JPH09147109A - Method and device for specific mark detection - Google Patents

Abstract

PROBLEM TO BE SOLVED: To cope with the movement, rotation, and power variation of a specific mark without increasing the capacity of a memory needed to detect the specific mark by making use of Zernike moment. SOLUTION: A radial polynomial table (106) corresponding to the size of the specific mark when the enlargement rate of an input image is maximum is prepared. This table contains only values in a 1/4 area of (x>=0 and y>=0). This table is enlarged or reduced according to the enlargement/reduction rate of the input image and values in the remaining 3/4 area are determined on the basis of the values of the table to complete a table (108). The value of the Zernike moment is calculated (114) by performing product-sum operation between the table and a specific mark candidate image cut out of the input image and the degree of difference from a dictionary (118) is calculated (116), and when the degree of difference is smaller than a threshold value, it is judged (120) that the specific mark has been detected.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image recognition technique, and more particularly to a method and an apparatus for detecting a specific mark which may exist at an arbitrary position in an image.

[0002]

2. Description of the Related Art In an image reading apparatus, when trying to detect a specific mark in an image read from a document, the position, orientation, and reading magnification of the document are not always constant, so that the following conditions must be met. desired. 1) Not affected by the position (movement) of the specific mark. 2) It is not affected by the rotation of the specific mark. 3) It is not affected by the scaling of the specific mark (including unidirectional scaling in only one direction).

As a conventional technique that can be used for such a purpose, reference [1] (Whoi-Yul Kimand Po Yuan, "A
Practicla Pattern Recognition System for Tran
slation, Scale amd Rotation Invariance, "Pro
c. CVPR'94, pp. 391-396, June 19
94), there is a method using a Zernike moment. This document [1] also describes speeding up of processing by forming a table of Radial Polynominal (radial polynomial).

Further, the character image is divided into eight regions by eight line segments coming out from the center of gravity, and the product of the value of the pixel inside and the distance from the center of gravity is calculated as a moment for each divided region, and the region is calculated. Japanese Unexamined Patent Publication No. 6-195512 describes a method of using the cumulative value of moments for each as a feature for character recognition. This method requires very complicated control to deal with image rotation and scaling.

[0005]

The method using the Zernike moment as described in the above-mentioned document [1] can deal with the position, rotation and scaling of an image, and the radial polynomial of the Zernike moment. It is also possible to speed up the process by creating a table. However, in order to realize a practical specific mark detection apparatus that can cope with various enlargement / reduction ratios of an image by this method, it is necessary to devise such as reducing the memory required for storing the radial polynomial table.

It is an object of the present invention to provide a practical method and apparatus for detecting specific marks in an image using Zernike moments.

[0007]

According to the first aspect of the present invention,
In the method of detecting a specific mark in an image using the Zernike moment, a table of radial polynomials corresponding to the size of the specific mark is prepared in advance, and the table is used as a reduction / enlargement ratio of the image at the time of detection processing. Expand accordingly /
The Zernike moment of the image is obtained by generating a reduced table and calculating the product sum of the generated table and the image.

According to a second aspect of the present invention, in a method of detecting a specific mark in an image using a Zernike moment, a radial polynomial table corresponding to the size of the specific mark when the enlargement ratio of the image is maximum is provided. The table is prepared in advance, and a table obtained by enlarging / reducing the table according to the reduction / enlargement ratio of the image is generated at the time of detection processing, and the sum of products of the generated table and the image is calculated. It is characterized by finding the Zernike moment of.

According to a third aspect of the invention, in the specific mark detection method according to the first or second aspect of the invention, the radial polynomial table prepared in advance is (x ≧ 0, y ≧ 0).
It holds that only the value of the 1/4 area of the above is held, and the value of the remaining area is determined at the time of the detection processing based on the held value.

According to a fourth aspect of the present invention, there is provided first table storage means for storing a table of radial polynomials according to the size of the specific mark, and image input means for inputting an image at a designated reduction / enlargement ratio. An image storage means for storing the input image by the image input means, an enlargement / reduction means for enlarging / reducing the table stored in the first table storage means according to a reduction / enlargement ratio of the input image, Second table storing means for storing the table enlarged / reduced by the enlarging / reducing means, and a black connected component having substantially the same size as the specific mark from the input image stored in the image storing means is the specific mark candidate area. The mark cutting means for extracting the image and the center of the image of the specific mark candidate area and the center of the table stored in the second table storing means are combined to form the image and the table. Zernike for calculating the size of the Zernike moment by the product-sum calculation of the Le
Moment calculating means, dictionary storing means for storing the average and standard deviation of the magnitudes of the Zernike moments calculated in advance for specific marks as a dictionary, the dictionary stored in the dictionary storing means, and the Zernike moment calculating A difference degree calculating unit for calculating the degree of difference between the Zernike moment and the magnitude of the Zernike moment calculated by the means, and detection determining means for determining detection / non-detection of the specific mark based on the difference degree calculated by the difference degree calculating unit. It is characterized by having.

