JP2003271974A - Method and detector for detecting rotation angle of image, and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To precisely calculate a rotation angle in response to an image condition in an input image of a broken or blurred image and the like. <P>SOLUTION: In this rotation angle detecting method for detecting the rotation angle of a pattern image in the image using Zernike moment, learning patterns different respectively in the image conditions, and the rotation angles found based on the learning patterns are stored preliminarily, and the rotation angle is detected in response to the image condition in the input image. A level of the Zernike moment, the number of black picture elements in the image, a length of an outline in the image, a ratio of the length of the outline in the image to an area of the black picture elements, an average picture element value in the image, a dispersion of picture element values in the image, an average of edge strengths in the image, and a dispersion of the edge strengths in the image are used as the image conditions in the input image. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a rotation angle detecting method, apparatus and recording medium for detecting a rotation angle of a specific pattern in an image.

[0002]

2. Description of the Related Art Conventionally, as a rotation angle detecting method, apparatus and recording medium for detecting a rotation angle of a pattern image in an image using Zernike moment, Japanese Patent Laid-Open No. 11-2
The one described in Japanese Patent No. 57947 is known. This one uses multiple Zernike polynomial tables of different orders and ranks to find the Zernike moment of the input pattern image, and calculates the rotation angle of the image based on this Zernike moment to accurately determine the rotation angle. It is supposed to be calculated.

[0003]

However, in the conventional technique, when the input image is blurred or blurred and the quality of the image is deteriorated, the value of the Zernike moment fluctuates, and therefore it is calculated. There is a problem that an error occurs in the rotation angle.

An object of the present invention is to solve the above problems and to accurately calculate a rotation angle corresponding to the state of an input image such as blurring or blurring.

[0005]

The present invention is directed to Zernike
A rotation angle detection method for detecting a rotation angle of a pattern image in an image using a moment, wherein a learning pattern having different image states and a rotation angle obtained from the learning pattern are stored in advance, and an image of the input image is stored. The feature is that the rotation angle is detected according to the state.

The present invention is also characterized in that the magnitude of Zernike moment is used as the image state of the input image.

The present invention is also characterized in that the number of black pixels in the image is used as the image state of the input image.

Further, the present invention is characterized in that the length of the contour in the image is used as the image state of the input image.

Further, the present invention is characterized in that the ratio of the length of the contour in the image to the area of the black pixel is used as the image state of the input image.

Further, the present invention is characterized in that an average pixel value in the image is used as the image state of the input image.

Further, the present invention is characterized in that the variation of pixel values in the image is used as the image state of the input image.

Further, the present invention is characterized in that the average of edge strengths in the image is used as the image state of the input image.

Further, the present invention is characterized in that a variation of edge strength in the image is used as the image state of the input image.

According to the present invention, the image input means for inputting the pattern image, the Zernike polynomial table storage means for storing the Zernike polynomial table, the center of the pattern image and the center of the Zernike polynomial table are combined to form the pattern image. Of the Zernike moment, the Zernike moment calculating means for calculating the Zernike moment, the Zernike moment calculating means for calculating the magnitude of the Zernike moment, and the angle calculated from the standard pattern in which the image states are different from each other. It is characterized in that it is provided with a storage means for storing each, a rotation angle calculation means for calculating the rotation angle of the image from the Zernike moment, and a rotation angle output means for outputting the calculated rotation angle.

The present invention also has a function of inputting a pattern image and a function of storing a Zernike polynomial table.
The function of calculating the Zernike moment of the pattern image by combining the center of the pattern image and the center of the Zernike polynomial table, the function of calculating the magnitude of the Zernike moment, and the image state are calculated from different standard patterns. The computer is made to realize a function of storing the angle of rotation for each magnitude of the Zernike moment, a function of calculating the rotation angle of the image from the Zernike moment, and a function of outputting the calculated rotation angle. I am trying.

[0016]

DESCRIPTION OF THE PREFERRED EMBODIMENTS First, the Zernike moment will be described.

