KR20130091366A - Method and system for inspecting object using 3-dimensional patern kernel, computer-readable recording medium and computer apparatus - Google Patents

Abstract

PURPOSE: An object inspection method using a three-dimensional pattern kernel and a system, a computer readable recoding medium, and a computer device are provided to obtain second kernel computation data by computing a binary-code result and a second three-dimensional pattern kernel of a reference image according to a determined directional pattern of an object image, thereby efficiently inspecting an object in which a symbol is imprinted on the surface. CONSTITUTION: An object inspection method using a three-dimensional pattern kernel is as follows. An object inspection system translates images of an object into binary code and determines a directional pattern of the object by using the result of the binary code (S100, S200). The system computes the binary-code result and a second three-dimensional pattern kernel of a reference image according to the determined directional pattern and obtains second kernel computation data (S300). The system calculates an error score based on the second kernel computation data, thereby determining a quality of the object (S400). [Reference numerals] (AA) Start; (BB) Finish; (S100) Translate images of an object into binary code; (S200) Determine a directional pattern of the object by using the result of the binary code; (S300) Kernel computation data by computing the result of the binary code and the three-dimensional pattern kernel duce the directional pattern; (S400) Determine a quality of the object by calculating an error score based on the second kernel computation data

Description

METHOD AND SYSTEM FOR INSPECTING OBJECT USING 3-DIMENSIONAL PATERN KERNEL, COMPUTER-READABLE RECORDING MEDIUM AND COMPUTER APPARATUS}

The present invention relates to a method and system for inspecting an object using a three-dimensional pattern kernel, a computer readable recording medium, and a computer device. Specifically, a computer-readable recording medium that includes an object inspection method and system using a three-dimensional pattern kernel and a program that performs an object inspection method using a three-dimensional pattern kernel that examines an object whose symbol is imprinted on a surface by using an image. And a computer device.

A method of inspecting an object using an image input from a camera has been proposed. The technique of acquiring and binarizing an image of an object and comparing the reference image with a reference image value has been widely used, and a technique of inspecting an object in which a symbol is imprinted on a surface using an image has been proposed.

When checking the quality of an object imprinted with a symbol on the surface, for example, a character, etc., if the symbol is deeply imprinted on the surface of the object, the symbol is easily judged through binarization, and whether foreign matter is attached from the binarization data value for the rest of the symbol. B. It is relatively easy to judge the damage.

However, in the conventional simple binarization technique, when a character or the like is shallowly imprinted on the surface of an object, the imprint is not clear, or the depth of the imprint is irregular, an image for recognizing a pattern, for example, a character is not easily extracted. For example, in order to recognize the acquired image and determine the front and back or top and bottom of the object, it is necessary to correctly extract a symbol, for example, a character imprinted on the object. However, in the conventional simple binarization technique, when the imprint is generally formed deep, the character pattern region is well extracted, but when the imprint is shallow, it is not easy to extract an image for recognition of a pattern, for example, a character.

Even when the adaptive binarization technique is applied to solve this problem, the degree of binarization can be changed by changing the parameter (N, sigma) when the imprinting is shallow. The data in the remaining areas can also change, reducing the overall performance of the defect inspection.

The present invention is to solve the above-described problem, to provide a technique for more efficiently inspecting the object is stamped on the surface using the image.

According to the present invention, the second three-dimensional pattern kernel of the reference image according to the determined direction pattern for the object image is computed and the second kernel operation data is calculated to more efficiently inspect the object whose symbol is imprinted on the surface. An object inspection method and system using a 3D pattern kernel are provided.

In order to solve the above problem, according to a first embodiment of the present invention, a method for inspecting an object in which a symbol is imprinted on a surface using an image, comprising: binarizing an image obtained from the object; Determining a direction pattern of the object using the binarization result; Compute the binarization result with the second three-dimensional pattern kernel of the reference image according to the determined direction pattern, wherein the second three-dimensional pattern kernel has different weights between the symbol part and the remaining area, and calculates the second kernel operation data from the operation result. Making; And determining whether the object is good or not by calculating an error score from the second kernel operation data. An object inspection method using a 3D pattern kernel including a is proposed.

Also, in one example, the binarizing may adaptively binarize the acquired image.

According to an example, in the determining of the direction pattern, the inclination of the object may be determined from the obtained video image or the binarized image, and the direction pattern of the object may be determined using the binarization result whose slope is corrected.

According to another example, the determining of the direction pattern may include: calculating a first three-dimensional pattern kernel for each direction according to a predetermined top, bottom, back and forth of the first binarization result and the reference image as a binarization result, The three-dimensional pattern kernel is a pattern kernel for each direction in which a symbol portion of an object is weighted higher than the remaining surface area, and generating first kernel operation data in each direction according to each operation; Calculating a matching score from the first kernel operation data corresponding to each direction and determining a direction pattern from a pattern kernel corresponding to the first kernel operation data for which the maximum matching score is calculated; . ≪ / RTI > In this case, in the calculating of the second kernel operation data, the second binarization result may be calculated as the binarization result with the second 3D pattern kernel according to the determined direction pattern.

According to one example, the second three-dimensional pattern kernel may be a three-dimensional pattern kernel that is weighted differently in the order of the boundary line of the object> the remaining surface area excluding the symbol> the symbol portion. .

In another example, calculating the second kernel operation data includes: inverting the first three-dimensional pattern kernel corresponding to the determined direction pattern such that the weight of the 'symbol portion' is lower than the 'rest surface area'. Forming a second three-dimensional pattern kernel having a weight in order of 'border line of object'>'remaining surface area excluding symbols'>'symbolpart'; Calculating second kernel operation data by calculating a second three-dimensional pattern kernel and a second binarization result; . ≪ / RTI >

에서 산출되고, 제1 커널연산 데이터로부터 산출되는 매칭스코어는 식,

에서 산출되고, 제2 커널연산 데이터는 식,

에서 산출되고, 제2 커널연산 데이터로부터 산출되는 에러 스코어는 식,

에서 산출될 수 있고, 여기에서, 커널 사이즈는 P×Q이고, I1(i,j)는 제1 이진화 결과의 (i,j) 좌표위치의 픽셀 값이고, I2(i,j)는 제2 이진화 결과의 (i,j) 좌표위치의 픽셀 값이고, M1(k,l)은 4개의 방향별 패턴 마스크 각각의 제1의 3차원 패턴 커널 M1의 커널 행렬 내의 (k,l) 위치의 데이터 값이고, M2(k,l)은 제2의 3차원 패턴 커널 M2의 커널 행렬 내의 (k,l) 위치의 데이터 값이고, B_ij는 s=0, 1, 2, 3의 4개의 방향별 M1 커널 행렬의 연산 각각에 따른 제1 커널연산 데이터 각각의 (i,j) 좌표위치의 데이터 값이고, S_ij는 제2 커널연산 데이터의 (i,j) 좌표위치의 데이터 값이고, Mch_s 는 s=0, 1, 2, 3의 4개 방향별 제1 커널연산 데이터 각각으로부터 산출되는 각각의 매칭스코어이고, T_err는 제2 커널연산 데이터 S_ij 로부터 산출되는 에러 스코어이다.
Further, according to one example, the first kernel operation data may be represented by an equation,