According to a fifth aspect of the present invention, in the specific mark detecting apparatus according to the fourth aspect of the invention, the first table storage means stores a radial mark corresponding to the size of the specific mark when the magnification of the input image is maximum. It is characterized in that a table of polynomials is stored.

According to a sixth aspect of the invention, in the specific mark detecting apparatus according to the fourth or fifth aspect of the invention, the table stored in the first table storing means is (x ≧ 0, y ≧
Only the value of 1/4 area of 0) is held, and the value of the remaining area is determined by the enlarging / reducing means based on the held value.

[0013]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will be described below with reference to the drawings in order to clarify the embodiments of the present invention.

FIG. 1 is a block diagram of a specific mark detecting apparatus according to an embodiment of the present invention. It should be noted that this specific mark detection device is used as a single device, or is incorporated in another device such as a digital copier.

In FIG. 1, reference numeral 100 denotes a scanner for scanning a document as a target of specific mark detection to read image information, and reference numeral 102 denotes an image for inputting image data of a document as a target of mark detection using the scanner 100. An input unit 104 is an image buffer for storing the image data input from the image input unit 102. 106 is a buffer for storing a radial polynomial table according to the size of a specific mark to be detected, and 108 is a buffer 1
A radial polynomial table enlarging / reducing unit for generating a table in which the radial polynomial table in 06 is enlarged / reduced according to the reduction / enlargement ratio 122 at the time of image input, and 110 stores the expanded / reduced radial polynomial table. Is the buffer. Reference numeral 112 denotes a mark cutout unit that cuts out a mark candidate area according to the size of a specific mark to be detected from the input image stored in the image buffer 104, 114
Using the expanded / reduced radial polynomial table stored in the buffer 110,
This is a Zernike moment calculation unit that calculates the magnitude of the Zernike moment of the mark candidate area cut out by. Reference numeral 118 denotes a dictionary memory that stores a dictionary of a specific mark to be detected, and reference numeral 116 compares the magnitude of the Zernike moment calculated by the Zernike moment calculating unit 114 with the dictionary of the specific mark in the dictionary buffer 118 to determine the degree of difference. A dissimilarity calculating unit 120 calculates a detection determining unit that determines whether the specific mark is detected or not based on the calculated dissimilarity.

Now, the Zernike moment and its application to mark detection will be described. The Zernike moment is defined by the following equation, as described in detail in the above-mentioned document [1].

[0017]

(Equation 1)

[0018]

(Equation 2)

[0019]

(Equation 3)

In the above equations (1) to (3), n is the order of the Zernike moment and m is the rank. The absolute value of m must be less than or equal to n, and the difference between the absolute values of n and m must be even.

Rnm (ρ) in the above equation (1) is called a radial polynomial, and is a value calculated from the rank, the order and the distance from the center. Re _{(Anm) in the} above equation (2) is the real part of the Zernike moment, and Im (equation 3) above
_(Anm) is the imaginary part of the Zernike moment. The magnitude of the Zernike moment Re _(Anm) ² + Im _(Anm) ² is a value that is invariant to the rotation of the image, and therefore can be used for detecting a specific mark. When used, the values of x and y need to be normalized according to the size of the specific mark. That is,
The coordinate system is used by enlarging / reducing so that the distance from the center of gravity of the specific mark to the furthest point is 1.

In order to actually calculate the Zernike moment, if the calculation of the above equations (2) and (3) is executed every time the mark candidate area is cut out, a lot of processing time is required, so that it is practical. Not relevant. Therefore, a table (radial polynomial table) of values calculated by the following equations (4) and (5) according to the size of the specific mark is created in advance.

[0023]

(Equation 4)

[0024]

(Equation 5)

If this radial polynomial table is used, the Zernike moment can be calculated by taking the product sum of each pixel value and the value of the radial polynomial table. That is, instead of the above equations (2) and (3), the following equation (6),
The Zernike moment is calculated by the equation (7).