The Zernike moment is described in the paper (Who
i-Yul Kim and Po Yuan, 'A Pr
acticlaPattern Recognitio
n System for Translation, S
calle and Rotation Invarian
ce, 'Proc. CVPR'94, pp. 391-3
96, June 1994).
It is defined by the following formula.

[Equation 1]

[Equation 2]

[Equation 3] In the above formulas (1) to (3), n is Zernike.
The order of the moment, 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 (An in the above equation (3)
m) 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 the specific mark f (x, y). At this time, the values of x and y need to be normalized according to the size of the specific mark. That is, the distance from the center of the specific mark to the furthest point is 1
The coordinate system is used by enlarging / reducing so that

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, it takes a lot of processing time. 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.

[Equation 4]

[Equation 5] By using this radial polynomial table, 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, the Zernike moment is calculated by the following equations (6) and (7) instead of the equations (2) and (3), and the processing speed is increased.

[Equation 6]

[Equation 7] [First Embodiment]

FIG. 1 shows the configuration of a rotation angle detecting device according to a first embodiment of the present invention. In FIG. 1, the rotation angle detection device includes an image input unit 1 to which a pattern image is input,
A product-sum operation unit 2 that performs a product-sum operation of the pattern image and the Zernike polynomial table 3, and the Zernike polynomial table 3
And a Zernike moment calculating unit 4 for calculating the magnitude of Zernike moment, and a dictionary 5 for holding the angle θ calculated from the standard pattern of different image states such as blurring and crushing for each Zernike moment magnitude.
And a rotation angle calculation unit 6 that calculates the rotation angle of the image from the Zernike moment, and a rotation angle output unit 7 that outputs the calculated rotation angle.

FIG. 2 is a flow chart showing the operation of the rotation angle detecting device according to the first embodiment of the present invention.

A pattern image is input to the image input section 1. When the pattern image is input, its center coordinates are given, and the pattern image is assumed to be normalized (step 101).

The product-sum operation unit 2 aligns the center image of the pattern image with the center of the Zernike polynomial table 3 (step 102).

After the alignment, the sum of products of each pixel value of the pattern image and the corresponding value of the above table is performed to calculate the Zernike moment (ZM) (step 10).
3).

Here, the pattern image and one type of Zernike polynomial table (ZPT) prepared in advance.
Performs a product-sum operation with 3. FIG. 3 shows an example of the Zernike polynomial table (ZPT) 3. 3 (a) and 3
In (b), the white pixel has a value of -1, the black pixel has a value of 1, and the gray of the halftone has a value of 0. Further, the center of the circle is the center coordinates (0, 0) of the Zernike polynomial table (ZPT) 3, the right side of the X axis is positive, and the upper side of the Y axis is positive.

That is, in this embodiment, since the input image is a binary image, the sum of the pixel values of the Zernike polynomial table (ZPT) 3 corresponding to the black pixels of the input image is calculated.

In this embodiment, the Zernike polynomial table (ZPT) 3 shown in FIG. 3 is used to calculate the ZM of the real part (Re) and the imaginary part (Im) of the frequency 1.

The ZM calculator 4 calculates the size M of ZM using the following formula (step 104). The M = Re ² + Im ² rotation angle calculation unit 6 first calculates the rotation angle θ _Z from ZM using the following formula (step 105). θ _Z = arctan (−Im / Re)

When the rotation angle θ _P is similarly calculated from the detection target pattern, the rotation angle θ _P is not always 0, so the rotation angle θ _O between the detection target pattern and the input image is It is calculated by the formula of θ _O = θ _Z −θ _P , but in the present embodiment, a plurality of θ _P are previously stored in the dictionary 5 according to the size of ZM in order to deal with the fluctuation of the input image. The rotation angle θ _O between the detection target pattern and the input image is calculated by the following formula (step 106). θ _O = θ _Z − θ _{P (M)}

Then, the rotation angle output unit 7 outputs the calculated rotation angle θ _O (step 107).

Here, a method for obtaining θ _{P (M)} will be described with reference to FIG.