The matching score calculated from and calculated from the first kernel operation data is

The second kernel operation data is calculated by

And an error score calculated from the second kernel operation data is

Wherein the kernel size is P × Q, I1 (i, j) is the pixel value at the (i, j) coordinate position of the first binarization result, and I2 (i, j) is the second Is the pixel value of the (i, j) coordinate position of the binarization result, and M1 (k, l) is the data of position (k, l) in the kernel matrix of the first three-dimensional pattern kernel M1 of each of the four direction pattern masks. Value, M2 (k, l) is the data value at position (k, l) in the kernel matrix of the second three-dimensional pattern kernel M2, and B _ij is four directions of s = 0, 1, 2, 3 M i s is the data value of the (i, j) coordinate position of each of the first kernel operation data according to each operation of the M1 kernel matrix, S _ij is the data value of the (i, j) coordinate position of the second kernel operation data, and Mch _s Is each matching score calculated from each of the first kernel operation data for each of four directions of s = 0, 1, 2, and 3, and T_err is an error score calculated from the second kernel operation data S _ij .

According to one example, in determining the good or bad, the error score may be normalized to determine a defect of the object surface.

In addition, according to one embodiment, the object may be a tablet drug.

Next, in order to solve the above-described problem, according to a second embodiment of the present invention, in the inspection system for inspecting an object whose symbol is imprinted on the surface using an image, a binarization module for binarizing an image obtained from the object ; The second 3D pattern kernel generates a 3D pattern kernel of the reference image according to the direction pattern of the object determined by the calculation and determination module. Generating kernel generation module; And determining the direction pattern of the object using the binarization result, calculating the binarization result and the second three-dimensional pattern kernel, calculating second kernel operation data from the operation result, and calculating an error score from the second kernel operation data. An operation and determination module for determining whether the object is good or bad; An object inspection system using a three-dimensional pattern kernel is proposed.

In one example, the binarization module may perform adaptive binarization on the obtained image.

Also, in one example, the calculation and determination module may determine the inclination of the object from the obtained image image or the binarized image and determine the direction pattern of the object by using the binarization result with the inclination corrected.

In addition, according to one example, the kernel generation module further generates a first three-dimensional pattern kernel for each direction along the top and bottom of the reference image, the first three-dimensional pattern kernel has a symbol portion of the object than the remaining surface area It is a highly weighted pattern kernel, and the arithmetic and judging module computes the first binarization result and the first three-dimensional pattern kernel as a result of binarization, respectively, to generate first kernel operation data in each direction and to generate the first kernel in each direction. Each matching score is calculated from one kernel operation data, and the direction pattern of the object is determined from the pattern kernel corresponding to the first kernel operation data for which the maximum matching score is calculated, and the second three-dimensional pattern kernel and the second binarization result are determined. The second kernel operation data may be calculated to calculate the error score, and the fitness of the object may be determined by calculating the error score.

In addition, in one example, the second 3D pattern kernel may be a pattern kernel that is weighted differently in the order of the boundary line of the object> the remaining surface area excluding the symbol> the symbol portion.

According to one example, the kernel generation module inverts the first three-dimensional pattern kernel corresponding to the direction pattern determined by the calculation and determination module so that the weight of the 'symbol portion' is lower than the 'rest surface area', so that the 'object' A second three-dimensional pattern kernel having a weight that is changed in the order of the boundary lines of '>' remaining surface areas excluding symbols'>'symbolparts' may be generated.

에서 산출되고, 제1 커널연산 데이터로부터 산출되는 매칭스코어는 식,

에서 산출되고, 제2 커널연산 데이터는 식,

에서 산출되고, 제2 커널연산 데이터로부터 산출되는 에러 스코어는 식,

에서 산출될 수 있다. 이때, 커널 사이즈는 P×Q이고, I1(i,j)는 제1 이진화 결과의 (i,j) 좌표위치의 픽셀 값이고, I2(i,j)는 제2 이진화 결과의 (i,j) 좌표위치의 픽셀 값이고, M1(k,l)은 4개의 방향별 패턴 마스크 각각의 제1의 3차원 패턴 커널 M1의 커널 행렬 내의 (k,l) 위치의 데이터 값이고, M2(k,l)은 제2의 3차원 패턴 커널 M2의 커널 행렬 내의 (k,l) 위치의 데이터 값이고, B_ij는 s=0, 1, 2, 3의 4개의 방향별 M1 커널 행렬의 연산 각각에 따른 제1 커널연산 데이터 각각의 (i,j) 좌표위치의 데이터 값이고, S_ij는 제2 커널연산 데이터의 (i,j) 좌표위치의 데이터 값이고, Mch_s 는 s=0, 1, 2, 3의 4개 방향별 제1 커널연산 데이터 각각으로부터 산출되는 각각의 매칭스코어이고, T_err는 제2 커널연산 데이터 S_ij 로부터 산출되는 에러 스코어이다.
Further, according to one example, the first kernel operation data may be represented by an equation,

The matching score calculated from and calculated from the first kernel operation data is

The second kernel operation data is calculated by

And an error score calculated from the second kernel operation data is

Can be calculated from In this case, the kernel size is P × Q, I1 (i, j) is the pixel value at the (i, j) coordinate position of the first binarization result, and I2 (i, j) is (i, j) of the second binarization result. ) Is the pixel value of the coordinate position, M1 (k, l) is the data value of the position (k, l) in the kernel matrix of the first three-dimensional pattern kernel M1 of each of the four direction-specific pattern masks, and M2 (k, l) l) is the data value at position (k, l) in the kernel matrix of the second three-dimensional pattern kernel M2, and B _ij is applied to each of the four directions of the M1 kernel matrix s = 0, 1, 2, and 3. Is the data value of the (i, j) coordinate position of each of the first kernel operation data, S _ij is the data value of the (i, j) coordinate position of the second kernel operation data, Mch _s is s = 0, 1, 2 and 3 are matching scores calculated from each of the first kernel operation data for each of four directions, and T_err is an error score calculated from the second kernel operation data S _ij .

In another example, the computation and determination module may normalize the error score to determine defects on the object surface.

Also, in one embodiment, the object may be a tablet.