[0026]

(Equation 6)

[0027]

(Equation 7)

When the radial polynomial table is used, when the input image is univariably scaled, the prepared radial polynomial table is enlarged / reduced according to the magnification of univariate magnification, and the enlarged / reduced radial polynomial table is input. By applying it to an image, the Zernike moment of a univariately magnified image can be calculated.

In the present embodiment, in the radial polynomial table stored in the buffer 106, when the reduction / enlargement ratio of the input image is 1 in the X direction and 2 in the Y direction (2 times enlargement only in the Y direction). In the case of univariable magnification was done) was created in a form corresponding to. By enlarging / reducing this radial polynomial table according to the reduction / enlargement ratio when an image is actually input, a radial polynomial table necessary for detecting a specific mark on an input image is generated and used.

Since the maximum value / minimum value of the enlargement ratio / reduction ratio of the image is usually determined by the image input device such as the scanner 100, the radial polynomial table corresponding to the case where the enlargement ratio is the maximum is prepared and stored in advance. By doing so, it is possible to expect an improvement in the detection accuracy when the image is enlarged and input in the range below the maximum enlargement ratio (because the resolution of the radial polynomial table is improved). In the present embodiment, such an advantage of improving the resolution can be obtained up to double the enlargement in the Y direction.

Further, since the radial polynomial table is a periodic function with respect to ρ, it is not always necessary to prepare the values of the radial polynomial table for all areas, that is, for all (x, y). It is possible to prepare only the value of and to easily generate the values of the remaining areas based on that value. Therefore, in the present embodiment, the radial polynomial table stored in the buffer 106 is a table that holds the values of only the 1/4 area of (x ≧ 0, y ≧ 0).

For example, if the image of the specific mark to be detected is 4
If the size is 8 × 48 pixels, 2 of (x ≧ 0, y ≧ 0)
A radial polynomial table of 4 × 24 pixels may be prepared. However, as described above, in the present embodiment, Y
The radial polynomial table is saved in a form in which only the directions are enlarged by a factor of 2. Therefore, the size of the radial polynomial table stored in the buffer 106 is 24 pixels in the x direction and Y
The direction is 48 pixels.

Specific processing contents of this embodiment will be described below. Reduction / enlargement ratio 122 before inputting images
Is input by the user using, for example, a ten-key pad.
Next, in the image input unit 102, the scanner 100 is used to read an image at a specified reduction / enlargement ratio, and the image data is stored in the image buffer 104. here,
The reduction / enlargement ratios in the x and Y directions are both specified as 1,
It is assumed that the image is input without being enlarged or reduced.

When the reduction / enlargement ratio of the input image is designated,
The processing of the radial polynomial table enlarging / reducing unit 108 is executed. The outline of the process is shown in FIG. In the radial polynomial table enlarging / reducing unit 108, the buffer 10
The radial polynomial table is read from 6 and enlarged / reduced according to the reduction / enlargement ratio of the image input (step 400). As described above, in this embodiment, the buffer 10
The radial polynomial table stored in 6 is 2 in the Y direction.
It is a table corresponding to the case of univariable magnification of double, and it is necessary to reduce the table to 1/2 in the Y direction in order to match the input image with a reduction / enlargement ratio of 1.

FIG. 2 schematically shows an example of enlargement / reduction of such a radial polynomial table. 2B corresponds to the case where the reduction / enlargement ratio of the input image is 1,
The values in the radial polynomial table 200 prepared in advance are thinned out by one pixel in the Y direction to obtain 1 in the Y direction.
A radial polynomial table 204 reduced to / 2 is obtained. 2A, the input image is 2/3 in the Y direction.
This is an example in the case of being reduced by a factor of two, and the table 202 is obtained by thinning out the values of the table 200 in the Y direction at a rate of 2 pixels for every 3 pixels. 2C shows an example in which the input image is magnified 8/3 times in the Y direction.
6 is a table enlarged in the Y direction. The pixel values of the table 206 are the same as the pixel values of the upper and lower ends of the table 200 for the upper and lower end pixels, but the corresponding values of 2 or 3 pixels of the table 200 are used for the other pixels. Calculated.

In this way, the radial polynomial table enlarged / reduced according to the reduction / enlargement ratio of the input image is generated, but as described above, in the present embodiment, it is stored in the buffer 106. The value of the radial polynomial table that is present is only the value of the quarter area of (x ≧ 0, y ≧ 0), and the expanded / reduced radial polynomial table is similarly (x ≧ 0.
It has only the value of the quarter area of 0, y ≧ 0). Therefore,
The radial polynomial table enlargement / reduction unit 108
1/4 of the reduced radial polynomial (x ≧ 0, y ≧ 0)
Based on the value of the area, the value of the remaining 3/4 area is determined and the table is completed (step 402). How to determine the remaining values will be described with reference to FIG. The radial polynomial table shown in FIG. 3 is an example of a table for the real part of the first and first orders.

