JPH10162139A - Image processor, and medium where image processing program is recorded - Google Patents

Abstract

PROBLEM TO BE SOLVED: To easily and fast search a reference image from in an input image as to an image processor, which searches a specified reference image from an externally supplied input image, and a medium where the image processing program for actualizing the processor by using a computer. SOLUTION: The image processor, which searches, from the given input image, the specified reference image, is equipped with a reference means 1, which previously stores a frequency distribution Sρθ in a ρθ Hough space found by processing and input image by Hough transformation, a Hough transforming means 2, which transforms the input image into the frequency distribution Tρθ in the ρθ Hough space by Hough transformation, and a search means 3 which decides the correlation between the said frequency distribution Tρθ and frequency distribution sρθ and judges 'whehther or not a reference image in the input image' from whether or not they correlate to each other.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing apparatus for searching for a predetermined reference image from a given input image,
And a medium recording an image processing program for realizing the apparatus using a computer.

[0002]

2. Description of the Related Art Conventionally, a laser processing apparatus or a stepper has been put to practical use in which alignment is performed by automatically recognizing an alignment mark or the like to be processed. Also, a device for automatically recognizing characters and a device for automatically sorting articles flowing on a belt conveyor have been put to practical use.

In these apparatuses, it is important to search for a pre-registered reference image from an input camera image or character image. For use in such image processing, an image processing apparatus for searching for a reference image from an input image and an image processing program for the same have been developed. Recently, as this type of image processing apparatus, an apparatus using a generalized Hough transform is well known.

FIG. 13 is a diagram for explaining image processing (conventional example) using this kind of generalized Hough transform. Here, a case where the reference image 71 is searched from the input image will be described.

First, a reference image 71 is registered in the image processing apparatus in advance. At the time of this registration, the image processing apparatus sets a reference point Rg (Xg, Yg) on the reference image 71. The reference point Rg is set at the position of the center of gravity of the reference image 71 or another appropriate position. Next, the image processing apparatus obtains a gradient for each pixel with respect to the density value of the reference image 71, and extracts an edge line from the magnitude (edge strength) of the gradient.

[0006] The image processing apparatus selects some points on the edge line as points Fj (hereinafter, referred to as "characteristic points Fj") indicating the characteristics of the image. Here, the image processing apparatus sequentially executes the following processes (1) to (4) for each of the feature points Fj.

(1) Based on the gradient direction of the feature point Fj, the tangential direction θ of the edge line at the feature point Fj
j is calculated. (2) An azimuth φj formed by a tangent to the edge line at the feature point Fj and a straight line connecting the feature point Fj and the reference point Rg.
Is calculated. (3) The distance rj between the feature point Fj and the reference point Rj is calculated. (4) The values θj, φj, and rj thus calculated are grouped and stored in the reference table 72 on the internal memory. Through the above processing, the reference table 72 that represents the reference image 71 as a model is completed.

The search for the reference image 71 is performed using the reference table 72 as follows. First, the image processing apparatus takes in an input image from the outside and obtains a gradient of density values of the input image. The image processing apparatus extracts an edge line from the size of this gradient (edge intensity). The image processing apparatus selects some points (X, Y) on the edge line as the feature points Fi of the input image.

Here, the image processing apparatus executes the following processes (1) to (6) for each of the feature points Fi of the input image. (1) The rotation angle θr and the magnification S are set. (2) The tangent direction θi of the edge line at the feature point Fi is calculated based on the gradient direction of the feature point Fi. (3) Search the reference table 72 for a tangent direction θj close to “the sum of the tangent direction θi and the rotation angle θr”. (4) From the lookup table 72, a numerical value (rj, φj) corresponding to the searched tangential direction θj is obtained. (5) Based on these numerical values, Xc = X + rj · S · cos (θj + φj) (1) Yc = Y + rj · S · sin (θj + φj) (2) is calculated, and the reference point candidate Rc is calculated. (Xc, Yc) is obtained. (6) Parameter values Xc, Yc, θ thus obtained
The cumulative value A (Xc, Yc, θr, S) set in the four-dimensional parameter space is incremented corresponding to r and S.

With the above processing, the input image is subjected to a generalized Hough transform on a four-dimensional parameter space. The image processing apparatus repeats such a generalized Hough transform while changing the rotation angle θr and the magnification S by a predetermined step value.

If the input image contains a reference image, the frequency distribution of the accumulated value A has a maximum point. Therefore, the image processing apparatus determines that the cumulative value A (X
c, Yc, θr, S) are extracted as detection points.
The parameter values (Xc, Yc, θr,
Based on S), it is estimated that the reference image linearly mapped with the rotation angle θr and the magnification S exists at the position of the reference point (Xc, Yc) in the input image.