Next, in order to solve the above-described problem, in the computer-readable recording medium according to the third embodiment of the present invention, the object inspection method using the three-dimensional pattern kernel according to any one of the first embodiments described above A computer readable recording medium containing a program for performing the above is proposed.

Further, in order to solve the above-described problem, according to the fourth embodiment of the present invention, in a computer device, a computer readable recording medium according to the third embodiment described above is mounted, and for reading a computer readable recording medium. Accordingly, a computer device in which the recorded program is performed is proposed.

According to an embodiment of the present invention, the second three-dimensional pattern kernel of the reference image according to the determined direction pattern for the object image is computed to calculate the second kernel operation data to view the object whose symbol is imprinted on the surface. It can be inspected efficiently.

In addition, according to one embodiment of the present invention, by separately inspecting the symbol region and the remaining region, foreign matter and / or damage inspection strength of the remaining portions except the symbol region may be adjusted.

It is apparent that various effects not directly referred to in accordance with various embodiments of the present invention can be derived by those of ordinary skill in the art from the various configurations according to the embodiments of the present invention.

1 is a flowchart schematically illustrating a method for inspecting an object using a 3D pattern kernel according to a first embodiment of the present invention.
2 is a flowchart schematically illustrating a part of an object inspecting method using a 3D pattern kernel according to an example of the present invention.
3 is a flowchart schematically illustrating a method for inspecting an object using a 3D pattern kernel according to another example of the present invention.
4 is a flowchart schematically illustrating a part of an object inspecting method using a 3D pattern kernel according to another example of the present invention.
5 is a block diagram schematically illustrating an object inspection system using a 3D pattern kernel according to a second embodiment of the present invention.
6 illustrates a pattern mask of four first three-dimensional pattern kernels along an object image and up, down, front, and rear directions used in an embodiment of the present invention.
7 illustrates a first image result and a first three-dimensional pattern kernel for calculating first kernel operation data used in an embodiment of the present invention.
8 illustrates a matching score calculated from four first kernel operations data according to an embodiment of the present invention.
9 illustrates a second three-dimensional pattern kernel to be computed with a second image result in one embodiment of the invention.
10 illustrates an error score calculated from second kernel operation data in one embodiment of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a block diagram showing the configuration of a first embodiment of the present invention; Fig. In the description, the same reference numerals denote the same components, and a detailed description may be omitted for the sake of understanding of the present invention to those skilled in the art.

As used herein, unless an element is referred to as being 'direct' in connection, combination, or placement with other elements, it is to be understood that not only are there forms of being 'directly connected, They may also be present in the form of being connected, bonded or disposed. In addition, the same is true when terms including 'contact' such as 'up', 'up', 'lower', 'below' are included. Directional terms may be construed to encompass corresponding relative directional concepts as the reference element is inverted or its direction is changed.

It should be noted that, even though a singular expression is described in this specification, it can be used as a concept representing the entire plurality of constitutions unless it is contrary to, or obviously different from, or inconsistent with the inventive concept. It is to be understood that the description of 'comprising', 'having', 'comprising', 'comprising', etc., in this specification includes the possibility of the presence or addition of one or more other components or combinations thereof.

First, an object inspection method using a 3D pattern kernel according to a first embodiment of the present invention will be described in detail with reference to the accompanying drawings. In this case, reference numerals not described in the accompanying drawings may be reference numerals in other drawings showing the same configuration.

1 is a flowchart schematically illustrating a method for inspecting an object using a 3D pattern kernel according to a first embodiment of the present invention, and FIG. 2 is a diagram of an object inspecting method using a 3D pattern kernel according to a first embodiment of the present invention. 3 is a flowchart schematically showing a part, and FIG. 3 is a flowchart schematically showing a part of an object inspecting method using one 3D pattern kernel according to the first embodiment of the present invention, and FIG. 4 is another example of the present invention. Is a flowchart schematically illustrating a part of an object inspection method using a 3D pattern kernel according to the present invention. FIG. 6 illustrates a pattern mask of four first three-dimensional pattern kernels along an object image and up and down directions, used in an embodiment of the present invention, and FIG. 7 illustrates a first kernel used in an embodiment of the present invention. FIG. 8 illustrates a first image result and a first three-dimensional pattern kernel for calculating arithmetic data. FIG. 8 illustrates a matching score calculated from four first kernel operations data according to an embodiment of the present invention. Shows an inverted kernel to be computed with the second image result in one embodiment of the invention, and FIG. 10 shows an error score calculated from the second kernel operation data in one embodiment of the invention.

1 to 4, an object inspection method using a 3D pattern kernel inspects an object in which a symbol is imprinted on a surface by using an image.

1 and / or 3, the object inspection method using the three-dimensional pattern kernel, the binarization step (S100), direction pattern determination step (S200, S1200), the second kernel operation data calculation step (S300, S1300) and It may include the determination step (S400, S1400).

According to one embodiment, the object inspected by the object inspection method using a three-dimensional pattern kernel may be a tablet.

In the binarization step S100 of FIGS. 1 and / or 3, an image obtained from an object is binarized.

In this case, in one example, in the binarizing operation S100, the acquired image may be adaptive binarized.

Adaptive binarization techniques are already known techniques. The binarization result is used in the calculation with the 3D pattern kernel. For example, the case of the adaptive binarization modified from the k-means clustering algorithm having the adaptive binarization parameters (N, sigma) will be described. Adaptive binarization, which is a modification of the k-means clustering algorithm described below, is exemplary, and various other methods of adaptive binarization may be applied to the present embodiment. In the adaptive binarization parameter, N denotes a size of a block for performing adaptive binarization, and sigma is a value for determining sensitivity of lightness.

In this case, T (x, y) is 'average of pixel values in N by N blocks centered on (x, y)-sigma', and Dst (x, y) is adapted to (x, y) pixel position. If the result of the positive binarization is a pixel value, the adaptive binarization result value Dst (x, y) allocates 255 if src (x, y)> T (x, y), and 0 otherwise. Here, the assignment of the adaptive binarization result to '0' and '255' is an example of assigning the minimum and maximum values of 8-bit grayscale, and assigning the allocation values to '0' and '1'. Do. src (x, y) is an 8-bit grayscale value of (x, y) pixels. At this time, the larger the sigma, which determines the sensitivity of the brightness, is insensitive to the brightness change, while the stronger the noise, and the smaller the sigma, the more sensitive the change in brightness is but the more vulnerable to the noise. For example, the sigma value can be determined by testing the object to be inspected.

