JP3223385B2 - Pattern matching device for grayscale images - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a gray-scale image pattern matching apparatus for performing pattern matching on a gray-scale image obtained by imaging an object using a model image of the object.

[0002]

2. Description of the Related Art Conventionally, the position of an object is determined from a grayscale image obtained by imaging the object by using a pattern matching technique. FIG. 8 is a diagram for explaining the normalized correlation density matching method, in which 1 is a rectangular model image prepared in advance, 2 is an input density image (hereinafter referred to as “input image”). It is).
The input image is scanned one pixel at a time over the entire screen by the model image 1, and the degree of coincidence between the model image 1 and the input image 2 is obtained by a correlation operation at each scanning position _Pj, and the model image 1 having the maximum correlation value is obtained. From the scanning position _Pj , the position of the object is obtained.

When the model image 1 is at the j-th scanning position P _j (X _j , Y _j ) on the input image 2, the density data of the i-th pixel in the model image 1 is M _i , Assuming that the density data of the corresponding pixel is I _i and the number of constituent pixels of the model image 1 is n, the correlation value r ² (X _j , Y _j ) at the scanning position P _j is given by the following equation (1).

[0004]

(Equation 1)

[0005]

As apparent from equation (1), conventionally, the density data M _i of all the constituent pixels of the model image 1 are compared with the density data I _i of the corresponding pixels of the input image 2. However, since all the constituent pixels of the model image 1 are subjected to pattern matching, the amount of calculation of the correlation operation becomes enormous and the processing time is long.

Therefore, the inventor recently selects pixels near the edge position of the model image 1 to perform pattern matching of the input image 2 using the model image 1 of the object.
A method for calculating a correlation value between the model image 1 of the target object and the input image 2 only for each pixel near the edge position has been developed.

FIG. 9 and FIG. 10 are explanatory diagrams of the principle when this method is concretely implemented. FIG. 9A shows a model image 1 obtained by imaging a model of a target object (in this case, the character “A”). In FIG. 9, reference numeral 3 denotes a whitish image portion of the character A; Is a dark image portion of the background.

FIG. 10A shows the density data (0 to 255 gradations) of each pixel along a certain horizontal line 6 of the model image 1 as shown in FIG. The value obtained by differentiating the density data of each pixel is as shown in FIG. 10 (2), and the secondary differential value is as shown in FIG. 10 (3). The absolute value of the second derivative is calculated as shown in FIG.
The shape is as shown in

The absolute value of the second derivative of each pixel is compared with a predetermined positive threshold value TH, and a pixel having a value larger than the threshold value TH is selected as a target pixel for pattern matching.

FIG. 10 (5) shows that the threshold value TH is 2
The binarized data, that is, the range of target pixels for pattern matching, is shown. When the density data of the model image 1 is sampled by the binarized data, sampling data as shown in FIG. Can be

FIG. 9 (2) shows a gray-scale image 5 composed of the density data sampled in this manner. The sampled density data includes the density data of the image portion 3 of the character A, that is, the inside of the edge. The density data of the neighboring pixel row 3a and the density data of the background image portion 4, that is, the density data of the pixel row 4a near the outside of the edge are included to the same extent.

The input image 2 is scanned one pixel at a time over the entire screen by the model image 1, and the density data of the model image 1 sampled by the binary data and the density data of the input image are scanned at each scanning position. Perform a correlation operation,
The position of the object is obtained from the scanning position of the model image 1 at which the correlation value becomes maximum.

However, according to this method, there is no problem when an object having a proper posture is imaged, but the posture of the object is inclined and the input image 2 is moved in any direction as shown by a broken line a in FIG. If it is slightly rotated, correct pattern matching cannot be performed, and position measurement may be difficult.

SUMMARY OF THE INVENTION The present invention has been made in view of the above-mentioned problem. By efficiently reducing the number of pixels to be subjected to pattern matching, the amount of correlation calculation can be reduced to shorten the processing time. It is an object of the present invention to provide a gray-scale image pattern matching apparatus that can perform correct pattern matching even when the image is rotated.