[0012]

In the above-described image processing apparatus, the number of feature points Fi of the input image is set to Nf.
If the number of steps of the rotation angle θr is Nθ and the number of steps of the magnification S is Ns, there is a problem that a huge amount of calculation processing (Nf · Nθ · Ns) must be executed.

Further, in a case where the search is performed while scanning the input image in pixel units, if the pixel size of the input image is horizontal Sx pixels and vertical Sy pixels, (Sx · Sy ·
(Nf · Nθ · Ns).

As described above, since a huge amount of arithmetic processing is performed, the time required for searching for a reference image is long, and there is a problem that the method cannot be applied to a field requiring high-speed image processing. Therefore, an object of the present invention is to provide an image processing apparatus that can easily and quickly search for a reference image from an input image.

According to the second aspect of the present invention, in addition to the object of the first aspect, even when the reference image in the input image is rotated or scaled, the reference image can be easily and quickly searched. It is an object of the present invention to provide an image processing apparatus that can perform the processing.

According to the third aspect of the present invention, in addition to the object of the first aspect, even when the number of feature points of the reference image is different from the number of feature points of the input image, the search of the reference image can be accurately performed. It is an object of the present invention to provide an image processing apparatus that can perform the processing.

A fourth object of the present invention is to provide, in addition to the object of the first embodiment, an image processing apparatus capable of searching a reference image from an input image more quickly.

According to a fifth aspect of the present invention, a medium on which an image processing program for realizing the image processing apparatus of the first to third aspects using a computer is specified as one aspect of the present invention. In a sixth aspect of the present invention, a medium in which an image processing program for realizing the image processing apparatus of the fourth aspect using a computer is recorded is explicitly described as one aspect of the present invention.

[0019]

FIG. 1 is a principle block diagram corresponding to the first and second aspects of the present invention. Claim 1
The invention described in is an image processing apparatus for searching for a predetermined reference image from a given input image, the frequency distribution Sρθ on the ρθ Hough space obtained by performing a Hough transform on the reference image,
Reference means 1 stored in advance, Hough transform means 2 for performing an Hough transform on an input image to convert it into a frequency distribution Tρθ in a ρθ Hough space, and a frequency distribution Tρθ converted by the Hough transform means 2 A search means for determining a correlation with the stored frequency distribution Sρθ and determining “the presence or absence of a reference image in the input image” based on the presence or absence of the correlation;

According to a second aspect of the present invention, in the image processing apparatus according to the first aspect, the search means 3 is based on: ρ = ρ / (average value of ρ) θ = θ− (average value of θ) The correlation between the frequency distribution Tρθ of the input image and the frequency distribution Sρθ of the reference image is determined in the ρθ Hough space standardized in this manner, and the presence or absence of the reference image is determined based on the presence or absence of the correlation.

FIG. 2 is a principle block diagram corresponding to the third aspect of the present invention. According to a third aspect of the present invention, in the image processing apparatus according to the first or second aspect, the search means 3 includes a frequency distribution Tρθ converted by the Hough transform means 2 and a frequency distribution stored in the reference means 1. A correlation calculating unit 3a for calculating a cross-correlation coefficient with the distribution Sρθ, and a correlation for determining a correlation based on the cross-correlation coefficient calculated by the correlation calculating unit 3a and determining the presence or absence of a reference image based on the presence or absence of the correlation. And determining means 3b.

FIG. 3 is a block diagram showing the principle corresponding to the fourth aspect of the present invention. According to a fourth aspect of the present invention, in the image processing apparatus according to any one of the first to third aspects, a gradient of a density value is obtained for an input image, and the gradient directions are coincident and continuous. Continuous edge extracting means 4 for extracting a group of edges to be extracted, and counting the number of pixels of the edge group extracted by the continuous edge extracting means 4,
An edge determination unit (5) for determining a correlation with the number of pixels of an edge group in the reference image, wherein the search unit (3) determines the correlation between the input image and the reference image determined to be correlated by the edge determination unit (5). 3 is performed.

According to a fifth aspect of the present invention, there is provided a medium recording an image processing program for realizing the image processing apparatus according to any one of the first to third aspects using a computer, This is a medium on which an image processing program for causing a computer to function as the above-mentioned reference means 1, Hough conversion means 2, and search means 3 is recorded.

According to a sixth aspect of the present invention, there is provided a medium storing an image processing program for realizing the image processing apparatus according to the fourth aspect by using a computer, wherein the computer is provided with the above-mentioned reference means 1, This is a medium on which an image processing program for functioning as the Hough transform unit 2, the search unit 3, the continuous edge extracting unit 4, and the edge determining unit 5 is recorded.