For example, according to one example, binarization of the acquired video image may be performed in a large number while varying sensitivity of brightness. For example, as the first binarization and the second binarization, sensitivity of lightness can be varied. For example, when adaptive binarization is performed, a first binarization result and a second binarization result of different sigma may be obtained. In this case, for example, the first binarization result used to determine the direction pattern may have a relatively large sigma, and the second binarization result used to determine whether the object may be relatively smaller after a direction pattern is determined may have a relatively small sigma. have. In the first binarization result for determining the direction pattern and the second binarization result for determining the quality of the object, the binarization parameter N may be the same and may be different according to embodiments.

Next, the direction pattern determination steps S200, S200 ′, and S1200 will be described with reference to FIGS. 1 to 3.

In the direction pattern determination step S200 of FIG. 1, the direction pattern of the object may be determined using the binarization result obtained in the binarization step S100. The reason for determining the direction pattern of the object is that the direction pattern must be matched when comparing or matching the image of the inspected object or its binarization result with the reference image or binarization result of the object to determine whether the object is good or bad. .

At this time, with reference to Figure 2 looks at another example. Referring to FIG. 2, in one example, the direction pattern determination step S200 ′ may include an object tilt determination step S200a and an object direction pattern determination step S200b. That is, in the determining of the direction pattern (S200 ′), the inclination of the object may be determined from the obtained image image or the binarized image, and the direction pattern of the object may be determined using the binarization result whose slope is corrected.

In this case, in the object tilt determination step (S200a), the inclination of the object may be determined from the obtained image image or the binarization image. For example, consider the method of determining the tilt of an object. The example in the following is illustrative and it is also possible to determine the tilt of the object in other ways not illustrated. In one example, when the object is not circular, the boundary may be determined through binarization of the object image, the long direction may be found from the determined boundary, and the inclination of the object may be determined from the inclination of the long direction. This approach is applicable when the object is a tablet. In another example, when the boundary of the object includes a straight line segment, the inclination of the object may be determined from the boundary of the predetermined linear component through binarization of the object image. Alternatively, the inclination of the object may be determined from an image of a symbol imprinted on the object, regardless of the shape of the object. For example, if the long direction can be determined from the image region of the symbol portion among the reference image or the standard image of the object, the long direction of the symbol portion is found from the result of binarization of the image of the symbol portion of the object image, and the inclination of the object from the inclination of the long direction. Can be determined. This method is applicable when the object is a circular drug and the long direction of the imprinted symbol portion can be determined. Alternatively, when the reference line for determining the inclination from the image of the object is imprinted, the inclination of the object may be determined from the binarization result based on the reference line.

The direction pattern determination step S1200 will be further described with reference to FIG. 3.

Referring to FIG. 3, in one example, the direction pattern determination step S1200 may include a first kernel operation data generation step S1210 and a direction pattern determination step S1230.

Specifically, in the first kernel operation data generation step (S1210), as a result of binarization, a first three-dimensional pattern kernel for each direction is calculated according to a predetermined vertical direction of the first binarization result and the reference image. In this case, the first kernel operation data in each direction may be generated according to each operation. Here, the first three-dimensional pattern kernel for each direction is four direction-specific pattern kernels in which the symbol portion of the object is weighted higher than the remaining surface areas. By making the weight of the symbol portion higher than the rest of the surface area, the degree of matching in the symbol portion can be easily known from the kernel operation data obtained by calculating the first binarization result and the first three-dimensional pattern kernel for each direction. Direction patterns can be determined. In addition, at this time, since portions other than the object need not be determined, the weight may be set to '0'. Here, the reference image refers to an image that is a comparison reference. The reference image may be an image generated from a design drawing of an object, or may be an image obtained from an average obtained by inspecting good products of an actual product. Or it may be an average image obtained from the learning results of the objects determined to be good through the present invention.

In one example, the weight of the symbol portion in the first three-dimensional pattern kernel for each direction may be higher than the remaining surface area, which matches the symbol portion of the first binarization result with the symbol portion of the first three-dimensional pattern kernel. It is suitable for the case where the direction is determined. That is, when an object has a symmetrical structure, it is difficult to determine whether the front and rear sides are reversed or upside down using only the boundary obtained from the object image, and thus the direction pattern is determined by matching the symbol portions. At this time, the weight of the surface area excluding the symbol portion may be set very low.

In this case, the first binarization result may be a result of adaptive binarization, and the sensitivity of the brightness may be different from the second binarization result described later. Since the first binarization result is used in the step S1200 of determining the direction pattern, the sensitivity may be lowered. That is, the first binarization result computed with the first three-dimensional pattern kernel is not to check whether the object is damaged or adhered to the foreign matter, but to determine the direction pattern. It can be matched well with the weighted portion of the first three-dimensional pattern kernel. Since the first three-dimensional kernel pattern has a change in a value insensitive to a symbol, for example, a small shift of a letter position or an inclination of an object, a false reading rate may be lowered.

In FIG. 6, a pattern mask is illustrated as four first three-dimensional pattern kernels of M1-0, M1-1, M1-2, and M1-3 along directions of up, down, front, back, and back of the reference object. M1-0 represents the forward direction of the front surface, M1-1 represents the up and down reverse direction of the front surface, M1-2 represents the forward direction of the rear surface, and M1-3 represents the up and down reverse direction of the rear surface. That is, the direction pattern determined in the direction pattern determination step S1200 is one of the direction patterns of M1-0, M1-1, M1-2, and M1-3. In this case, in the first three-dimensional pattern kernel of M1-0, M1-1, M1-2, and M1-3, the weight of the outside of the object is set to '0', and the weight of the symbol portion is the remaining surface area except for the symbol. It is set higher than the weight of. In FIG. 6, the weight of the symbol portion is set to approximately two times higher than the remaining surface area.

FIG. 7A shows the first binarization result to be calculated with the first three-dimensional pattern kernel, and FIG. 7B shows the first three-dimensional pattern kernel M to be calculated with the first binarization result of (a). 7 (c) shows a part of kernel data having a size of 252 x 252 as a data array of the first three-dimensional pattern kernel M. As shown in FIG. In FIG. 7C, the region indicated by the numeral 63 represents a surface region near the symbol 'T', and the weight of the surface region except for the symbol is 63. FIG. The binarization result of (a) of FIG. 7 is a binarization result when the adaptive binarization parameters N = 11 and sigma = 12 in the adaptive binarization technique in which the k-average algorithm is illustrated.

In addition, the first kernel operation data generation step S1210 will be described in more detail. In one example, the first kernel operation data may be calculated from the following equation (1).

식 (1)

Formula (1)

Here, the kernel size is P × Q, I1 (i, j) is the pixel value at the (i, j) coordinate position of the first binarization result, and M1 (k, l) is the value of each of the four direction pattern masks. Is a data value of the position (k, l) in the kernel matrix of the first three-dimensional pattern kernel M1, and B _ij is the first according to each of the operations of the M1 kernel matrices for four directions of s = 0, 1, 2, and 3; It is the data value of (i, j) coordinate position of each kernel operation data. 6, M1-0 is the first three-dimensional pattern kernel when s = 0, M1-1 is s = 1, M1-2 is s = 2, and M1-3 is s = 3 days For each first 3D pattern kernel.