As shown in FIG. 3A, (x≤0, y
For the ¼ area of ≧ 0), the value of the ¼ area of (x ≧ 0, y ≧ 0) is copied for each column. In the case of the real part table, when m is an odd number, Multiply the value by -1 before copying m
When is an even number, the value is copied as it is. The same applies to the imaginary part table, but when m is an odd number, the value is copied as it is, and when m is an even number, the value is multiplied by -1 before copying. Next, as shown in FIG. 3B, (x ≧ 0, y ≦
0) for the 1/4 region of 1 / (x ≧ 0, y ≧ 0)
The values of the four areas are copied for each row for the 1/4 area of (x ≦ 0, y ≦ 0). If the table for the imaginary part is
Copy it after multiplying it by 1.

By such a process, 2 in the Y direction.
1x of (x ≧ 0 ,, y ≧ 0)
From the radial polynomial table that has values only in four areas,
A complete radial polynomial table for calculating the Zernike moments of a 1x input image is generated and is temporarily stored in buffer 110 (step 4).
041).

Next, in the mark cutting section 112,
A connected component of black pixels is extracted from the input image stored in the image buffer 104, and a connected component of black pixel determined to have a size substantially equal to the size of the specific mark to be detected is set as a mark candidate area. The coordinates, for example, the diagonal point coordinates of the circumscribed rectangle are sent to the Zernike moment calculation unit 114. Specifically, for example, when the threshold for determination is LTh and the size of the specific mark to be detected is MSize, when the width BlW and the height BlH of the black pixel connected component satisfy the following two equations. , The black pixel connected component is detected as a mark candidate area, and its coordinates are Zernike /
It is sent to the moment calculation unit 114.

[0040]

(Equation 8)

[0041]

(Equation 9)

Here, it is assumed that the black pixel connected components A, B and C as shown in the following table are extracted from the input image.

[0043]

[Table 1]

Assuming that MSize = 48 and Lth = 4, the result of mark cutout is as shown in the following table, and only the black pixel connected components A and B are detected as mark candidate areas.

[0045]

[Table 2]

In the Zernike moment calculation section 114, the mark candidate area image is fetched from the image buffer 104 from the coordinates passed from the mark cutting section 112,
The center of gravity of the mark candidate area image is calculated. Then, after the center of gravity and the center of the radial polynomial table stored in the buffer 110 are aligned, the sum of the products of the pixel value of the mark candidate area image and the corresponding value of the table is calculated. Radial polynomial table has 2 for real part and imaginary part
Since there are two, the sum of products calculation is performed for each. Then, the sum of the square of the product sum of the real part and the square of the product sum of the imaginary part is calculated to obtain the magnitude of the Zernike moment. FIG. 3C schematically shows how the real part 304 of the Zernike moment is obtained by multiplying the value of the radial polynomial table (here, the table for real part) 300 and the corresponding pixel value of the image 302. It is shown in the figure.

In the present embodiment, assuming that the magnitudes of the Zernike moments up to the third order are used for mark detection, the Zernike moment calculation results shown in the following table are assumed.

[0048]

[Table 3]

In the dictionary buffer 118, with respect to the specific mark to be detected, the average value Dic () of the magnitudes of Zernike moments obtained from a plurality of images and the standard deviation Std.
() Is stored as a dictionary. Dissimilarity calculation unit 116
At the Zernike moment calculation unit 116, the Zernike moment magnitude F () and
The degree of difference Dcb between the input image (mark candidate area image) and the specific mark is calculated by the following formula using the dictionary of the specific mark.

[0050]

(Equation 10)

The following table shows an example of a dictionary of specific marks and the calculation result of the degree of difference for the black pixel connected components A and B.

[0052]

[Table 4]

The detection determination unit 120 determines whether or not the specific mark is detected based on the calculated dissimilarity Dcb. Specifically, a certain threshold ThD is compared with the difference Dcb, and it is determined that the specific mark is detected when ThD> Dcb. ThD =
Assuming 4.00, the detection judgment result of the black pixel connected components A and B is as shown in the following table.

