now on, the calculation of the sum of brightness for each pixel of the image pattern included within the window frame of the standard pattern will be referred to as area calculation.
6.

Based on the above assumptions, the operation of this technical means will be explained with reference to the drawings.

Figure 2 shows a standard pattern 22 with a size of MxM and an image pattern 1 with a size of NxN.
23 shows matching reference points 24 provided every n pixels on the image pattern 22. The total number of matching reference points 24 existing on the image pattern 23 is given by the following equation (3) in consideration of the size of the standard pattern 2°2. Symbol Ca]
represents the largest integer not exceeding a.

N-Ad+1 ((knee)+1)2 ・・・・・・・・・・・・・・・
(3) The total number of reference points 24 is, for example,
N=512°M=16. If n=4, about 15,60
It becomes a huge value of 0. As shown in FIG. 2, these reference points include a matching reference point 25 located at a position such that the window frame of the standard pattern is completely included in the area "1" of the image pattern, and a reference point 25 of the image pattern. There are a large number of matching reference points 26 located at positions that are included in the "0" region of . The matching process between the image pattern 23 and the standard pattern 22 at the matching reference point 25 is a process of calculating the area of the "1" area included in the standard pattern 22, and the matching process at the matching reference point 26 is a process for calculating the area of "1" included in the standard pattern 22. "0"
This is the area calculation process for the area. The matching process is usually performed by two processes: reading out the 0 image pattern from the storage means, and calculating the degree of matching by performing logical operations on each pixel of the read image pattern and the standard pattern.
Depending on the size of the standard pattern, these steps ① and ② are divided into several steps, so in this case, ③ cumulatively adding the matching degrees obtained as a result is added. be done. However, in the matching process at the matching reference points 24 and 25, as in the above case, the logical operation (2) is completely meaningless and wasteful processing. At matching reference points where matching processing has meaning,
The proportion of 10" and "1" mixed within the window frame of the standard pattern is the initially assumed "0" of the standard pattern itself.
Therefore, in the present invention, at the matching reference points roughly set on the image pattern, the area obtained by the processing of Calculation is performed first, and if the area obtained as a result is smaller than the smaller of the two pre-given reference values or larger than the larger of the two, the matching process is performed. The area is moved to the next reference point without performing this, and only when the area is greater than or equal to the smaller of the two pre-given reference values and less than or equal to the larger of the two pre-given reference values, the area is We decided to perform the matching process at the matching reference point. Also,
Area calculation is not closely related to the position detection of two-dimensional patterns, which is the ultimate goal of the present invention, so
Even if it is done strictly, the effect will not be fully reflected. Therefore, in the present invention, as shown in FIG. 3, we decided to perform all area calculations only for the partial region 32 of the standard pattern 31. but,
Depending on the degree of thinning of the window frame when calculating the area of the standard pattern 32 and the spacing between matching reference points provided on the image pattern, information regarding the area of the image pattern may not be obtained correctly. Therefore, when setting these values, the safety factor must be taken into consideration. In addition, the proportion of mixed "o" and "1" expressed by the two reference values used for judgment when calculating the area is sufficiently lower than the proportion of mixed "0" and "1" in the initially assumed standard pattern. This is set in advance so as to be able to deal with errors in area caused by providing matching reference points every few pixels and by thinning out the area within the window frame of the standard pattern.

As described above, the time required for matching processing at a huge number of matching reference points provided every few pixels on an image pattern can be significantly reduced.

EXAMPLE Hereinafter, a pattern matching method according to an example of the present invention will be described with reference to the drawings.