Subsequently, in the direction pattern determination step S1230 of FIG. 3, each matching score is calculated from the first kernel operation data corresponding to each direction, and the direction pattern is determined from the pattern kernel corresponding to the first kernel operation data for which the maximum matching score is calculated. Can be determined.

In FIG. 8, a matching score calculated from the first kernel operation data value for each direction is illustrated as a value of er0 to 3. It can be seen that the first value is 411060, the maximum value, and the pattern kernel corresponding to the maximum matching score is the M1-0 kernel.

Specifically, in one example, the matching score calculated from the first kernel operation data may be calculated by the following equation (2).

식 (2)

Equation (2)

At this time, Mch _s is each matching score calculated from each of the first kernel operation data for each of four directions of s = 0, 1, 2, and 3.

1, 3 and / or 4, the second kernel operation data calculation step (S300, S1300, S2300) will be described.

Referring to FIG. 1, in the second kernel operation data calculating step S300, a binarization result is calculated with a second three-dimensional pattern kernel of a reference image according to a direction pattern determined in the previous direction pattern determination step S200. The second kernel operation data may be calculated from the operation result. In this case, the second three-dimensional pattern kernel has a weight different from that of the symbol portion. The reference image refers to an image that is a comparison standard of object inspection. In addition, as described above, the reference image may be an image generated from a design drawing of an object, or may be an image obtained from an average obtained by inspecting good products of an actual product. The expression "second" in the second three-dimensional pattern kernel is to distinguish it from the "first three-dimensional pattern kernel" in another embodiment and does not have an order meaning.

In one example, the binarization result computed with the second three-dimensional pattern kernel may be an adaptive binarization result. Also, in one example, the binarization result calculated with the second 3D pattern kernel may be a second binarization result, where the second binarization result may have a brightness sensitivity different from that of the first binarization result. For example, the value of the binarization parameter sigma of the second binarization result may be smaller than the sigma value of the first binarization result. By reducing the sigma value of the binarization parameter, it is possible to increase the sensitivity of the lightness, thereby effectively inspecting the surface of the object for damage or foreign matter.

In one example, the weight of the second 3D pattern kernel may cause the remaining area to be higher than the symbol portion. The object inspection mainly checks the surface damage of the object and whether there is foreign matter attached to the surface, and the damage of the imprinted symbol is also linked to the damage of the surface near the boundary with the symbol. By further adding, the accuracy of the inspection can be increased. In this case, since the background parts other than the object need not be determined, the weight may be "0".

For example, referring to FIG. 9, referring to one example, the second three-dimensional pattern kernel has 3 weighted differently in the order of the boundary line of the object> the remaining surface area excluding the symbol> the symbol part. It may be a dimensional pattern kernel. For example, whether or not damage on the boundary portion of the object is an important factor in determining whether the object is good, the weight of the boundary line of the object can be set to the highest.

In addition, referring to FIG. 3, in the second kernel operation data calculation step S1300, the second binarization result as a result of binarization with the second three-dimensional pattern kernel according to the direction pattern determined in the aforementioned direction pattern determination step S1230. May be calculated to calculate second kernel operation data. In the present embodiment, by distinguishing the first binarization result computed from the first three-dimensional pattern kernel and the second binarization result computed from the second three-dimensional pattern kernel, a check portion and a symbol pattern for recognizing a symbol pattern are different. The inspection strength of the inspection portion to determine the foreign matter and / or damage of the can be different, so that more accurate foreign material and / or damage inspection can be performed.

In addition, according to an example, the second kernel operation data calculation step will be described in detail with reference to FIG. 4. In FIG. 4, the second kernel operation data calculation step S2300 may include a second three-dimensional pattern kernel formation step S2310 and a second kernel operation data calculation step S2330.

In the second three-dimensional pattern kernel forming step (S2310) of FIG. 4, the first three-dimensional pattern kernel corresponding to the direction pattern determined in the direction pattern determination step (S1230) of FIG. 3 is inverted to weight the symbol part. Can form a second three-dimensional pattern kernel lower than the 'rest surface area'. Further, in this case, the second 3D pattern kernel to be formed may vary the weight in the order of the boundary line of the object> the remaining surface area excluding the symbol> the symbol part.

FIG. 9A illustrates a first three-dimensional pattern kernel corresponding to the direction pattern determined in the direction pattern determination step S1230 of FIG. 3, and FIG. 9B is a view of FIG. 9A. The second three-dimensional pattern kernel is shown by inverting the first three-dimensional pattern kernel and adjusting the weights in the order of 'object boundary line'> 'symbol remaining surface area'> 'symbol part'. At this time, the background area outside of the object has a weight of "0".

Next, in the calculating of the second kernel operation data of FIG. 4 (S2330), the second kernel is calculated by calculating a second binarization result with a second 3D pattern kernel formed in the second 3D pattern kernel forming step (S2310). Calculation data can be calculated. In this case, the second binarization result may be an adaptive binarization result, and may have a higher sensitivity than the first binarization result. For example, the value of the adaptive binarization parameter sigma is lower than the sigma value of the first binarization result. As a result, the second binarization result has higher sensitivity than the first binarization result, so that it is possible to more precisely determine whether the surface of the object is damaged or adhered to the foreign material.

Specifically, in one example, the second kernel operation data may be calculated from the following equation (3).

식 (3)

Equation (3)

In this case, the kernel size is P × Q, I2 (i, j) is the pixel value of the (i, j) coordinate position of the second binarization result, and M2 (k, l) is the size of the second three-dimensional pattern kernel M2. A data value at position (k, l) in the kernel matrix, and S _ij is a data value at coordinate position (i, j) of the second kernel operation data.

Referring to FIG. 1 and / or 3 again, a description will be given of the determination process (S400, S1400).

Referring to FIGS. 1 and / or 3, in the acceptance determination steps S400 and S1400, an error score may be calculated from the second kernel operation data to determine whether an object is valid.

Specifically, in one example, the error score calculated from the second kernel operation data may be calculated from the following equation (4).

식(4)

Equation (4)

In this case, T_err is an error score calculated from the second kernel operation data S _ij .