[0015]

According to the present invention, there is provided an apparatus for performing pattern matching of an input gray-scale image using a model image of an object, wherein the expanded image is generated by expanding the model image to generate an expanded image. A generation unit, a contraction image generation unit that contracts the model image to generate a contraction image, a pixel selection unit that selects a pixel near an edge position of the dilation image and the contraction image, and a pixel selection unit. Calculating means for calculating a correlation value between the model image of the object and the inputted grayscale image for the pixels near the edge position.

According to a second aspect of the present invention, there is provided the pattern matching apparatus for a grayscale image according to the first aspect, wherein the pixel selecting means obtains the density data of each of the constituent pixels of the expanded image by performing second-order differentiation. First pixel selecting means for selecting a pixel whose value exceeds a first threshold value; and a second threshold value obtained by secondarily differentiating the density data of each constituent pixel of the contracted image. Second pixel selecting means for selecting a pixel exceeding the value.

[0017]

Since the correlation value between the model image and the input grayscale image is calculated for pixels near each edge position of the dilated image and the contracted image of the model image, the input grayscale image is rotated in either direction. However, correct pattern matching is performed, and position measurement is possible. Further, since the target of the correlation calculation is limited to pixels near each edge position of the dilated image and the contracted pixel, the number of target pixels for pattern matching is efficiently reduced, the calculation amount of the correlation calculation is reduced, and the processing time is shortened. .

In the pattern matching device of the second aspect,
A pixel whose value obtained by secondarily differentiating the density data of each constituent pixel of the dilated image exceeds a first threshold value, and a value obtained by secondarily differentiating the density data of each constituent pixel of the contracted image Is the second
Is selected, pixels near both sides of the edge of the model image are reliably extracted.

[0019]

1 to 3 are explanatory diagrams for explaining the principle of a pattern matching apparatus for a gray-scale image according to the present invention. FIG. 1A is a model image 1 obtained by capturing an image of a model of a target object (in this case, the character “A”), where 3 is an image portion of the character A and 4 is an image portion of the background. In this example, the image portion 3 of the character A is white and the image portion 4 of the background is black. However, the present invention is not limited to this, and a black-and-white inverted model image can be used according to the input image 2.

FIG. 1B shows an expanded image 7 generated by expanding the model image 1.
(2) 'shows a contracted image 8 generated by subjecting the model image 1 to contraction processing. 2 (2) (2) '
In FIG. 7, a broken line b is an edge of the model image 1 before the expansion or contraction processing. Each processing of expansion and contraction is performed a plurality of times according to the degree of rotation of the input image.

FIG. 2A shows the density data (0 to 255 gradations) of each pixel along a certain horizontal line 9 of the expanded image 7. FIG. 3A shows the density data (0 to 255 gradations) of each pixel along the same horizontal line 9 of the contracted image 7.

The dilated image 7 and the contracted image 8 are generated by subjecting the model image 1 to image processing using a predetermined operator. FIG. In FIG. 4, reference numeral 30 denotes a rectangular mask of 3 × 3 pixels for scanning the model image 1 in the direction of the arrow, and the pixel of the model image 1 located at the center of the mask 30 is a target pixel 31. It is.

The expanded image 7 is obtained by sequentially replacing the density data of the target pixel 31 of the model image 31 with the density data of the brightest pixel among the nine pixels in the mask 30 in the mask scanning of the model image 1. Generated.
The contracted image 7 is generated by sequentially replacing the density data of the target pixel 31 of the model image 31 with the density data of the darkest pixel among the nine pixels in the mask 30 in the mask scanning of the model image 1. You.

The value obtained by differentiating the density data of each pixel of the expanded image 7 is as shown in FIG. 2 (2), and the secondary differential value is as shown in FIG. 2 (3). The value obtained by differentiating the density data of each pixel of the contracted image 8 is shown in FIG.
This is as shown in (2), and its secondary differential value is as shown in FIG. 3 (3).

The second derivative of each pixel of the expanded image 7 is compared with a first positive threshold TH1, and a pixel having a value larger than the first threshold TH1 is selected. FIG. 2 (4) shows that the first threshold value TH1
The binarized data that has been digitized is shown.