(Function) In the image processing apparatus according to the first aspect, the Hough transform means 2 performs a Hough transform on an externally input image to convert it into a frequency distribution Tρθ in a ρθ Hough space. On the other hand, the frequency distribution Sρθ in the ρθ Hough space obtained by performing the Hough transform on the reference image is stored in the reference unit 1 in advance.

The exploration means 3 is based on the ρθ Huff space,
The correlation between these frequency distributions Tρθ and Sρθ is determined. The search means 3 determines whether or not a reference image exists in the input image according to the level of the correlation. In the image processing apparatus according to the second aspect, the search means 3 determines the correlation between the frequency distribution Tρθ of the input image and the frequency distribution Sρθ of the reference image on the ρθ Hough space standardized based on the following equation.

Ρ = ρ / (average value of ρ) (3) θ = θ− (average value of θ) (4) In the above expression (3), the frequency distribution Tρθ of the input image is By normalizing the frequency distribution Sρθ of the reference image in the ρ-axis direction, a difference in magnification between the input image and the reference image is suppressed.

On the other hand, in the above equation (4), the frequency distribution Tρθ of the input image and the frequency distribution Sρθ of the reference image are normalized in the θ-axis direction, so that a difference between the input image and the reference image is obtained. The difference in rotation angle is suppressed.

As described above, in the standardized ρθ Hough space, the mutual correlation can be determined without considering the difference between the magnification and the rotation angle which originally exist between the input image and the reference image. . Therefore, it is possible to easily and quickly search the rotated or enlarged / reduced reference image from the input image.

In the image processing apparatus according to the third aspect, the correlation calculating means 3a takes in the frequency distribution Tρθ of the input image and the frequency distribution Sρθ of the reference image, and calculates a cross-correlation coefficient for these frequency distributions. . The correlation determining unit 3b determines whether or not a reference image exists in the input image, according to the magnitude of the cross-correlation coefficient.

Here, the cross-correlation coefficient is represented by the frequency distribution Tρ
This is a value obtained by normalizing the covariance between θ and Sρθ by the standard deviation of each frequency distribution. Therefore, in the cross-correlation coefficient, a difference in pixel density between the reference image and the input image, a difference in the number of feature points, and the like are appropriately suppressed. For example,
The number of feature points of the reference image and the number of feature points of the input image are different,
And when the correlation between the frequency distributions Tρθ and Sρθ is extremely high, the constant A (Ａ) which almost satisfies {Tρθ− (average value of Tρθ)} = A · {Sρθ− (average value of Sρθ)} A> 0) exists. The value of the constant A is a value generated due to a difference in the number of feature points. Even in such a case, the value of the cross-correlation coefficient takes a value of almost “1” regardless of the value of A.

As described above, according to the third aspect of the present invention, by performing the correlation judgment using the cross-correlation coefficient, the frequency distributions Tρθ and Sρθ can be obtained without considering the difference in the pixel density and the number of feature points.
Can be accurately determined. In the image processing apparatus according to the fourth aspect, the continuous edge extracting means 4 obtains a gradient of a density value with respect to the input image. The continuous edge extracting means 4 extracts a group of edges having the same gradient direction and continuous.

The edge determining means 5 counts the number of pixels of each of the edge groups, and determines a correlation with the number of pixels of the edge group in the reference image. Such correlation determination simply and quickly compares only the distribution of the number of pixels of the edge group. Therefore, based on the determination of “no correlation” in the correlation determination, it can be immediately determined that there is no reference image in the input image.

However, since the comparison is performed only for a part of the features of the image, it is difficult to immediately determine that the reference image exists in the input image based on only the “correlation” determination in the correlation determination. It is. Therefore, the correlation determination regarding the number of pixels of the edge group is used for narrowing down search candidates in the search means 3. By narrowing down the search candidates, an image having a low possibility of correlation can be easily and quickly excluded, and the processing time required for the search processing can be significantly reduced.

In the medium according to claim 5, a program for realizing the above-mentioned reference means 1, Hough transform means 2 and search means 3 using a computer is recorded. When the program on this medium is executed by a computer, all of the constituent requirements according to any one of claims 1 to 3 are prepared, and the computer functions as an image processing device.

A program for realizing the above-mentioned reference means 1, Hough transform means 2, search means 3, continuous edge extraction means 4 and edge determination means 5 using a computer is recorded on the medium according to claim 6. You. By executing the program on this medium by a computer, all of the components described in claim 4 are completed, and the computer functions as an image processing device.

[0037]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 4 is a diagram showing a hardware configuration of the embodiment according to the first to sixth aspects. In FIG. 4, a CPU board 11 having an arithmetic processing circuit and a control circuit (not shown) includes a main bus 12 for exchanging data and the like at a normal speed, and a high-speed local bus for exchanging data and the like at a high speed. 13 is connected.