In addition, in one example, in determining the good or bad (S400, S1400) it is possible to determine the defect of the surface of the object by normalizing the error score. For example, it is possible to determine whether there is a defect by comparing the normalized error score with a reference score as a criterion for defect determination. In this case, the reference score as a reference for the defect determination may be adjusted according to the sensitivity of the binarization result calculated with the second three-dimensional pattern kernel in the second kernel operation data calculation step (S300, S1300, S2300). For example, when the sensitivity of the binarization result in the second kernel operation data calculation steps S300, S1300, and S2300 is increased, the normalized error score is high in the determination process of this foster care step S400, S1400. It can also be judged high. Alternatively, even if the sensitivity of the binarization result is adjusted, the reference score, which is a criterion for defect determination, may maintain the preset value as it is in the determination process (S400, S1400).

Fig. 10 shows the result of normalizing the error score calculated from the second kernel operation data. In this case, the second three-dimensional pattern kernel shown in FIG. 9 (b) with respect to the direction of the first three-dimensional pattern kernel corresponding to the maximum matching score, that is, the front forward direction, is calculated with the second binarization result. The error score was calculated and normalized. In this case, the second binarization result computed with the second 3D pattern kernel of FIG. It is the binarization result when = 11 and sigma = 8. In Fig. 10, normalization indicates normalization by calculating the maximum value of the degree of object damage or foreign matter adhesion to 100. In Fig. 10, the normalized error score is shown as 25.

Next, the third embodiment will be described before explaining the second embodiment of the present invention.

The computer readable recording medium according to the third embodiment of the present invention is a computer readable recording medium containing a program for performing the object inspection method using the three-dimensional pattern kernel according to any one of the first embodiments described above. In this case, the recording medium may be an optical recording medium such as a CD, a DVD, a Blu-ray, or a memory device such as a ROM or a RAM. For example, it may be a memory device for operating a program stored in a computer.

Next, a fourth embodiment of the present invention will be described. The computer device according to the fourth embodiment is equipped with the computer readable recording medium according to the third embodiment described above, and according to the reading of the computer readable recording medium, the recorded program is performed.

Next, a second embodiment of the present invention will be described in detail with reference to FIG. 5. Referring to the second embodiment of the present invention, reference will be made to the object inspection method using the three-dimensional pattern kernel according to the first embodiment described above, as well as the following Figure 5 and Figures 1 to 4, 6 to 10.

5 is a block diagram schematically illustrating an object inspection system using a 3D pattern kernel according to a second embodiment of the present invention.

An object inspection system using a 3D pattern kernel according to a second embodiment of the present invention is an inspection system for inspecting an object in which a symbol is imprinted on a surface by using an image.

At this point, in one embodiment, the object may be a tablet.

Specifically, referring to FIG. 5, an object inspection system using a 3D pattern kernel according to an embodiment includes a binarization module 100, a kernel generation module 300, and an operation and determination module 500.

First, the binarization module 100 of FIG. 5 binarizes an image obtained from an object. At this time, the binarization of the obtained video image may be performed a large number while varying the sensitivity of the brightness. For example, the first binarization and the second binarization can be performed with different sensitivity of lightness.

In one embodiment, the binarization module 100 may perform adaptive binarization on the obtained image. In this case, a plurality of adaptive binarizations may be performed while varying the sensitivity of the brightness. For example, when adaptive binarization is performed, a first binarization result and a second binarization result of different sigma may be obtained. In this case, for example, the first binarization result used to determine the direction pattern may have a relatively large sigma, and the second binarization result used to determine whether the object may be relatively smaller after a direction pattern is determined may have a relatively small sigma. have.

Also, in one example, the first binarization result is calculated with the first three-dimensional pattern kernel for each direction generated by the kernel generation module 300, and the second binarization result is the second binary formation formed by the kernel generation module 300. Can be computed with a three-dimensional pattern kernel. At this time, the sensitivity of the second binarization result is higher than the first binarization result, and for example, the sigma value, which is an adaptive binarization parameter, is lowered to further determine whether the object surface is damaged or adhered to the object in determination of object acceptance in the calculation and determination module 500. You can make a precise judgment.

In one example, by distinguishing the first binarization result computed from the first three-dimensional pattern kernel and the second binarization result computed from the second three-dimensional pattern kernel, other than the check portion and the symbol pattern for recognizing the symbol pattern The inspection intensity of the inspection portion to determine the foreign matter and / or damage of the can be different, so that more accurate foreign matter and / or damage inspection can be performed.

Next, the kernel generation module 300 of FIG. 5 will be described in detail.

The kernel generation module 500 generates a second 3D pattern kernel of the reference image according to the direction pattern of the object determined by the calculation and determination module 500 to be described later. As described above, the reference image may be an image used as a reference for determining whether the object is good, an image generated from a design drawing of the object, or an image obtained from an average obtained by inspecting good products of an actual product. Or it may be an average image obtained from the learning results of the objects determined to be good through the present invention.

In this case, the second 3D pattern kernel may vary the weight of the symbol portion and the remaining region. For example, in one example, the weight of the second three-dimensional pattern kernel may be higher in the remaining area than in the symbol portion. The object inspection mainly checks the surface of the object for damage or whether foreign matter adheres to the surface, and thus, the weight of the surface area of the object can be given more weight than the symbol portion to increase the accuracy of the inspection. In this case, since the background parts other than the object need not be determined, the weight may be "0".

Referring to FIG. 9, in one example, the second three-dimensional pattern kernel is a three-dimensional pattern kernel weighted differently in the order of the boundary line of the object> the remaining surface area excluding the symbol> the symbol portion Can be.

In addition, according to one example, the kernel generation module may further generate a first three-dimensional pattern kernel for each direction along the top and bottom of the reference image. In this case, the first 3D pattern kernel may be used to determine the direction pattern in the operation and determination module which will be described later. The first three-dimensional pattern kernel may vary the weight of the symbol portion and the rest of the region. In one example, the first three-dimensional pattern kernel is a pattern kernel in which the symbol portion of the object is weighted higher than the remaining surface regions. Can be. By making the weight of the symbol portion higher than the rest of the surface area, the matching in the symbol portion from the kernel operation data that computes the first binarization result and the first three-dimensional pattern kernel for each direction in the operation and determination module 500 to be described later. The degree can be easily grasped and the direction pattern can be easily determined. In addition, at this time, since portions other than the object need not be determined, the weight may be set to '0'. At this time, the weight of the surface area excluding the symbol portion may be set very low.

In FIG. 6, a pattern mask is illustrated as a first three-dimensional pattern kernel for each direction. Shows a three-dimensional pattern kernel. In this case, in the first three-dimensional pattern kernel of M1-0, M1-1, M1-2, and M1-3, the weight of the outside of the object is set to '0', and the weight of the symbol portion is the remaining surface area except for the symbol. It is set higher than the weight of. FIG. 7B illustrates a first three-dimensional pattern kernel M, and FIG. 7C is a data array of the first three-dimensional pattern kernel M, and a part of kernel data having a size of 252 × 252. It is shown. In FIG. 7C, the region indicated by the numeral 63 represents a surface region near the symbol 'T', and the weight of the surface region except for the symbol is 63. FIG.

