JPH06337939A - Method and device for identifying picture - Google Patents

Abstract

PURPOSE:To provide a new picture identifying method capable of effectively shortening processing time. CONSTITUTION:Respective directions of reference axes X, Y mutually meeting at right angles are integrated by the information value of a reference pattern and respective integrated results are differentiated by the directions rectangular to respective integrating directions to prepare X direction differential reference data and Y direction differential reference data. In an identification searching area picture, the reference axes X and Y directions are integrated by an information value and respective integrated results are differentiated by the directions rectangular to respective integrating directions to calculate X direction differential data and Y direction differential data. The X and Y direction differential reference data and the X and Y direction differential data are respectively processed by pattern matching operation.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image identifying method and an image identifying apparatus.

[0002]

2. Description of the Related Art "Pattern matching" is known as one of methods for identifying a two-dimensional image. In pattern matching, a part of the “reference image” that shows the characteristics of the reference image is extracted as a search region, and an image in this search region, that is, a part of the “search region image” is Set as "reference pattern".

When an image to be identified is captured by reading, an image of a portion of the captured image corresponding to the search area is extracted as an "identification search area image", and a reference pattern is applied to the identification search area image. Is two-dimensionally moved, and the correlation between the “identification search area image portion overlapping the reference pattern” and the reference pattern is calculated at each movement position of the reference pattern.

Although various equations can be considered for calculating the correlation,
Basically, the product of the “information value” of each pixel in the reference pattern and the information value of the pixel in the identification search area that overlaps this pixel is added for all pixels in the reference pattern, and added as necessary. And multiply it by a constant. When this value is called the "correlation value", "the same pattern as the reference pattern" is displayed in the search area.
And when the reference pattern overlaps this pattern,
The correlation value becomes the “autocorrelation value” of the reference pattern and takes the maximum value.

Therefore, when a "threshold level" close to the autocorrelation value and smaller than the autocorrelation value is set, and the maximum value of the correlation value in the identification search area is larger than the threshold level, the captured image captured is referred to. It contains the same pattern as the pattern and can therefore be identified as being part of the same image as the reference image.

Image discrimination by such pattern matching has good discriminability and high reliability of discrimination, but since the information of all the pixels of the reference pattern is used for calculating the correlation value, the amount of calculation required for discrimination becomes large. There was a problem that the processing time was long.

In order to reduce the amount of calculation, a method such as reducing the size of the identification search area or reducing the size of both the identification search area and the reference pattern can be considered. With the reduction of these sizes, the identification accuracy is reduced. Is also not preferable because it also decreases.

[0008]

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and provides a novel image identifying method capable of effectively reducing the processing time and an apparatus for implementing this method. Is.

[0009]

According to a first aspect of the present invention, there is provided an image identification method, comprising: "an identification search area image extracted from a two-dimensional captured image captured by image reading; and a two-dimensional reference pattern prepared in advance. Is a method for identifying whether or not a reference pattern is included in the captured image by pattern matching with the above. ”And is characterized by the following points.

That is, for a reference pattern prepared in advance, the information value of the reference pattern is integrated in each direction of the reference axes X and Y which are orthogonal to each other, and each integration result is differentiated in the direction orthogonal to the integration direction. Then, X direction differential reference data and Y direction differential reference data are prepared. Further, the identification search area image is subjected to the integration operation of the information values in the reference axes X and Y directions, and each integration result is operated in the direction orthogonal to the integration direction to perform the differentiation operation to differentiate the X direction differential data and the Y direction differential. Calculate the data. Pattern matching operations are performed on the X-direction differential reference data and the X-direction differential data, and the Y-direction differential reference data and the Y-direction differential data, respectively.

An image identifying apparatus according to a second aspect is an apparatus for implementing the image identifying method according to the first aspect, and has an image capturing means, a reference pattern extracting means, and a computing means.

