JP7360092B2 - Inspection equipment, inspection method, and inspection program - Google Patents

Description

The present invention relates to an inspection device, an inspection method, and an inspection program.

Conventionally, as a learning-type defect detection device, a plurality of partial regions having different sizes are set for a first evaluation region, and for each partial region, an image vector generated from its image information and a corresponding unique The distance to the space is calculated, a first response vector whose elements are distances for each size of each partial region is generated, and this first response vector is calculated by referring to a predetermined discriminant function. It is known that the presence or absence of a defect in a first evaluation area is determined by evaluating (see Patent Document 1). This defect detection device determines the presence or absence of a defect after integrating the evaluation results of the first evaluation area in units of multiple partial areas of different sizes. This suppresses the effects of size variations and increases defect detection accuracy.

JP2010-203845A

By the way, in a method of detecting a defect in an object to be inspected based on the difference between a reference image and an inspection image, the difference may become large near the edge, and defects may be over-detected. That is, even among inspection images taken of non-defective inspection objects, quantization errors such as density (brightness) at the time of imaging, and variations in the inspection object such as position and angle at the time of imaging may occur. For this reason, depending on the determination criteria for the difference, an inspection target object that is originally a good product may be erroneously determined to have a defect, that is, the rate of over-detection of defects may increase.

Therefore, an object of the present invention is to provide an inspection apparatus, an inspection method, and an inspection program that can suppress over-detection of defects.

An inspection device according to one aspect of the present invention is an inspection device that inspects whether or not an object to be inspected has a defect based on an inspection image taken of the object to be inspected, and based on a plurality of learning images, a restored image generation unit that generates a restored image from an inspection image; a difference image generation unit that generates a difference image based on the inspection image and the restored image; and an edge that indicates the direction of change in density gradient in pixels included in the difference image. and a determination unit that determines whether a pixel of the difference image is a defective pixel based on the code and an edge code of the pixel included in the restored image.

According to this aspect, a restored image is generated from a test image based on a plurality of learning images. As a result, compared to the case where a learning image that is a non-defective image is used as a reference image, it is possible to eliminate quantization errors such as density (brightness) at the time of shooting, and the position and angle at the time of shooting, which may occur even in a non-defective image. A restored image RG close to the inspection image in which variations in the inspection object are reduced can be obtained as a reference image. Furthermore, it is determined whether or not a pixel in the differential image is a defective pixel based on the edge code of the pixel included in the differential image and the edge code of the pixel included in the restored image. Thereby, it is possible to separate the original edge pixels of the object to be inspected from the edge pixels of defects that may be included in the object to be inspected, and it is possible to determine that the original edge pixels of the object to be inspected are not defective pixels. Therefore, over-detection of defects can be suppressed.

In the above-described aspect, the difference image further includes an extraction unit that extracts a pixel whose density gradient is equal to or greater than a predetermined value as a pixel of interest, and the determination unit includes an edge code of the pixel of interest and an edge code of the pixel included in the restored image. It may be determined whether the pixel of interest is a defective pixel based on the edge code.

According to this aspect, a pixel whose density gradient is equal to or greater than a predetermined value is extracted as a pixel of interest. Here, in the edge portion of the image that is the original boundary of the object to be inspected and the edge portion of a defect that may be included in the object to be inspected, the pixels constituting these edge portions have a density gradient of a predetermined magnitude. It is considered to be greater than the value. Thereby, pixels forming the edge portion can be extracted. Therefore, the number of pixels to be inspected can be reduced, and inspection time can be shortened.

The above-described aspect further includes a setting section that sets a mask of a predetermined size centered on the coordinates of a pixel corresponding to the pixel of interest in the restored image, and the determination section includes an edge code of the pixel of interest and a pixel within the mask. It may be determined whether the pixel of interest is a defective pixel based on the edge code of the pixel of interest.