The second derivative of each pixel of the contracted image 8 is compared with a negative second threshold value TH2, and a pixel having a value smaller than the second threshold value TH2 is selected. FIG. 3D shows that the second threshold value TH2
The binarized data that has been digitized is shown.

When the density data of the model image 1 is sampled by the binarized data shown in FIG.
A gray image 32 as shown in (3) is obtained. In the grayscale image 32, reference numeral 34 denotes an image portion generated based on the sampled density data.
4 is located outside the vicinity of the edge position of the dilated image 7.

When the density data of the model image 1 is sampled by the binarized data shown in FIG.
A gray image 33 as shown in FIG. 1 (3) 'is obtained. In the gray-scale image 33, reference numeral 35 denotes an image portion generated based on the sampled density data, and the image portion 35 is located inside the vicinity of the edge position of the contracted image 8.

The image portion 34 in FIG.
1 (3) 'corresponds to the image portion 3 of the character A near the inside of the edge in the model image 1, and these images FIG. 1 with parts 34 and 35 synthesized
In the grayscale image 36 shown in (4), the density data of the background image portion 4 of the model image 1 and the image portion 3 of the character A
And the same concentration data as the concentration data. FIG. 1
In (4), the dashed line b is the edge of the model image 1, and the image portions 34 and 35 are located near the edge b but are separated from each other by an amount that allows the rotation of the input image.

The model image 1 is shifted one pixel at a time and the input image is scanned over the entire screen.
A correlation operation is performed between the density data of the model image 1 sampled by the digitized data and the density data of the input image 2, and the position of the object is obtained from the scanning position of the model image 1 at which the correlation value is maximum. . In this case, even if the input image 2 is slightly rotated as shown by the broken line a in FIG. 8, the target pixel of the correlation operation is selected from near the edges of the dilated image 7 and the eroded image 8 of the model image 1. Therefore, for both the model image 1 and the input image 2, the density data is sampled from the image portion of the object and the image portion of the background, and correct pattern matching is performed.

FIG. 5 shows an example of a circuit configuration of a pattern matching apparatus based on the above principle. The apparatus in the illustrated example includes a camera 10, a gray image processing unit 11, and a monitor 12. The gray image processing unit 11 includes an A / D converter 13, an expansion / contraction processing unit 25, a differential measurement unit 14, and an LUT 15. , Image memories 16, 17, binary memory 18, display control unit 19,
It includes a D / A converter 20, a control unit 21, and the like.

The A / D converter 13 receives an analog signal of a grayscale image from the camera 10 and converts it into a digital signal. The density data of each pixel constituting this digital signal takes, for example, any value of 256 gradations. The grayscale image (model image 1) obtained by imaging the model of the object is
The data is input to the expansion / contraction processing unit 25 via the A / D converter 13 and stored in one image memory 16. The input image 2 obtained by imaging the measurement object is stored in the other image memory 17 via the A / D converter 13. Each of the image memories 16 and 17 stores density data in pixel units.

The expansion / contraction processing unit 25 generates an expansion image 7 by expanding the model image 1 one or more times, and generates an expansion image 8 by contracting the model image 1 one or more times. . The differential measuring unit 14 obtains a binary differential value by secondarily differentiating the density data of each pixel with respect to the expanded image 7 and the contracted image 8.

The LUT 15 binarizes the secondary differential value of each pixel of the dilated image 7 given by the differential measuring section 14 with the above-mentioned first threshold value TH1, and converts the secondary differential value of each pixel of the dilated image 8 This is for binarizing the differential value with the above-mentioned second threshold value TH2. When an address signal corresponding to the secondary differential value is given to the LUT 15, binary data is output from the corresponding address area. The binarized data for the dilated image 7 and the binarized data for the contracted image 8 are combined and stored in the binary memory 18 in pixel units.

The display control section 19 is for controlling the image display operation of the monitor 12, and inputs a grayscale image signal or binary data of a model or an object to be measured, and performs D / A conversion on one of them. Output to the container 20. D / A converter 2
0 converts the input signal into an analog signal and outputs it to the monitor 12.