Also, according to one example, the kernel generation module 300 may invert the first three-dimensional pattern kernel to form a second three-dimensional pattern kernel. In this case, the first 3D pattern kernel is a pattern kernel corresponding to the direction pattern determined by the operation and determination module 500 to be described later. When the second three-dimensional pattern kernel is inverted from the first three-dimensional pattern kernel to form the second three-dimensional pattern kernel, the weight of the 'symbol portion' may be lower than the 'rest surface area'.

Further, the second 3D pattern kernel may vary the weight in the order of the boundary line of the object> the remaining surface area excluding the symbol> the symbol part. For example, FIG. 9A illustrates the first three-dimensional pattern kernel, and FIG. 9B inverts the first three-dimensional pattern kernel of FIG. A second three-dimensional pattern kernel is shown in which weights are adjusted in order of '>' remaining surface areas excluding symbols'>'symbolparts'.

Returning again, the operation and determination module 500 of FIG. 5 will be described. The operation and determination module 500 of FIG. 5 may perform the kernel operation for determining the object direction pattern, calculating the second kernel operation data, and determining the quality of the object. Look at each of them in detail.

First, the determination of the object direction pattern in the calculation and determination module 500 will be described. The calculation and determination module 500 of FIG. 5 may determine the direction pattern of the object by using the binarization result in the binarization module 100.

In addition, in one example, the calculation and determination module 500 may determine the inclination of the object from the obtained video image or the binarized image and determine the direction pattern of the object by using the binarization result whose slope is corrected. For the method of determining the inclination of the object, refer to the description of the above-described first embodiment.

In determining the direction pattern of the object, the calculation and determination module 500 of FIG. 5 uses the first binarization result and the first three-dimensional pattern kernel obtained from the kernel generation module 300 as the binarization result obtained from the binarization module 100. Each operation may generate first kernel operation data in each direction. In this case, the first binarization result may be a result of adaptive binarization, and the sensitivity may be lower than a second binarization result to be calculated with the second three-dimensional pattern kernel.

At this time, in one example, the first kernel operation data may be calculated from the above equation (1). Reference is made to the description of the above-described first embodiment.

Further, in determining the direction pattern of the object, the calculation and determination module 500 calculates each matching score from the first kernel operation data according to each direction, and thus the pattern corresponding to the first kernel operation data for which the maximum matching score is calculated. You can determine the direction pattern of an object from the kernel.

In addition, in one example, the matching score calculated from the first kernel operation data may be calculated from Equation (2) described above. Equation (2) and a description thereof refer to the first embodiment described above. Referring to FIG. 8, a matching score calculated from the first kernel operation data value for each direction is illustrated as a value of er0 to 3, wherein a pattern kernel corresponding to the maximum matching score is determined and the determined first 3 It can be seen that the direction of the dimensional pattern kernel is the direction of the object. The kernel generation module 300 may form a second 3D pattern kernel based on the direction determined by the calculation and determination module 500.

Next, the kernel operation for calculating the second kernel operation data in the operation and determination module 500 will be further described.

The calculation and determination module 500 of FIG. 5 may calculate the binarization result and the second 3D pattern kernel obtained by the binarization module 100. The operation and determination module 500 may calculate the second kernel operation data from the binarization result and the operation result of the second 3D pattern kernel.

In this case, according to an example, the second kernel operation data may be calculated by calculating the second binarization result and the second three-dimensional pattern kernel as the binarization result. In this case, the second binarization result may be an adaptive binarization result, and may have a higher sensitivity than the first binarization result. For example, the value of the adaptive binarization parameter sigma for obtaining the second binarization result is lower than the sigma value of the first binarization result. As a result, the second binarization result has higher sensitivity than the first binarization result, so that it is possible to more precisely determine whether the surface of the object is damaged or adhered to the foreign material.

Specifically, in one example, the second kernel operation data may be calculated from Equation (3) described above. Equation (3) and a description thereof will be referred to the first embodiment described above.

Next, look at the determination of whether the object in the operation and determination module 500.

The operation and determination module 500 of FIG. 5 may determine whether the object is good or not by calculating an error score from the second kernel operation data calculated from the result of the kernel operation. In this example, in one example, the operation and determination module 500 may determine whether the object is good or not by calculating an error score from the second kernel operation data calculated from the operation of the second three-dimensional pattern kernel and the second binarization result. have.

Specifically, in one example, the error score may be calculated from the above equation (4). Equation (4) and a description thereof will be referred to the first embodiment described above.

In addition, in one example, the computation and determination module 500 may normalize the error score to determine a defect on the object surface.

The foregoing embodiments and accompanying drawings are not intended to limit the scope of the present invention but to illustrate the present invention in order to facilitate understanding of the present invention by those skilled in the art. Embodiments in accordance with various combinations of the above-described configurations can also be implemented by those skilled in the art from the foregoing detailed description. Accordingly, various embodiments of the invention may be embodied in various forms without departing from the essential characteristics thereof, and the scope of the invention should be construed in accordance with the invention as set forth in the appended claims. Alternatives, and equivalents by those skilled in the art.

100: binarization module
300: kernel generation module
500: operation and judgment module

Claims (20)

In the method of inspecting an object in which the symbol is imprinted on the surface using an image,
Binarizing an image obtained from the object;
Determining a direction pattern of the object using the binarization result;
Compute the binarization result with the second three-dimensional pattern kernel of the reference image according to the determined direction pattern, wherein the second three-dimensional pattern kernel has a weight difference between the symbol part and the remaining area, and a second kernel operation from the operation result. Calculating data; And
Determining whether the object is good or not by calculating an error score from the second kernel operation data; Object inspection method using a three-dimensional pattern kernel comprising a.
The method according to claim 1,
In the binarizing step, the object inspection method using a three-dimensional pattern kernel, characterized in that for adaptive binarization.
The method according to claim 1,
The determining of the direction pattern may include determining the inclination of the object from the obtained image image or the binarized image and determining the direction pattern of the object by using the binarization result of correcting the inclination. How to check objects using the kernel.
The method according to claim 1,
The determining of the direction pattern may include:
As a result of the binarization, a first three-dimensional pattern kernel for each direction is calculated according to a predetermined vertical direction of the first binarization result and the reference image, and the first three-dimensional pattern kernel for each direction has a surface area in which the symbol portion of the object remains. Generating a first kernel operation data according to each direction according to a calculation, the kernel being four weighted directions kernels which are weighted higher; And
Calculating each matching score from the first kernel operation data corresponding to each direction and determining a direction pattern from a pattern kernel corresponding to the first kernel operation data for which the maximum matching score is calculated; Lt; / RTI >
In the calculating of the second kernel operation data, the second 3D pattern kernel according to the determined direction pattern and the second binarization result are calculated as the binarization result. Way.
The method according to claim 1,
The second three-dimensional pattern kernel is a three-dimensional pattern kernel characterized in that the weight is given differently in the order of the 'boundary line of the object'>'the remaining surface area excluding the symbol'>'symbolpart' Object inspection method used.
The method of claim 4,
Computing the second kernel operation data:
The first three-dimensional pattern kernel corresponding to the determined direction pattern is inverted so that the weight of the 'symbol portion' is lower than the 'rest surface region', so that the boundary line of the object > Forming the second three-dimensional pattern kernel with different weights in the order of 'symbol portion'; And
Calculating the second kernel operation data by calculating the formed second 3D pattern kernel and the second binarization result; Object inspection method using a three-dimensional pattern kernel, characterized in that it comprises a.
The method of claim 4,
The first kernel operation data is an equation,