A disk drive unit 14 is connected to the main bus 12, and a recording medium 15 on which an image processing program and the like are recorded is attached to and detached from the disk drive unit 14. The main bus 12 includes a D / A converter 1
The display device 17 is connected via the display device 6. Further, an imaging device 19 is connected to the main bus 12 via an A / D converter 18.

The main bus 12 includes a memory 20 for storing image information, frequency distribution, and the like, and a CPU board 11.
And a high-speed operation processing unit 21 for executing high-speed operation based on an instruction from the CPU. On the other hand, the CPU board 11, the memory 20, and the high-speed processing unit 21 are alternately connected via the high-speed local bus 13.

FIG. 5 is a block diagram relating to the arithmetic function of this embodiment. FIG. 5 shows the memory 20 and the high-speed arithmetic processing unit 21 when the image processing program recorded on the recording medium 15 is executed on the CPU board 11.
The functions realized by using the above are described in block units. In FIG. 5, the input image captured by the imaging device 19 and the new reference image
1 is input to the differential processing unit 33.

The output of the differential processing unit 33 is input to a Hough conversion unit 34 and a continuous edge extraction unit 35, and the output of the Hough conversion unit 34 is input to a normalization processing unit 36. The output of the normalization processing unit 36 is individually input to the reference data storage unit 37 and the correlation coefficient calculation unit 38, while the output of the continuous edge extraction unit 35 is input to the pixel counting unit 39.

The output of the pixel counting section 39 is individually input to the reference data storage section 37 and the edge similarity determination section 40, and the output of the reference data storage section 37 is output to the correlation coefficient calculation section 3.
8 and the edge similarity determination unit 40 are individually input.
The output of the correlation coefficient calculator 38 and the edge similarity determiner 40
Is input to the image recognition processing unit 41 individually.

Here, when the edge similarity determination unit 40 outputs a determination that “there is no similarity in the edge portion”, the image recognition processing unit 41 issues the following instruction (dotted line portion shown in FIG. 5). Emits. That is, the image recognition processing unit 41
It instructs the Hough transform unit 34, the normalization process unit 36, and the correlation coefficient calculation unit 38 to interrupt the “process related to Hough transform”, and instructs the input process unit 31 to immediately start the next process.

As for the correspondence between the invention described in claim 1 and the present embodiment, the reference means 1 corresponds to the reference data storage section 37, the Hough conversion means 2 corresponds to the Hough conversion section 34, and The means 3 corresponds to the correlation coefficient calculator 38 and the image recognition processor 41. Regarding the correspondence between the invention described in claim 2 and the present embodiment, the search means 3 includes a normalization processing unit 36, a correlation coefficient calculation unit 38, and an image recognition processing unit 4
Corresponds to 1.

As for the correspondence between the invention described in claim 3 and the present embodiment, the correlation calculating means 3a corresponds to the correlation coefficient calculating section 38, and the correlation determining means 3b corresponds to the image recognition processing section 41.
Corresponding to Regarding the correspondence between the invention described in claim 4 and the present embodiment, the continuous edge extraction unit 4 corresponds to the continuous edge extraction unit 35, and the edge determination unit 5 corresponds to the pixel counting unit 3
9 and the edge similarity determination unit 40.

Regarding the correspondence between the invention described in claim 5 and the present embodiment, the medium on which the image processing program is recorded corresponds to the recording medium 15. Regarding the correspondence between the invention described in claim 6 and the present embodiment, the medium on which the image processing program is recorded corresponds to the recording medium 15. FIG. 6 is a diagram illustrating a data flow according to the present embodiment.

The operation of this embodiment will be described below with reference to FIGS. (Activation of Image Processing Program) First, the CPU board 11
Reads the image processing program from the recording medium 15 using the disk drive unit 14. CPU board 11
Starts this image processing program and executes the following operation.

(Creation of Reference Data) First, the CPU board 11 executes creation of reference data for a new reference image as follows. The CPU board 11 captures a new reference image from the imaging device 19, the disk drive unit 14, the memory 20, or the like.

The reference image is subjected to averaging compression processing in the input processing section 31 (step S1). Here, the “averaging compression process” means, for example, that an image is 2 × 2
This is a process of reducing the number of pixels by dividing into blocks of about pixels and taking an average value of density values, a median value, and the like for each block. By this processing, the number of pixels of the reference image is reduced to a necessary and sufficient degree for image recognition.

The differential processing unit 33 applies a Prewitt X shown in FIG.
The direction operator is operated to calculate the X component Gx of the gradient of the density value (step S2). Note that this Gx is
This is the same as the Y component in the pattern edge direction. Further, the differential processing unit 33 applies the Prewitt Y-direction operator shown in FIG. 7A to the reference image after the averaging compression, and calculates the Y component Gy of the gradient of the density value (step S3). Gy is the same as the X component in the pattern edge direction.