The "image capturing means" reads the image to be identified and extracts the image information of the characteristic portion area to be searched. The area thus extracted is the "identification search area", but is the search area when the captured image is the "reference image". The extracted image information is referred to as an “identification search area image” for the identification search area and a “search area image” for the search area. The “reference pattern extraction means” extracts a desired portion in the characteristic portion (search area image) as a reference pattern when the standard image is captured. "Calculation means"
When the image to be identified is fetched, the information values of the identification search area image of the area extracted as the identification search area and the image information of the reference pattern are calculated, and integration is performed in each of the X and Y axis directions orthogonal to each other. Operation, and each integration result is differentiated in the direction orthogonal to the integration direction, and X
Directional differential data, Y-direction differential data, X-direction differential reference data, and Y-direction differential reference data are calculated, and pattern matching is performed for the X-direction differential reference data and X-direction differential data, and the Y-direction differential reference data and Y-direction differential data, respectively. It is a means of operating. The pattern matching operation is performed by examining the correlation.

Since both the captured image from which the image to be identified and the reference pattern are image information decomposed for each pixel, the information values separated for each pixel are spatially discontinuous. In the above description, the operation of integrating the image information in the X and Y directions means an operation of adding the information values of the pixel information arranged in these directions in each of the above directions. In addition, differentiating each integration result means taking the difference (or a constant multiple) of the integrated values of adjacent pixel rows in the direction orthogonal to the integration direction, or, as described later, Constant multiple "
Means to take the absolute value of.

Both the reference pattern and the identification search area image are digitized, and the integration operation and the differentiation operation are performed digitally.

[0015]

In FIG. 1, (a) shows a reference pattern prepared in advance. X and Y orthogonal to each other as shown
Set the direction. For simplification of description, it is assumed that the reference pattern is binary information including “white” and “black” as shown in the figure, and the information value of each pixel is “white” and “black”.
Is set to 0.

The information value distribution of the reference pattern is f (X,
Y) (actually, a set of information values f (X _i , Y _j ) for each pixel), and integrating this in the X direction,
Integration result: ∫f (X, Y) dX (= Σf (X _i , Y _j );
The sum is a subscript: i) is a function of Y: F (Y) (actually, it is a set of the above sums regarding the order of the pixel row in the X direction in the Y direction: Y _j ). It will be as shown in b). Horizontal axis: F (Y) represents “luminance level (sum of information values of“ white ”)”. This operation is an integration operation in the X direction.

Similarly, when the above f (X, Y) is integrated in the Y direction, the integration result: ∫f (X, Y) dY (= Σf
(X _i , Y _j ); sum takes index: j) is a function of X:
G (X) is as shown in FIG. 1 (c). Vertical axis: G
(X) represents a "brightness level". This operation is an integration operation in the Y direction.

Next, when the above F (Y) is differentiated with respect to the Y direction orthogonal to the X direction which is the integration direction for obtaining F (Y), the differential result: d {F (Y)} / The absolute value of dY (actually, the difference: F (Y _{j + 1} ) −F (Y _j ) or this difference multiplied by a constant) = fy (Y) is shown in FIG.
As shown in. fy (Y) is “Y-direction differential reference data”.

Similarly, when G (X) is differentiated in the X direction, the differential result: d {G (X)} / dY (actually,
The absolute value of the difference: G (X _{i + 1} ) −G (X _i ) or the difference multiplied by a constant = fx (X) is as shown in FIG. fx (X) is “X-direction differential reference data”.

In FIG. 2, (a) shows an identification search area image extracted from the captured image captured as the identification target. Corresponding to the reference pattern (FIG. 1A), the X and Y directions orthogonal to each other are set. Similar to the reference pattern, the identification search area image is also binary information including “white” and “black”, and the information value of each pixel is 1 for “white” and 0 for “black”.

The information value distribution of the identification search area image is defined as P
(X, Y) (Actually, the information value for each pixel: P (X _m ,
Y _n )), and performing an integration operation in the X direction, the integration result: ∫P (X, Y) dX (= ΣP
(X _m , Y _n ): Sum is a subscript: m) is a function of Y:
Q (Y) (actually, the order of the pixel rows in the X direction in the Y direction: a set of the above sums with respect to Y _n ).
As shown in (b). Horizontal axis: Q (Y) represents "luminance level".

