JPS5831027B2 - pattern matching device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a pattern matching device that matches patterns with variations in pattern quality or local differences with high accuracy.

A pattern matching method using digital processing is often used to compare and match the patterns of seal impressions, characters, printed matter, etc., with the patterns on the original register.

Figure 1 a and b are the binarized patterns of the seal frame,
On the two-dimensional xy coordinates, the point with a signal is 1, the point without a signal is O, and the black circles are 10 points.

If these two patterns in figures a and b are superimposed and compared, the result will be as shown in figure 0.

Here, the white circles are points where the patterns overlap, and the 1Δ"1×" marks are points that are on one side but not on the other.

The degree of matching F between these two patterns is as follows.

However, So is the area of the overlapping part of the two patterns (0
(white circles in the figure), S4, and S2 are the areas of the two patterns (black circles in figures a and b).By the way, in this kind of verification, the pattern of the original record is generally filed, and the input pattern matches the original record. This is done in order to determine whether or not the pattern is true, but the input pattern is generally accompanied by various noises.

In particular, patterns using ink, vermilion, etc. bleed and become blurred when printed, which results in a lower degree of matching during matching.

Even when checking seals, parts of the seal often become blurred or smeared due to the pressure applied when stamping or the way the ink is applied.

Also, it is necessary to judge that patterns with slight local differences are different, but since the above formula considers the degree of matching as the overall average, even if there is a significant difference in some patterns, it will be judged as different. If the parts match well, it may be determined that they are the same.

The present invention was made in order to eliminate these drawbacks, and provides a pattern matching device that increases the matching degree by taking into account differences in pattern quality and local differences.

In this invention, the plane of the entire pattern is divided into small areas of, for example, 16 x 16 or 32 x 32, and the two patterns in each area are compared and judged. I will explain about it.

In Figure 2, the plane of the entire pattern is divided into M small areas in the
m), D (n-1, m), D (n+1, m)...
・It is divided into ・

Matching degree of two patterns in each region FD (n, rTl)
is calculated independently using the following formula.

Areas where the matching degree is low contribute to lowering the overall matching degree, and we will investigate these areas in detail.

The reasons for the markedly low matching degree are (A): The pattern difference is really significant in that area, and (B) The pattern quality of one or both of them is extremely poor due to blurring. Differences can be seen here (N should make the contribution that lowers the matching degree to the overall matching degree even stronger, and by doing this, there are many similarities overall, but there are differences in some parts. Can detect patterns.

Regarding (B) above, the contribution that lowers the degree of matching should be reduced, so that even if there is slight variation in the quality of the patterns, it can be determined that they match.

For example, when the matching degree of D(m, n) is extremely poor, the following method is used to determine whether the reason is either the above-mentioned error or CB).

Although FIG. 1 shows a binarized pattern, here a pattern with shading information is used.

Generally, the input data is an analog value corresponding to the density level, but this can be digitized into, for example, 4 bits (
Used as data for 16 stages).

It is assumed that the value O is a level with no signal at all, and the higher the value, the higher the concentration level.

To perform binarization, points where the density level is above a certain value are set to 1, and points below that are set to 0.

Using this data, the average value of the level of the binarized points with a signal is determined as "sum of levels of points with a signal/number of points with a signal".

When this value is determined for the entire pattern, the level average value: Lt for the entire pattern is determined.

However, S is the area of a point binarized using TH as a threshold.

L(x,y) is the density level of each point This value varies to some extent depending on the degree of vermilion coating.

Next, when the matching degree of D(m, n) is extremely poor, the average level in this area is determined.

Let the overall average level of the first and second patterns to be compared be Lt,,Lt2, and the average level at D(m,,n) be L+-D(m,n)L2-D(m,n). When L1゛ L1゛D (m゛n) are compared and they are almost equal, it is determined that there is no bleeding or blurring between - and Lt2 L2- D (m, n), which is the reason for the above (N).

Also, if there is a significant difference in this, it is determined to be due to the reason (B) above.

In some cases, if one of the patterns is blurred and its area is extremely small, the value of D(m, n) in the peripheral area is used.

Let the average level in the surrounding area be Ll, D', L2.0, and L
tL' then compare - and there is a difference Lt2
L2・D Determine whether the above reason is the prisoner's reason or the reason (B).