The differential processing unit 33 calculates the edge intensity | G | = (Gx ² + Gy ² ) ^1/2 (6) based on the gradient (Gx, Gy) of each pixel, and calculates Find the edge strength | G |. The differential processing unit 33 determines the edge intensity | G | of each pixel as a threshold value,
Pixels exceeding a predetermined threshold are extracted as feature points (step S4).

Next, the differential processing unit 33 calculates a gradient angle θ = tan ⁻¹ (Gy / Gx) (7) based on the gradient (Gx, Gy) of the feature point, and calculates this gradient angle. θ is stored as feature point data (step S5). The direction angle of the pattern edge is tan ^-1 (Gx / Gy). Continuous edge extraction unit 3
5 shows, based on the data of the feature points, feature points in which the gradient angle θ matches within a predetermined range and the pixel positions are continuous are classified into individual groups (hereinafter referred to as “edge groups”) for each edge group shown in FIG. "). Further, the continuous edge extraction unit 35 calculates an average value θav for the gradient angles θ of these edge groups, and calculates a normalized gradient angle (θ−θav).
Ask for.

On the other hand, the pixel counting section 39 counts the number N of feature points included in each edge group, and calculates a distribution ratio (N / M) divided by the total number M of feature points. The pixel counting unit 39
The set of the distribution ratio (N / M) of the number of feature points and the standardized gradient angle (θ-θav) is stored in the reference data storage unit 37 (step S6). As a result, information relating to the edge group as shown in FIG.

On the other hand, the Hough transform unit 34 calculates the barycentric position of the feature point by averaging the position coordinates of the feature point in both the X and Y directions. Using the position of the center of gravity as the origin, the position coordinates (x, y) of each feature point are obtained.

The Hough transform unit 34 calculates a perpendicular length ρ = x · cos θ + y · sin θ (8) based on the position coordinates (x, y) and the gradient angle θ for each feature point. . Note that, as shown in FIG. 9A, the perpendicular here is a perpendicular that is lowered from the origin toward the tangent of the edge located at the feature point.

As a result, the Q feature points aligned in a straight line on the XY space are converted into the cumulative frequency Q at one point (ρ, θ) on the ρθ Hough space, as shown in FIG. Is performed (step S7). FIG. 10 is an explanatory diagram three-dimensionally showing a frequency distribution on the ρθ Hough space. The normalization processing unit 36 calculates ρ = ρ / ρav (9) θ = θ−θav (10) based on the average value ρav of the length ρ and the average value θav of the gradient angle θ. Is calculated, and the ρθ Hough space is normalized in the ρ-axis direction and the θ-axis direction (step S8).

By performing such a standardization process,
For example, in the various images shown in FIG. 11, differences in magnification and rotation angle are appropriately suppressed, and the same frequency distribution is obtained. The standardization processing unit 36 stores the frequency distribution Sρθ of the reference image in the standardized ρθ Hough space in the reference data storage unit 37. Further, the normalization processing unit 36 uses the frequency distribution Sρθ of the reference image,

(Equation 1)

(Equation 2) Is calculated, the standard deviation s of the frequency distribution Sρθ is obtained, and stored in the reference data storage unit 37 (step S9). Note that n in the expression is the number of samples on the ρθ Huff plane.

With the above operation, the creation of the reference data for the new reference image is completed. (Image Recognition Processing for Input Image) Next, the processing operation of image recognition for an input image will be described.

First, the CPU board 11 converts the input image into
Captured from the imaging device 19. This input image is subjected to averaging compression processing in the input processing unit 31 (step S11). The differential processing unit 33 applies an X direction operator of Prewitt to the input image after the averaging compression, and calculates an X component Gx of a gradient of the density value (step S12).

Further, the differential processing section 33 applies the Prewitt Y-direction operator to the input image after the averaging compression, and calculates the Y component Gy of the gradient of the density value (step S13). The differential processing unit 33 calculates the magnitude of the gradient (Gx, Gy) for each pixel, and obtains the edge strength | G |.

The differential processing unit 33 calculates the edge intensity | of each pixel.
G | is determined as a threshold, and pixels exceeding a predetermined threshold are extracted as feature points (step S14). Next, the differential processing unit 3
3 is the gradient angle θ of the gradient (Gx, Gy) of the feature point
Is calculated, and the gradient angle θ is stored as feature point data (step S15). The continuous edge extraction unit 35 groups feature points having the same gradient angle θ within a predetermined range and continuous pixel positions into edge groups from the feature point data. Further, the continuous edge extraction unit 35 calculates an average value θav for the gradient angles θ of these edge groups,
The normalized gradient angle (θ−θav) is obtained for each edge group.