The character "A" shown on the left side of FIGS. 4A to 4C is an example of a learning pattern, and FIG. 4B shows the character "A" as an appropriate image. 4 (a) is shown in FIG.
An image that looks fainter than in (b) is shown in FIG.
An image that is more crushed than in (b) is shown.

The graphs shown on the right side of FIGS. 4A to 4C are ZM (real part Re) -rotation when the image ("A") shown on the left side is rotated to calculate ZM. It is a graph of an angle. M ₁ , m ₂ , and m _{3 in} these graphs are Z
The square root of the size of M, θ _{P (M1)} , θ _{P (M2)} , θ
_{P (M3)} is the rotation angle θ _P obtained from the learning pattern “a” on the left side.

Generally, when an image is crushed or blurred, the size of ZM changes as well as θ _P in this way. Therefore, the relationship between the size of ZM and θ _P is shown in FIG. By obtaining it as shown in, it is possible to use an appropriate θ _P corresponding to the input image, and the detection accuracy of the rotation angle is improved. Further, at that time, by setting the maximum value and the minimum value of the size of ZM as Mmin and Mmax in FIG. 5, respectively, an image of a very low quality in which crushing and blurring are remarkable and a wrong pattern are input. If
It is also possible to take a method of interrupting the processing.

Further, in the present embodiment, the size of ZM is used as the parameter representing the change of the input image, but the number of black pixels in the mark can be used in addition to this, and in that case, in FIG. It suffices to replace m ₁ , m ₂ , and m ₃ with parameters such as the number of black pixels in the input image to generate the correspondence relationship as shown in FIG. 5 and store it as a dictionary. As an alternative to the ZM calculation unit 4 (see FIG. 1) for calculating the magnitude of Zernike moment, the number of black pixels in the image, the length of the contour in the image, the length of the contour in the image and the area of the black pixel. Ratio, average pixel value in image, variation in pixel value in image (variance / standard deviation), average of edge strength (pixel value gradient) in image, variation in edge strength (pixel value gradient) in image, etc. The calculation unit may calculate the parameter of

The above-mentioned contour length is the number of adjacent boundaries between white pixels and black pixels in the input image. Also,
The average pixel value is the average value of the pixel values in the input image (when the input is a multi-valued image).

As for the edge strength in the image, many methods for measuring the edge strength have been proposed.
Document "Basics of image recognition [II] -feature extraction, edge strength, texture analysis-", Ohmsha, Shunji Mori, P1
06 The edge strength as shown in the lower part can be used. That is, there are two methods to determine the direction and strength of the edge of each pixel.

(1) When the edge strengths in the east (x) and south (y) directions are sx and sy, respectively, {(sx)
+ (Sy)} ^1/2 , or a method in which the sum of the absolute value of sx and the absolute value of sy is the edge strength, and tan ⁻¹ [sx / sy] is the edge direction,

(2) There is a method in which the strength of the edge in eight directions is calculated, the maximum value among them is taken as the strength of the edge, and the direction in which the strength of the edge is maximized is taken as the direction of the edge.

Then, using the method of (1), Sob
el, Prewitt (differential type) operator,
Examples of the method (2) include Prewitt (template type), Robinson, and 3-level mask.

Prior art (Japanese Patent Laid-Open No. 11-257947)
In Fig. 4, one typical learning pattern,
Since only the learning pattern shown in (b) is used, when a faint or crushed pattern is given as the input image, θ _{P (M1)} −θ
An error of about _{P (M2)} , θ _{P (M3)} −θ _{P (M2)} occurred, but according to the present invention, this error is eliminated or can be made extremely small. It is possible to accurately calculate the rotation angle corresponding to the image state of the input image such as scrapes.

In addition, the amount of calculation can be reduced as compared with the method of performing the affine transformation on the image to match with the dictionary image.