The control section 21 includes a CPU 22 which is a main body of control and calculation, and a ROM 23 in which programs are stored.
And a RAM 24 for storing various data. The CPU 22 controls the operation of the entire apparatus according to a program, and stores image memories 16, 17, a binary memory 18, a RAM 2
4 while performing various calculations and processes relating to pattern matching while reading and writing data.

FIG. 6 shows the flow of processing for selecting a target pixel for pattern matching in steps 1 to 6.
(In the figure, each step is indicated by “ST”).

In step 1 of FIG.
In each of the 15 address areas, binary data for performing the binarization processing on the expanded image 7 and the contracted image 8 is set. When the model of the object is imaged by the camera 10 in step 2, the model image 1 is captured by the grayscale image processing unit 11 and stored in the image memory 16, and the model image 1 is An expansion image 7 and a contraction image 8 are generated (Step 3).

In the next step 4, the differential measuring section 14 subjects the dilated image 7 and the eroded image 8 to second-order differential processing, and
Calculate the next differential value. An address signal corresponding to the second derivative of each pixel of the expanded image 7 and the contracted image 8 is L
When given to the UT 15, the LUT 15 outputs binary data from the corresponding address area (step 5). The binarized data for the dilated image 7 and the binarized data for the contracted image 8 are combined and stored in the binary memory 18 in pixel units (step 6).

FIG. 7 shows a procedure of controlling the pattern matching by the CPU 22 in steps 1 to 17. In step 1 of the figure, the CPU 22 clears and initializes the contents of the work area of the RAM 24, takes in the input image 2 obtained by imaging the measurement target object into the grayscale image processing unit 11, and after the A / D conversion It is stored in the image memory 17 (steps 2 and 3).

In the next step 4, the counter j in the CPU 22 for counting the scanning position P _j of the model image 1 is set to zero, and the model image 1 is set to the first scanning position P ₀ (X ₀₎ on the input image 2. , Y ₀ ). Next Step 5
After the CPU 22 sets the counter i for counting the pixel position in the model image 1 to zero, the CPU 22 refers to the binary memory 18 to determine the first pixel position (x ₀ , Read the binarized data of y ₀ ).

If the binarized data is "1",
The determination in step 7 proceeds to step 8 is "YES", the concentration of the first pixel position (x _0, y ₀₎ and density data M _i of the model image 1, the corresponding pixel position of the input image 2 is CPU22 The data I _i is read out from each of the image memories 16 and 17, and in the next step 9, I _i · M _i and I _i ² · M _i ² in the above equation are calculated.

Next, the CPU 22 uses the values calculated in step 9 to calculate the cumulative addition values {I _i · M _i , {I _i · ΣM _i ,}
I _i ² , (ΣI _i ) ² , ΣM _i ² , (ΣM _i ) ² are calculated and stored in the work area of the RAM 24 (step 10). If it is determined in step 7 that the binarized data is "0", steps 8 to 10 are skipped and the above calculation is omitted.

In the next step 11, the CPU 22 determines whether or not the value of the counter i has reached the total number of pixels of the model image 1. Increment the value of i, CP
U22 is the next pixel position (x ₁ ,
The binary data of y ₁ ) is read, and the same procedure as described above is executed.

When the above procedure is repeatedly executed for all the pixel positions in the model image 1, the determination in step 11 becomes "YES", and in the next step 13, the CPU
Numeral 22 executes the calculation of the expression (1) using the calculated value of step 10, and calculates the correlation value r at the first scanning position (X ₀ , Y ₀ ).
₂ (X ₀ , Y ₀ ) is calculated.

In the next step 14, the CPU 22 compares the correlation value r ₂ (X ₀ , Y ₀ ) with the maximum value of the correlation values. In this case, since the maximum value of the correlation value is initially set to zero, the determination in step 14 is “YES”,
In step 15, the correlation value r ₂ (X ₀ , Y ₀ ) is stored in the RAM 24 as the maximum value of the correlation value. In step 16, the coordinates (X ₀ , Y ₀ ) of the scanning position P ₀ are stored in the position of the object ( X ', Y') in the RAM 24. If the determination in step 14 is “NO”, steps 15 and 16 are skipped and the maximum correlation value is not updated.