On the other hand, the pixel counting section 39 counts the number N of feature points of each edge group, and calculates a distribution ratio (N / M) divided by the total number M of feature points. Here, the edge similarity determination unit 40 determines that the gradient angle (θ−θav) and the distribution ratio (N / M) on the input image side match the information on the edge group in the reference data storage unit 37 within the allowable range. It is determined whether or not.

If they do not match within the allowable range, the image recognition processing section 41 immediately determines that there is no reference image in the input image and suspends the image recognition processing operation (step S2).
2). On the other hand, if they match within the allowable range, the image recognition processing unit 41 determines that there is a high possibility that the reference image exists in the input image, and starts the image recognition processing on the ρθ Hough space as follows. .

First, the Hough transform unit 34 calculates the position of the center of gravity of the feature point by averaging the position coordinates of the feature point of the input image in both the X and Y directions. Using the position of the center of gravity as the origin, the position coordinates (x, y) of each feature point are obtained. Huff converter 34
Calculates the length ρ of the perpendicular by substituting the position coordinates (x, y) and the gradient angle θ for each feature point into equation (8) (step S17).

The standardization processing unit 36 calculates the average value ρav of the length ρ
And the average value θav of the gradient angle θ,
Equation (10) is calculated, and the ρθ Hough space is converted into the ρ axis direction and θ.
Normalization is performed in the axial direction (step S18). The normalization processing unit 36 uses the frequency distribution Tρθ of the input image on the standardized ρθ Hough space,

(Equation 3)

(Equation 4) To calculate the standard deviation σt of the frequency distribution Tρθ (step S19). Note that m in the equation is the number of samples on the ρθ Huff plane.

The correlation coefficient calculator 38 calculates a frequency distribution Sρθ of the reference image stored in the reference data storage 37 and a frequency distribution Tρθ of the input image.

(Equation 5) Is calculated, and a covariance Cst is obtained (step S20).
Note that k in the expression (15) is the number of samples on the ρθ Huff plane.

Further, the correlation coefficient calculating section 38 reads out the standard deviation s of the reference image stored in the reference data storage section 37 and, together with the standard deviation σt of the input image and the value of the covariance Cst,

(Equation 6) Is calculated, and a cross-correlation coefficient P between the frequency distributions Sρθ and Tρθ is obtained (step S21).

The image recognition processing section 41 determines the absolute value of the cross-correlation coefficient P as a threshold value (step S22). if,
When this absolute value exceeds a predetermined threshold (for example, about 0.8), the image recognition processing unit 41 determines that the reference image exists in the input image. On the other hand, when the absolute value does not exceed the predetermined threshold, the image recognition processing unit 41 determines that there is no reference image in the input image.

According to the operation described above, in this embodiment, it is possible to reliably determine whether or not a reference image exists in an input image. Further, in the ρθ Hough space, straight lines on the image are collected into one point, so that a complicated line segment forming the image is converted into a simple point frequency distribution. Therefore, in the present embodiment, it is only necessary to perform the correlation determination with respect to the frequency distribution of points, and the amount of calculation processing can be significantly reduced.

Further, according to the present invention, since the reference image is recorded in the form of a simple frequency distribution, the storage capacity required for storing the reference image can be remarkably reduced. In particular, an apparatus that performs character recognition records a large number of reference images, and thus the present invention is very suitable. In the present embodiment, the correlation is determined on the normalized ρθ Hough space. In the standardized ρθ Hough space, the mutual correlation can be determined by ignoring the difference in the magnification and the rotation angle that originally exist between the input image and the reference image.
Therefore, the rotated or scaled-up reference image can be easily and quickly searched from the input image.

Further, since the cross-correlation coefficient P is used for the correlation judgment, the difference between the reference image and the input image is not taken into account in the input image without considering the “difference in pixel density” and “difference in the number of feature points”. Can be determined whether or not a reference image exists.
Further, in the present embodiment, since information on edge groups is compared as preprocessing for image recognition, it is possible to accurately narrow down candidates to be searched and perform correlation determination of the frequency distribution. Therefore, the processing time required for image recognition can be significantly reduced.

In the above-described embodiment, only the determination as to whether or not the reference image exists in the input image is performed, but the application of the present invention is not limited to this. For example, as shown in FIG. 12, a partial region of an input image may be cut out, and whether or not a reference image exists in the partial region may be determined in the same manner as in the embodiment. In addition, by performing main scanning and sub-scanning of this partial area, a partial area including the reference image can be searched from the input image.