[Second Embodiment] The present invention is not limited to the above-mentioned embodiment, but can be realized by software. When the present invention is implemented by software, as shown in FIG. 6, a computer system including a CPU, a memory, a display device, a hard disk, a keyboard, a CD-ROM drive, a scanner, etc. is prepared, and a computer such as a CD-ROM is prepared. A program that realizes the rotation angle detection function of the present invention is recorded on the readable recording medium. An image input from an image input unit such as a scanner is temporarily stored in a hard disk or the like. When the program is started, the temporarily stored image data is read, the rotation angle detection process is executed, and the detection result is output to a display or the like.

[0044]

As described above, according to the present invention,
The learning patterns having different image states and the rotation angles obtained from the learning patterns are stored in advance, and the rotation angles are detected according to the image states of the input images. The rotation angle can be calculated accurately in accordance with the image state.

[Brief description of drawings]

FIG. 1 shows a configuration of a first embodiment of the present invention.

FIG. 2 shows a processing flowchart of the first embodiment of the present invention.

FIG. 3 (a) shows ZPT (real number part). (B) shows ZPT (imaginary part).

FIG. 4 (a) a slightly faint learning pattern image,
And ZM when this image is rotated to calculate ZM
The relationship of (real part Re) -rotation angle is shown. (B) shows a standard learning pattern image and the relationship between ZM (real part Re) and rotation angle when ZM is calculated by rotating this image. (C) ZM (real number part R) when the learning pattern image slightly crushed and ZM is calculated by rotating this image
e) -The relationship of a rotation angle is shown.

FIG. 5 shows an example of a relationship diagram between the magnitude of ZM and θ _P.

FIG. 6 shows a configuration of a second exemplary embodiment of the present invention.

[Explanation of symbols]

1 Image input section 2 Sum of products operation unit 3 Zernike polynomial table (ZPT) 4 Zernike moment (ZM) calculator 5 dictionary 6 Rotation angle calculator 7 Rotation angle output section

Claims (11)

[Claims] 1. A rotation angle detecting method for detecting a rotation angle of a pattern image in an image by using Zernike moment, wherein a learning pattern having different image states and a rotation angle obtained from the learning pattern are stored in advance. A rotation angle detection method characterized by detecting a rotation angle according to an image state of an input image. 2. The rotation angle detecting method according to claim 1, wherein the magnitude of the Zernike moment is used as the image state of the input image. 3. The rotation angle detecting method according to claim 1, wherein the number of black pixels in the image is used as the image state of the input image. 4. The rotation angle detecting method according to claim 1, wherein the length of the contour in the image is used as the image state of the input image. 5. The rotation angle detecting method according to claim 1, wherein a ratio of a length of a contour and an area of a black pixel in the image is used as an image state of the input image. 6. The rotation angle detecting method according to claim 1, wherein an average pixel value in the image is used as the image state of the input image. 7. The rotation angle detecting method according to claim 1, wherein a variation in pixel values in the image is used as the image state of the input image. 8. The rotation angle detecting method according to claim 1, wherein an average of edge intensities in the image is used as the image state of the input image. 9. The rotation angle detecting method according to claim 1, wherein a variation of edge strength in the image is used as the image state of the input image. 10. A Zernike moment of the pattern image by combining an image input means for inputting a pattern image, a Zernike polynomial table storage means for storing a Zernike polynomial table, and a center of the pattern image with a center of the Zernike polynomial table. And a Zernike moment calculating means for calculating the magnitude of the Zernike moment, and an angle calculated from a standard pattern in which image states are different for each Zernike moment magnitude. A rotation angle detecting device comprising: a storage unit, a rotation angle calculation unit that calculates a rotation angle of an image from the Zernike moment, and a rotation angle output unit that outputs the calculated rotation angle. 11. A function of inputting a pattern image, a function of storing a Zernike polynomial table, and a function of calculating the Zernike moment of the pattern image by aligning the center of the pattern image and the center of the Zernike polynomial table. A function to calculate the magnitude of the Zernike moment, a function to save the angle calculated from the standard pattern with different image states for each magnitude of the Zernike moment, and to calculate the rotation angle of the image from the Zernike moment A recording medium for causing a computer to realize a function and a function of outputting the calculated rotation angle.