FIG. 1 shows a block diagram of a pattern matching method according to an embodiment of the present invention. In FIG. 1, a video signal 2 obtained from an imaging means 1 such as a television camera is binarized by a binarization means 3 to become an image pattern 4 and stored in an image pattern storage means 5. The image pattern storage means 5 is controlled by an image pattern control means 6, and the input/output of the image pattern is performed by commands 8 from the central processing means 7. The image pattern control means 6 includes image pattern X-coordinate and y-coordinate storage means for reading out a specific image pattern from the image pattern storage means. On the other hand, the standard pattern 9 is cut out in the image pattern storage means 5 at the time of teaching prior to the pattern matching process, and is held in the standard pattern storage means 10. The standard pattern storage means 10 is controlled by the standard pattern control means 11.
The input/output of the standard pattern is performed by command 1 from the central processing means 7.
2. When the central processing means 7 performs the matching process between the image pattern and the standard pattern, the data for performing the logical operation is read out from the image pattern storage means 5 and the standard pattern storage means 1, respectively, and the data is read out from the logical operation means 13. It has the role of transmitting the command 8 to the image pattern control means 6 and the command 12 to the standard pattern control means 1o so as to be input to the image pattern control means 6, and the role of controlling each block in FIG. The two pattern data inputted to the logic operation means 13 are subjected to exclusive OR negation, and the two pattern data are
When the corresponding pixels in two pattern data match, the pit becomes 11". When the standard pattern 41 consists of A x B pixels as shown in FIG. The obtained A-bit data 14 is input to the pit counting means 15, where the number 1j, which is the number of matching pixels in the A-pit, is counted, and the resultant data 1'& degree 1 in the A-bit is counted.
6 is input by the cumulative addition means 17.

For the standard pattern 41 consisting of A×B pixels, the above processing is repeated for B cycles to complete the processing for the entire standard pattern. Image pattern L
The matching degree of 18, which is the processing result of : and the standard pattern, is medium.
1゜)□ If the degree of matching is high as determined by the central processing means 7, the
II) The coordinate values of the verification reference point at the time are stored in the verification result storage means 19.
and stored in [′. To calculate the area of the image pattern, the image 1° pattern is read out from the image pattern storage means 5.
1□I, I, □Yo, 3□fK□1,1ka,
This is done by inputting the A-bit information 2o to the input means 15.
Pattern matching method configured as above 1′i′ ′″′°“″′″tT@111″sQ＞!#IE°12
．． .

The operation will be explained using In Figure 6, the size is AxB
□Bare A71. Kajun tap-ys4g-t. picture. base
It represents a 1° single point 52 and a matching reference 1° point 64 provided on an image pattern 53 (however, the pattern is not shown). Assuming that the starting point of the matching process is the matching reference point 56 at the upper left corner of the image pattern 53, first align the matching reference point 5°2 of the standard pattern 61 with the matching reference point 65 of the image pattern 63, and then The area of the image pattern 63 contained within the window frame is calculated. At this time, area calculation is performed on a part of the window frame 61 of the standard pattern with the size of A×B pixels shown in the bottom cylinder 6.
In this figure, the image pattern is divided into four times of A bit each and input to the pit counting means 15 in FIG. 1, and the area is calculated by the cumulative addition means 17.

This area is then compared with two predetermined reference values in the central processing means 7, and if it is smaller than the smaller reference value or larger than the larger reference value, the area is compared with two predetermined reference values. The process proceeds to the next comparison reference point 56 and area calculation is performed again. If the calculated area is greater than or equal to the smaller of the two reference values and less than or equal to the larger of the two reference values, the image pattern 53 and the standard pattern 61 are compared at the matching reference point 55, and the result is obtained. The matching degree obtained is judged by the central processing means 7 in FIG. 1, and if a high matching degree is obtained, the matching result storage means 1
The coordinate values of the matching reference point 9 are stored in the matching reference point 65, and the first calculation at the matching reference point 65 in FIG. The method of moving the matching reference point on the image pattern 63 is that when scanning the screen vertically, it is first moved downward from the reference point 65 to the reference point 66, and then moved to the lowest reference point 57. Next highest reference point 6
Moving on to step 8, perform the above area calculation until the end point, reference point 69.
Verification processing is performed.