Is calculated from
The matching score calculated from the first kernel operation data is represented by equation,

Is calculated from
The second kernel operation data is an equation,

Is calculated from
An error score calculated from the second kernel operation data is represented by equation,

Is calculated from
Here, the kernel size is P × Q, I1 (i, j) is the pixel value of the (i, j) coordinate position of the first binarization result, and I2 (i, j) is the value of the second binarization result ( i, j) is a pixel value of the coordinate position, M1 (k, l) is a data value of position (k, l) in the kernel matrix of the first three-dimensional pattern kernel M1 of each of the four direction-specific pattern masks, M2 (k, l) is a data value of position (k, l) in the kernel matrix of the second three-dimensional pattern kernel M2, and B _ij is a M1 kernel for four directions of s = 0, 1, 2, and 3 Is the data value of the (i, j) coordinate position of each of the first kernel operation data according to each operation of the matrix, S _ij is the data value of the (i, j) coordinate position of the second kernel operation data, and Mch _s Is a matching score calculated from each of the first kernel operation data for each of four directions of s = 0, 1, 2, and 3, and T_err is an error score calculated from the second kernel operation data S _ij . The method object and examination with a three dimensional pattern kernel.
The method according to claim 1,
The object inspection method using a three-dimensional pattern kernel, characterized in that in the step of determining whether the defect is normalized by determining the defect of the object surface.
The method according to any one of claims 1 to 8,
The object inspection method using a three-dimensional pattern kernel, characterized in that the object is a tablet.
An inspection system for inspecting an object whose symbol is imprinted on a surface by using an image,
A binarization module for binarizing an image obtained from the object;
The second 3D pattern kernel generates a second 3D pattern kernel of the reference image according to the direction pattern of the object determined by the calculation and determination module, wherein the second 3D pattern kernel has different weights of the symbol portion and the remaining region. Kernel generation module for generating a kernel; And
The direction pattern of the object is determined using a binarization result, and the binarization result and the second 3D pattern kernel are calculated, and second kernel operation data is calculated from the calculation result, and an error score is calculated from the second kernel operation data. An operation and determination module configured to determine whether the object is good or bad; Object inspection system using a three-dimensional pattern kernel comprising a.
The method of claim 10,
The binarization module is an object inspection system using a three-dimensional pattern kernel, characterized in that for performing the adaptive binarization on the obtained image.
The method of claim 10,
The calculation and determination module may determine the inclination of the object from the obtained image image or the binarized image and determine a direction pattern of the object by using the binarization result of correcting the inclination. Object Inspection System.
The method of claim 10,
The kernel generation module further generates a first three-dimensional pattern kernel for each direction according to the top, bottom, front and back of the reference image, wherein the first three-dimensional pattern kernel has a pattern in which the symbol portion of the object is weighted higher than the remaining surface area. Kernel,
The operation and determination module generates a first kernel operation data in each direction by calculating the first binarization result and the first three-dimensional pattern kernel as the binarization result, and generates the first kernel operation data in each direction from the first kernel operation data in each direction. Compute each matching score to determine a direction pattern of the object from the pattern kernel corresponding to the first kernel operation data for which the maximum matching score is calculated, and calculate a second binarization result with the second three-dimensional pattern kernel. The object inspection system using the three-dimensional pattern kernel, characterized in that for calculating the second kernel operation data and the error score to determine whether the object.
The method of claim 10,
The second three-dimensional pattern kernel object weighting system using a three-dimensional pattern kernel, characterized in that the weight is differently assigned in the order of the 'boundary line of the object'>'the remaining surface area excluding the symbol'>'symbolpart'.
The method according to claim 13,
The kernel generation module inverts the first three-dimensional pattern kernel corresponding to the direction pattern determined by the operation and determination module so that the weight of the 'symbol portion' is lower than the 'surface surface region', and thus the boundary line of the object. And a second three-dimensional pattern kernel having a weight different in order of ">" symbol portion remaining surface area "" symbol portion ".
The method according to claim 13,
The first kernel operation data is an equation,

Is calculated from
The matching score calculated from the first kernel operation data is represented by equation,

Is calculated from
The second kernel operation data is an equation,

Is calculated from
An error score calculated from the second kernel operation data is represented by equation,

Is calculated from
Here, the kernel size is P × Q, I1 (i, j) is the pixel value of the (i, j) coordinate position of the first binarization result, and I2 (i, j) is the value of the second binarization result ( i, j) is the pixel value at the coordinate position, M1 (k, l) is the data value at the position (k, l) in the kernel matrix of the first three-dimensional pattern kernel M1 of each of the four direction-specific pattern masks, and M2 (k, l) is the data value of the position (k, l) in the kernel matrix of the second three-dimensional pattern kernel M2, and B _ij is the four direction M1 kernel matrix of s = 0, 1, 2, 3 The data value of the (i, j) coordinate position of each of the first kernel operation data according to each operation, S _ij is the data value of the (i, j) coordinate position of the second kernel operation data, and Mch _s is s = 0 And a matching score calculated from each of the first kernel operation data for each of four directions of 1, 2, and 3, and T_err is an error score calculated from the second kernel operation data S _ij . Used Body inspection system.
The method of claim 10,
The calculation and determination module normalizes the error score to determine a defect of an object surface.
The method according to any one of claims 10 to 17,
The object inspection system using a three-dimensional pattern kernel, characterized in that the object is a tablet.
A computer-readable recording medium,
A computer-readable recording medium containing a program for performing the object inspection method using the three-dimensional pattern kernel according to any one of claims 1 to 8.
In a computer device,
A computer readable recording medium according to claim 19 is mounted,
Wherein the recorded program is performed in accordance with the reading of the recording medium,
Computer device.

Images