Further, in the above-described embodiment, the case where a single reference image is used has been described, but the present invention is not limited to this. The same determination as in the embodiment may be sequentially performed on a plurality of reference images. Furthermore, in the above-described embodiment, the cross-correlation coefficient P is used for determining the correlation of the frequency distribution, but the present invention is not limited to this.
Any method can be used as long as it is a correlation determination method. For example, the correlation determination of the frequency distribution may be performed using a matching method such as a residual sequential test method (SSDA method) or a sum of squared difference method.

In the above-described embodiment, the cross-correlation coefficient P is calculated for determining the correlation of the frequency distribution.
The present invention is not limited to this, and a numerical value that can be converted into a cross-correlation coefficient may be calculated.

(Equation 7) For example, the square value (so-called contribution rate) of the cross-correlation coefficient P is calculated using the above equation, and the threshold value of this square value is determined.
It may be determined whether or not a reference image exists in the input image. Such a configuration is preferable because the square root operation required for calculating the cross-correlation coefficient P can be omitted.

Further, in the above embodiment, the number N of pixels in the edge group is normalized by the total number M of feature points, but the present invention is not limited to this. For example, a correlation determination may be made for each pixel number N of the edge group. In the above-described embodiment, the type of the medium on which the image processing program is recorded is not described, but the present invention is not limited to the type of the medium. For example, a ROM (read only memory), a semiconductor recording medium,
An optical recording medium, a magnetic recording medium or a magneto-optical recording medium may be used.

Furthermore, in the above-described embodiment, an edge portion of an image is extracted and used as a feature point, but the present invention does not necessarily require this processing. For example, in a black and white line drawing or the like, either white or black may be used as a feature point without extracting an edge portion. In the above-described embodiment, the gradient of the density value is obtained by using the operator of Prewitt, but any spatial differential filter can be used. For example, the gradient may be calculated using a Sobel operator as shown in FIG. 7B.

In the above-described embodiment, a series of operations are performed by executing the image processing program.
The present invention is not limited to this. For example, FIG.
A part or all of the configuration shown in FIG. 1 may be realized by replacing it with a gate array circuit, a custom IC, or another hardware configuration. Further, in the above-described embodiment, the apparatus including the high-speed operation processing unit 21 has been described, but the present invention is not limited to this. For example, data processing as shown in FIG. 6 may be performed using a general computer.

[0078]

As described above, according to the first aspect of the present invention, the correlation between the input image and the reference image is determined on the ρθ Hough space to determine whether or not the reference image exists in the input image. Can be reliably determined. Normally, in the ρθ Huff space, straight lines on the image are aggregated into one point,
Complex line segments that make up the image are converted to a simple point frequency distribution. Therefore, in the present invention, it is often necessary to perform the correlation determination on the frequency distribution of a simple point, and the amount of calculation processing required for the correlation determination can be reduced.

In addition, since the amount of calculation required for the correlation determination is reduced as described above, it is possible to efficiently reduce the time required for searching for the reference image. for that reason,
It is suitable for applications requiring high-speed image processing. Further, in comparison with the case where the reference image is recorded in a bitmap format or the like, in the present invention, the reference image may be recorded in a simple frequency distribution format on the ρθ Hough space. Therefore, the amount of information to be stored generally decreases. Therefore, it is possible to efficiently save the storage capacity of the reference means, and to record a larger number of reference images than before.

In particular, in a device for performing character recognition, a large number of reference images are recorded, so that the present invention is very suitable. According to the second aspect of the present invention, the standardized ρθ
The correlation of the frequency distribution is determined on the Hough space. In the standardized ρθ Hough space, differences in the magnification and the rotation angle that originally exist between the input image and the reference image are suppressed. Therefore, the mutual correlation can be accurately determined without being greatly affected by differences in magnification and rotation angle. As a result, even for rotated or scaled reference images,
It is possible to easily and quickly search from within the input image.

In particular, in the conventional image processing apparatus employing the generalized Hough transform, it is necessary to repeat complicated arithmetic processing many times while finely changing the magnification and the rotation angle over a wide range. However, according to the second aspect of the present invention, since the difference between the magnification and the rotation angle can be easily suppressed, it is sufficient to perform the correlation determination as it is or to perform the correlation determination with the magnification and the rotation angle slightly shifted. Therefore, the amount of calculation processing is further reduced, and the calculation processing time can be significantly reduced.

According to the third aspect of the invention, the correlation of the frequency distribution is determined based on the cross-correlation coefficient. Therefore, it is possible to accurately determine whether or not the reference image exists in the input image, without being largely influenced by the difference in pixel density or the difference in the number of feature points between the reference image and the input image. In the image processing device according to the fourth aspect, since the distribution of the number of pixels of the edge group is compared, the candidates for the image to be searched can be simply and quickly narrowed down.