Similarly, when the above P (X, Y) is integrated in the Y direction, the integration result: ∫P (X, Y) dY (=
ΣP (X _m , Y _n ): Sum is a subscript: n) is a function of X: R (X), which is as shown in FIG. Vertical axis: R (X) represents "luminance level".

Next, when Q (Y) is differentiated in the Y direction, the differential result: d {Q (Y)} / dY (actually,
The absolute value of the difference: Q (Y _{n + 1} ) −F (Y _n ) or this difference multiplied by a constant = Py (Y) is as shown in FIG. 2 (d). Py (Y) is “Y direction differential data”.

Similarly, when R (X) is differentiated in the X direction, the differential result: d {R (X)} / dY (actually,
The absolute value of the difference: G (X _{m + 1} ) −G (X _m ) or this difference multiplied by a constant = Px (X) is as shown in FIG. Px (X) is “X direction differential data”.

Thus, four one-dimensional data, that is, X
Directional differential reference data: fx (X), Y-directional differential reference data: fy (Y), X-directional differential data: Px (X), Y-directional differential data: Py (Y).

When the size of the reference pattern shown in FIG. 1A is (I × J) pixels and the size of the identification search area image shown in FIG. 2A is (M × N) pixels, fx (X) is I-
One element: fx (X _i ) and fy (Y) is J-
It consists of one element: fy (Y _j ). In addition, Px (X)
Consists of M-1 elements: Px (X _m ), Py (Y)
Consists of N-1 elements: Py (Y _n ).

Therefore, while fx (X) is superimposed on Px (X) and shifted by one pixel at a time, each time the shift is performed, the sum (or a constant multiple) of the products of the information values of the overlapping pixels is calculated as a correlation value. Perform a matching operation. This is an operation: K · Σ _i , where K is a constant value and k is an operation parameter.
It means that fx (X _i ) · Px (X _{i + k} ) (i = 1 to I-1) is repeated in the range of k = 1 to (MI) for the calculation parameter: k. The maximum value of the calculated correlation values (the X coordinate giving the maximum value (the position corresponding to the center coordinate of fx (X) is X ₀ ) is compared with a preset threshold value.

Similarly, while fy (Y) is superimposed on Py (Y) and shifted by one pixel at a time, each time the shift is performed, the sum of the products of the information values of the overlapping pixels (or a constant multiple thereof) is calculated as a correlation value. . This is L for a constant value and l for a calculation parameter.
As: L · Σ _{j, n} fy (Y _j ) · Py (Y _{j + l} )
(J = 1 to J-1) for the calculation parameter: l,
It means repeating in the range of l = 1 to (N−I).
Maximum value of the calculated correlation values (Y coordinate (f
The position (corresponding to the central coordinate of y (Y)) is set to Y ₀ ) is compared with a preset threshold value.

When both of the above two maximum values are equal to or more than the threshold value, the identification search area image extracted from the captured image has the same pattern as the reference pattern therein, and its position is (X ₀ , Y ₀ ). And in this case, the captured image is identified as being identical to the reference image.

Note that d {F (Y)} / dY = d {∫f
(X, Y) dX} / dy = ∫ [d {f (X, Y)} / d
Note that it is possible to switch the order of the integration operation and the differentiation operation because y] dX. Further, in the example described above, the absolute value of the calculation result is taken in the differential operation, but the result of the differential calculation (including positive / negative data) may be used as it is.

[0031]

FIG. 3 shows an embodiment of the image identifying apparatus according to the present invention. This device is, for example, a device for identifying a wiring pattern or the like of an electronic component such as an IC,
0, the A / D converter 12, the preprocessing unit 14, the feature extraction unit 16, the reference pattern extraction unit 20, the identification unit 18, and the control unit 22.