If the overall matching degree is F, the matching degree of the part determined as the area (A) or (B) above is F(11 or F(2), and the matching degree of other areas is FNORMAL, then α is set to 1. Let β be less than 1 for the dog.

An embodiment of the pattern matching device of the present invention having such a function is shown in FIG.

In Fig. 3, 1.2 is a pattern memory that stores the binarized pattern to be matched, and 4 and 5 are the binarized shading of the first and second patterns having the shading levels to be matched. Shade pattern memory that stores pattern signals,
3 is a verification calculation circuit, 6 is a data memory, and 7 is a determination circuit.

In the pattern matching device having such a configuration, the matching calculation circuit 3 performs each region D(n, m), D(n-1, m), . . . of the first and second two patterns to be matched. The pattern signals of are sequentially read out from the pattern memories 1 and 2 and compared and calculated, and the area St, D(n
, m)・sl, D(n-1, m)"""Area S2, D(n, m), S2, D(n-
1, m)...and the area of the overlapping portion of the first and second patterns in each region of the first and second patterns 5OzD (n, m), 504) (n-1 , m)...
. . , and calculates the matching degree F in each region by calculating the equation (2) above.Furthermore, from the gray pattern memories 4 and 5, each region D(n
, m), D(n-, , m)-, the first pattern's density level average value L]% D(n, m) N L
IN D(n-1, m)"""-average level of shading of the second pattern L2, D(n, m), L7, D(n-,
, m) . . . and the overall gray level average value Lt of the first pattern and the overall gray level average value Lt2 of the second pattern are determined and input into the data memory 6.

The determination circuit 7 reads out each data stored in the data memory, performs the arithmetic comparison using the above equation (2), detects an area with a low degree of matching, and if this area is now D(n, m), then Ll, D(n, m)/L2, D(n, m)
is calculated, compared with Lt and /Lt2, and the above equation (4) is calculated to determine the overall matching degree.

It is determined whether the matching patterns are the same based on the overall matching degree.

Although the above description has been made of the case where the average value of the shading level in the entire area and each area of the first pattern, which is the pattern of the original register, is different, in general, the average value of the above-mentioned shading level of the original pattern is approximately constant, In this case, this average value may be stored in a grayscale pattern memory, and the determination circuit 7 may read it.

As described above, the verification device according to the present invention determines the state of each small area and determines the overall verification degree, so blurring occurs.
It is also possible to match patterns with high accuracy even if the quality varies due to blurring or there are local differences in the pattern.

[Brief explanation of drawings]

Figure 1 is a diagram showing the binarized pattern of the frame of the seal;
FIG. 2 is a diagram showing the entire pattern divided into small areas, and FIG. 3 is a block diagram showing one embodiment of the present invention. In the figure, 1 and 2 are pattern memories, 3 is a matching calculation circuit, 4 and 5 are gray pattern memories, 6 is a data memory,
7 is a determination circuit.

Claims (1)

[Scope of Claims] 1. First and second pattern memories that store digitized pattern signals of each area of a first and second two patterns divided into a plurality of areas, and each of the above-mentioned patterns. The first one stores the digitized gray signals of the area and the entire area.
and a second gradation memory, the first and second pattern memories, and the gradation memory to read the stored data and calculate the total pattern area S1 in each region of the two patterns.
, S2 and the area S of the overlapping portion of the two patterns. , and use these data to find the degree of matching in each area, and also to calculate the average value Lt, , Lt2 of the overall shading level of at least one of the above two patterns, and the average value 7, L2 of the shading level in each area. The average value of the gray level in the same area of the two patterns in the area where the matching degree is lower than the data obtained by the matching calculation circuit and the ratio of L2 and the overall gray level L
A pattern matching device comprising a determination circuit that calculates an overall matching degree from the data obtained by comparing the ratios of t1 and Lt2. 2 The grayscale signals stored in the first and second grayscale memories are based on the average value of the digitized density level of the entire area of the first pattern with uniform density level and each area, and the average value of the digitalized density level of each area of the second pattern. 2. The pattern matching device according to claim 1, wherein the pattern matching device is a digitized gray pattern signal of the entire area.