[0047] At the next step 17, which determines whether the scanning position P _i of the model image 1 has reached the final position by the value of the counter j, in this case, the judgment is "NO"
Therefore, the CPU 22 increments the counter j in step 18 to set the model image 1 to the next scanning position P ₁ (X ₁ , Y ₁ ) on the input image 2 and executes the same procedure as described above. .

The scan position P at which the correlation value is maximized by performing the way the procedure at all scanning positions P _j while scanning the entire screen of the input image 2 by model image 1
_When the coordinates (X ′, Y ′) of _j are obtained and the determination in step 17 becomes “YES”, the CPU 22 ends the pattern matching.

In the above embodiment, the contents of the binary memory 18 are referred to, and when the binary data is "1", the model image 1
The While the density data M _i is read from the respective image memory 16, prior to the pattern matching, the binarized data is density data M of the pixels in the model image 1 of "1"
_i along with the coordinates (x _i , y _i ) of the pixel position
For example, it may be stored in a table.

[0050]

As described above, the present invention performs dilation and erosion processing on a model image to generate an dilated image and a eroded image, and then applies the model image to pixels near each edge position of the dilated image and the eroded image. Since the correlation value between the input grayscale image and the input grayscale image is calculated, correct pattern matching is performed even if the input grayscale image is rotated in either direction, and position measurement is possible. Further, since the target pixel of the correlation operation is limited to pixels near each edge position of the dilated image and the contracted pixel, the target pixel of the pattern matching can be efficiently reduced, and the calculation amount of the correlation operation is reduced.
Processing time can be reduced.

According to the pattern matching apparatus of the second aspect,
A pixel whose value obtained by secondarily differentiating the density data of each constituent pixel of the dilated image exceeds a first threshold value, and a value obtained by secondarily differentiating the density data of each constituent pixel of the contracted image Is the second
Pixels that exceed the threshold of
There is an effect that pixels near both sides of the edge of the model image can be reliably extracted.

[Brief description of the drawings]

FIG. 1 is a principle explanatory diagram for explaining the principle of a pattern matching apparatus for a grayscale image according to the present invention.

FIG. 2 is an explanatory diagram showing a processing result of each processing step for an expanded image.

FIG. 3 is an explanatory diagram showing a processing result of each processing step for a contracted image.

FIG. 4 is an explanatory diagram showing a method of generating an expanded image and a contracted image.

FIG. 5 is a block diagram showing a circuit configuration example of a gray-scale image pattern matching apparatus according to one embodiment of the present invention;

FIG. 6 is a flowchart illustrating a flow of a process of selecting a target pixel for pattern matching.

FIG. 7 is a flowchart showing a control procedure of pattern matching by a CPU.

FIG. 8 is an explanatory diagram showing a conventional normalized correlation density matching method.

FIG. 9 is a principle explanatory diagram showing the principle of a system in which a conventional example is improved.

FIG. 10 is an explanatory diagram showing processing results at each processing stage for the improved method of FIG. 9;

[Explanation of symbols]

 Reference Signs List 10 camera 11 density image processing unit 14 differential measurement unit 15 LUT 21 control unit 22 CPU 23 ROM 24 RAM 25 expansion / contraction processing unit

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G06T ⁷ /00-7/60 G06K 9/36 G06K 9/62

Claims (2)

(57) [Claims] 1. An apparatus for performing pattern matching of an input grayscale image using a model image of an object, comprising: an expanded image generating unit configured to expand the model image to generate an expanded image; A contracted image generating unit that generates a contracted image by performing a contraction process; a pixel selecting unit that selects a pixel near an edge position of the dilated image and the contracted image; A pattern matching apparatus for a gray-scale image, comprising: arithmetic means for calculating a correlation value between a model image of an object and an input gray-scale image. 2. The first pixel selecting means for selecting a pixel whose value obtained by secondarily differentiating the density data of each constituent pixel of the expanded image exceeds a first threshold value. And a second pixel selecting means for selecting a pixel whose value obtained by secondarily differentiating the density data of each constituent pixel of the contracted image exceeds a second threshold value. A pattern matching device for the described gray image.