Therefore, an image having a low correlation can be easily and quickly excluded, and the processing time required for the search processing can be shortened efficiently. In the medium according to the fifth aspect, the image processing program on the medium is executed by a computer, so that all the components described in any one of the first to third aspects are completed. Therefore, claims 1 to
A computer may function as the image processing device according to any one of the items (3) to (3).

In the medium according to the sixth aspect, the image processing program on the medium is executed by a computer, so that all of the components described in the fourth aspect are completed. Therefore, the computer can function as the image processing device according to the fourth aspect.

[Brief description of the drawings]

FIG. 1 is a principle block diagram corresponding to the first and second aspects of the present invention.

FIG. 2 is a principle block diagram corresponding to the invention described in claim 3;

FIG. 3 is a principle block diagram corresponding to the invention described in claim 4;

FIG. 4 is a diagram showing a hardware configuration of an embodiment according to claims 1 to 6;

FIG. 5 shows, in block units, functions realized by using the CPU board 11, the memory 20, the high-speed processing unit 21, and the like when the image processing program recorded on the recording medium 15 is executed on the CPU board 11. FIG.

FIG. 6 is a diagram showing a data flow of the present embodiment.

FIG. 7 is a diagram illustrating an example of a differential filter on space.

FIG. 8 is a diagram illustrating information on an edge group.

FIG. 9 is a diagram illustrating Hough transform.

FIG. 10 is an explanatory diagram three-dimensionally showing a frequency distribution on a ρθ Hough space.

FIG. 11 is an explanatory diagram of a normalization process on a ρθ Hough space.

FIG. 12 is a diagram illustrating pattern matching in an XY space.

FIG. 13 is a view for explaining typical image processing (conventional example) by generalized Hough transform.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Reference means 2 Hough conversion means 3 Exploration means 3a Correlation calculation means 3b Correlation determination means 4 Continuous edge extraction means 5 Edge determination means 11 CPU board 12 Main bus 13 High-speed local bus 14 Disk drive unit 15 Recording medium 16 D / A converter 17 Display device 18 A / D converter 19 Imaging device 20 Memory 21 High-speed operation processing unit 31 Input processing unit 33 Differential processing unit 34 Hough transform unit 35 Continuous edge extraction unit 36 Normalization processing unit 37 Reference data storage unit 38 Correlation coefficient operation unit 39 pixel counting unit 40 edge similarity determination unit 41 image recognition processing unit 72 reference table

Claims (6)

[Claims] An image processing apparatus for searching a given reference image for a predetermined reference image from a given input image, wherein a frequency distribution in a ρθ Hough space obtained by performing a Hough transform on the reference image is stored in advance. Means, performing a Hough transform on the input image, a Hough transform means for converting to a frequency distribution on a ρθ Hough space, and a frequency distribution converted by the Hough transform means, and a frequency distribution stored in the reference means. A search unit that determines a correlation and determines the presence or absence of the reference image based on the presence or absence of the correlation. 2. The image processing apparatus according to claim 1, wherein the search means includes a ρθ Huff normalized based on ρ = ρ / (average value of ρ) θ = θ− (average value of θ). An image processing apparatus, wherein a correlation between a frequency distribution of the input image and a frequency distribution of the reference image is determined in space, and the presence or absence of the reference image is determined based on the presence or absence of the correlation. 3. The image processing apparatus according to claim 1, wherein said searching means is configured to determine a correlation between a frequency distribution converted by said Hough conversion means and a frequency distribution stored in said reference means. Correlation calculating means for calculating the number of relations, and correlation determining means for determining a correlation based on the cross-correlation coefficient calculated by the correlation calculating means, and determining the presence or absence of the reference image based on the presence or absence of the correlation. An image processing apparatus characterized by the above-mentioned. 4. The image processing apparatus according to claim 1, wherein a gradient of a density value is obtained for the input image, and a group of edges having the same gradient direction and being continuous is extracted. Continuous edge extracting means, and edge determining means for counting the number of pixels of the edge group extracted by the continuous edge extracting means and determining a correlation with the number of pixels of the edge group in the reference image; The image processing apparatus according to claim 1, wherein the search is performed on the input image and the reference image determined to have a correlation by the edge determination unit. 5. A medium storing an image processing program for realizing the image processing apparatus according to claim 1 using a computer, wherein the medium is a computer. A medium in which an image processing program for functioning as the Hough transform unit and the search unit is recorded. 6. A medium in which an image processing program for realizing the image processing apparatus according to claim 4 using a computer is recorded, wherein the computer is provided with the reference unit, the Hough conversion unit, the search unit, A medium in which an image processing program for functioning as the continuous edge extracting unit and the edge determining unit is recorded.