′″″″″″′1”;g ”)7'rLRV4-'41
°'nO”Jjhr('1'' i:,.

Among the matching reference points 54 existing in the turn 53, the matching reference that is thought to exist near the recognition point, aomii
”゛・M * (MoN****Ltl[19cl=+
, , will be stored in . Therefore, next, by performing detailed matching using a search method at each coordinate stored in the matching result storage means 19,
It is possible to detect a position near the matching reference point where the degree of matching is maximum. The detected position data 21 is then transmitted to a twin controller or the like that uses this position detection device.

As described above, according to this embodiment, by calculating the area of the image 1a pattern included in the window and frame of the standard pattern, the matching process at a huge number of matching reference points is reduced, and the pattern matching process is can be accelerated.

Effects of the Invention FIG. 7 is a diagram for explaining the present invention in detail. A case where a black cross pattern 72 represented by “0” exists in a white background area 71 represented by “1”;
A case is shown in which a white cross pattern 74 exists within a black background area 73. In either case, before the matching process, the area of the image pattern included within the window frame of the standard pattern at each matching reference point is calculated, and the value of this area is the smaller of two predetermined reference values. If the reference value is smaller than the reference value or larger than the larger reference value, the matching process will not be performed, thereby reducing the area where the matching reference point that needs matching processing exists to the entire screen. This can be narrowed down to a small area 75 near the outline of the cross pattern, and the pattern matching process can be sped up.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of a pattern matching method that is an embodiment of the present invention, FIG. 2 is a diagram showing the sizes of a standard pattern and an image pattern, and matching reference points existing on the image pattern. Figure 3 shows the area within the window frame of the standard pattern that is subject to area calculation, Figure 4 shows the size of the standard pattern, and Figure 5 shows the movement of the standard pattern relative to the matching reference point on the image pattern. FIG. 6 is a diagram showing the method, FIG. 6 is a diagram showing the area to be calculated within the window frame of the standard pattern, and FIG. 7 is a diagram showing the effect of the present invention. 5... Image pattern storage means, 6...
Image pattern control means, 7...Central processing means,
10...Standard pattern storage means, 11...
...Standard pattern control means, 13...Logic operation means, 15.Old...Bit counting means, 17...
... Cumulative addition means, 19 ... Verification result storage means, 22 ... Standard pattern, 23 ... Image pattern, 24 ... Verification reference point, 31 ...
...Picture 1 elephant pattern window frame, 32... Area calculation target area, ``S%1...Standard iJ-n, 51.
...Standard pattern, 53 ... Image pattern, 54 ... Verification reference point, 61 ... Standard pattern window frame, 62 ... Area calculation Target area 0 Name of agent Patent attorney Toshio Nakao and 1 other person 1st
Figure 2 Figure 3 Figure 4 Legal dimensions h To Nc'-

Claims (2)

[Claims] (1) The video signal obtained from the imaging means is digitized, and an image pattern obtained by numerical information of the brightness corresponding to each pixel is stored in the storage means, and the image pattern is stored in advance in the same way as the image pattern. In pattern matching processing, which compares each pixel with the stored standard pattern to find the position where the degree of matching is maximum, before performing pixel-by-pixel matching between the image pattern and the standard pattern, Calculate the sum of brightness for each pixel of the image pattern included within the window frame, and determine whether the sum of the brightness is smaller than the smaller of two predetermined reference values, or If it is larger than the larger reference value, the image pattern and the standard pattern are not compared pixel by pixel, and the sum of the luminances is greater than or equal to the smaller reference value and less than the larger reference value. A pattern matching method characterized by performing pixel-by-pixel matching between the image pattern and the standard pattern only if . (2) When calculating the sum of brightness for each pixel of the image pattern included within the window frame of the standard pattern, use means for comparing each pixel of the image pattern and the standard pattern and cumulatively adding the degree of matching. The pattern matching method according to claim 1, which is used as is.