[0054]

[Table 5]

[0055]

According to the invention described in each of claims 1 to 6, it is possible to detect with high accuracy a specific mark at an arbitrary position in an image at an arbitrary angle and with an arbitrary scaling factor. , High-speed processing is possible by using the radial polynomial table, and it is necessary to save the radial polynomial table because only one is prepared instead of multiple radial polynomial tables corresponding to various image enlargement / reduction ratios. Memory capacity is small. According to the inventions of claims 2 and 5, by improving the resolution of the radial polynomial table,
It is possible to improve the mark detection accuracy when the image is input in an enlarged size. According to the third and sixth aspects of the present invention, since only the value in the 1/4 area of the radial polynomial table is stored, the memory capacity required for table storage (memory capacity of the first table storage means) can be further reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram of a specific mark detection device according to an exemplary embodiment of the present invention.

FIG. 2A is a diagram showing an example of reduction of a radial polynomial table. (B) It is a figure which shows the example of reduction of a radial polynomial table. (C) It is a figure which shows the example of expansion of a radial polynomial table.

FIG. 3 (a) (x ≦ 0, y of a radial polynomial table)
It is a figure which shows the determination method of the value of the 1/4 area | region of> = 0. (B) It is a figure which shows the determination method of the value of the 1/4 area | region of (x ≧ 0, y ≦ 0) and the 1/4 area of (x ≦ 0, y ≦ 0) of the radial polynomial table. (C) It is a figure which shows typically a mode that a Zernike moment is obtained by multiplying a radial polynomial table and an input image.

FIG. 4 is a schematic flowchart of a process of a radial polynomial table enlarging / reducing unit.

[Explanation of symbols]

100 Scanner 102 Image Input Unit 104 Image Buffer 106 Radial Polynomial Table Storage Buffer 110 Enlargement / Reduction Radial Polynomial Table Storage Buffer 112 Mark Extraction Unit 114 Zernike Moment Calculation Unit 116 Dissimilarity Calculation Unit 118 Dictionary Buffer 120 Detection Judgment Unit

Claims (6)

[Claims] 1. A method of detecting a specific mark in an image using a Zernike moment, wherein a table of radial polynomials according to the size of the specific mark is prepared in advance, and the table is used for the detection process. A specific mark detection method characterized in that a table enlarged / reduced according to a reduction / enlargement ratio is generated, and a Zernike moment of the image is obtained by calculating a product sum of the generated table and the image. 2. A method for detecting a specific mark in an image using a Zernike moment, wherein a table of radial polynomials corresponding to the size of the specific mark when the enlargement ratio of the image is maximum is prepared in advance, A characteristic is that a Zernike moment of the image is obtained by generating a table obtained by enlarging / reducing the table according to a reduction / enlargement ratio of the image at the time of detection processing and calculating a product sum of the generated table and the image. The specific mark detection method. 3. The method for detecting a specific mark according to claim 1 or 2, wherein a table of radial polynomials prepared in advance holds only values in a quarter region of (x ≧ 0, y ≧ 0), A method of detecting a specific mark, wherein the value of the remaining area is determined at the time of detection processing based on the held value. 4. A first table storage means for storing a table of radial polynomials according to the size of a specific mark,
An image input unit for inputting an image at a designated reduction / enlargement ratio, an image storage unit for storing the input image by the image input unit, and an image input unit for storing in the first table storage unit according to the reduction / enlargement ratio of the input image. The enlarging / reducing means for enlarging / reducing the stored table, the second table storing means for storing the table enlarged / reduced by the enlarging / reducing means, and the input image stored in the image storing means. A mark cutout unit that extracts a black connected component having substantially the same size as the specific mark as a specific mark candidate region, a center of gravity of an image of the specific mark candidate region, and a center of the table stored in the second table storage unit are provided. The Zernike moment calculation means for calculating the magnitude of the Zernike moment by calculating the sum of products of the image and the table, and the specific mark. Dictionary storage means for storing the average value and standard deviation of the magnitude of the Zernike moment calculated in advance as a dictionary, the dictionary stored in the dictionary storage means, and the Zernike moment calculated by the Zernike moment calculation means. A dissimilarity calculating unit that calculates a dissimilarity with respect to the magnitude of the moment; and a detection determining unit that determines detection / non-detection of the specific mark based on the dissimilarity calculated by the dissimilarity calculating unit. Specific mark detection device. 5. The specific mark detection apparatus according to claim 4, wherein a table of radial polynomials according to the size of the specific mark when the enlargement ratio of the input image is maximum is stored in the first table storage means. A specific mark detection device characterized by: 6. The specific mark detection device according to claim 4 or 5, wherein the table stored in the first table storage means holds only a value in a quarter area of (x ≧ 0, y ≧ 0). The value of the remaining area is determined by the enlarging / reducing means based on the held value.