First, a subject 0 having a "reference image" is read by the CCD camera 10 and converted into an electric signal for each pixel. This electric signal is digitized by the A / D converter 12 and sent to the preprocessing unit 14.
The pre-processing unit 14 is a comparator, which is controlled by the control unit 22 (specifically, a computer or the like), and removes noise by comparing the set threshold value with the input signal.

The data from which the noise is removed is the feature extraction unit 1
Sent to 6. The feature extraction unit has a memory, and the input data is developed in the memory as two-dimensional information. The control unit 22 sets a predetermined search area in the two-dimensional information developed in the memory of the feature extraction unit 16 and extracts the information inside the search area as a search area image. Therefore, the CCD camera 1
0, the A / D converter 12, the pre-processing portion 14, the feature extracting portion 16 and the control portion 22 constitute "image capturing means".

The control means 22 also designates and extracts a desired portion in the extracted search area image as a reference pattern and stores it in the reference pattern extraction unit 20. Therefore, the control unit 22 and the reference pattern extraction unit 16 constitute "reference pattern extraction means".

Next, the subject to be identified is read, and the feature identifying section 16 extracts the "identification search region image". The extraction method is the same as in the case of extracting the “search area image”.

The control unit 22 controls the reference pattern extraction unit 20 to read the “reference pattern” stored in the extraction unit 20 and send it to the identification unit 18. At the same time, the feature extraction unit 16 is controlled, and the identification search area image extracted by the extraction unit 16 is also sent to the identification unit 18.

The identification section 18 receives the control of the control section 22,
The above integration operation / differentiation operation and pattern matching operation are performed on both the reference pattern and the identification search area image. Therefore, the identification unit 18 and the control unit 22 form a calculation unit.

Thus, whether or not the read subject contains the same pattern as the reference image is obtained as the output of the identifying section 18.

[0039]

As described above, according to the present invention, a novel image identifying method and apparatus can be provided. Since the present invention has the configuration described above, the amount of calculation required for pattern matching is effectively reduced, and high-speed image identification is possible.

[Brief description of drawings]

FIG. 1 is a diagram for explaining an integration operation and a differentiation operation for a reference pattern.

FIG. 2 is a diagram for explaining an integration operation and a differentiation operation on an identification search area image.

FIG. 3 is a block diagram for explaining an embodiment of the image identifying apparatus according to claim 2;

Front page continued (72) Inventor Hirokazu Watanabe 6100 Uehara, Ina City, Nagano Prefecture, Sankyo Seiki Co., Ltd.Ina Factory (72) Inventor Koki Nakamura 6100 Uehara, Ina City, Nagano Prefecture, Sankyo Seiki Co., Ltd. Ina factory

Claims (2)

[Claims] 1. A reference pattern in the captured image is obtained by pattern matching between an identification search area image extracted from a two-dimensional captured image captured by image reading and a two-dimensional reference pattern prepared in advance. In the method of identifying whether or not it is included, the information value of the reference pattern is integrated for each direction of the reference axes X and Y which are orthogonal to each other, and each integration result is operated for differentiation in the direction orthogonal to the integration direction. Then, X-direction differential reference data and Y-direction differential reference data are prepared, and an information value is integrated in the identification search area image and in the reference axis X and Y directions. The X-direction differential data and the Y-direction differential data are calculated by performing a differential operation in the direction orthogonal to
An image identification method characterized by performing pattern matching operation on each of the direction differential reference data and the Y direction differential data. 2. An apparatus for carrying out the image identifying method according to claim 1, wherein an image capturing means for reading an image to be identified and extracting a characteristic portion as a search area image, and inside the search area image. The reference pattern extracting means for extracting the reference pattern from the reference pattern and the extracted information value of the search area image and the reference pattern information value are respectively integrated and operated in respective directions of the X and Y axes which are orthogonal to each other. Is differentiated in the direction orthogonal to the integration direction to obtain X-direction differential data, Y
Directional differential data, X-direction differential reference data, and Y-direction differential reference data are calculated, and the X-direction differential reference data and X-direction differential data, and the Y-direction differential reference data and Y-direction differential data are subjected to pattern matching operation. An image identification device having a computing means.