According to this aspect, it is determined whether the pixel of interest is a defective pixel based on the edge code of the pixel of interest and the edge code of the pixel within the mask. Thereby, even if there are variations in the object to be inspected, for example, when an edge code corresponding to the edge code of the pixel of interest is extracted from the mask, it is possible to determine that the pixel of interest is not a defective pixel. Therefore, detection of defective pixels due to variations in the inspection object can be suppressed.

In the above-mentioned aspect, the difference image further includes a setting unit that sets a mask of a predetermined size centered on the coordinates of a pixel corresponding to a predetermined pixel included in the restored image, and the determination unit Based on the edge code and the edge code of the pixel within the mask, it may be determined whether the corresponding pixel is a defective pixel.

According to this aspect, it is determined whether the pixel corresponding to the predetermined pixel is a defective pixel based on the edge code of the predetermined pixel included in the restored image and the edge code of the pixel within the mask. . Thereby, even if there are variations in the object to be inspected, for example, when an edge code corresponding to the edge code of a predetermined pixel is extracted from the mask, it can be determined that the corresponding pixel is not a defective pixel. Therefore, detection of defective pixels due to variations in the inspection object can be suppressed.

In the aspect described above, in the restored image, a first mask of a predetermined size centered on the coordinates of a corresponding pixel corresponding to a predetermined pixel included in the difference image is set, and a pixel corresponding to the corresponding pixel in the difference image is set. The determination unit further includes a setting unit configured to set a second mask of a predetermined size centered on the coordinates of Then, it may be determined whether a predetermined pixel is a defective pixel.

According to this aspect, it is determined whether a predetermined pixel included in the difference image is a defective pixel based on the edge code of the pixel in the first mask and the edge code of the pixel in the second mask. be done. As a result, even if there are variations in the inspection target, for example, if the edge code corresponding to the edge code of a predetermined pixel included in the difference image is extracted from the first mask, or if the edge code corresponds to the edge code of the corresponding pixel, If the edge code is extracted from the second mask, it can be determined that the predetermined pixel is not a defective pixel. Therefore, detection of defective pixels due to variations in the inspection object can be suppressed.

In the aspect described above, the restored image includes a plurality of regions, and the restored image generation unit may generate the restored image from the inspection image for each region based on the plurality of learning images.

According to this aspect, the restored image includes a plurality of regions, and the restored image is generated from the test image for each region based on the plurality of learning images. Thereby, quantization errors and variations in the inspection object can be reduced for each region, and a restored image that is even closer to the inspection image can be obtained as a reference image.

Further, an inspection method according to another aspect of the present invention is an inspection method for inspecting whether or not an object to be inspected has a defect based on an inspection image taken of the object to be inspected. a restored image generation step of generating a restored image from the inspection image based on the inspection image; a difference image generation step of generating a difference image based on the inspection image and the restored image; and a change direction of the density gradient in pixels included in the difference image. and a determination step of determining whether a pixel of the difference image is a defective pixel based on an edge code indicating the pixel included in the restored image.

According to this aspect, a restored image is generated from a test image based on a plurality of learning images. As a result, compared to the case where a learning image that is a non-defective image is used as a reference image, it is possible to eliminate quantization errors such as density (brightness) at the time of shooting, and the position and angle at the time of shooting, which may occur even in a non-defective image. A restored image RG close to the inspection image in which variations in the inspection object are reduced can be obtained as a reference image. Furthermore, it is determined whether or not a pixel in the differential image is a defective pixel based on the edge code of the pixel included in the differential image and the edge code of the pixel included in the restored image. Thereby, it is possible to separate the original edge pixels of the object to be inspected from the edge pixels of defects that may be included in the object to be inspected, and it is possible to determine that the original edge pixels of the object to be inspected are not defective pixels. Therefore, over-detection of defects can be suppressed.

The above-described aspect further includes an extraction step of extracting a pixel whose density gradient is equal to or greater than a predetermined value in the difference image as a pixel of interest, and the determination step includes an edge code of the pixel of interest and a pixel of the pixel included in the restored image. The method may also include determining whether the pixel of interest is a defective pixel based on the edge code.

According to this aspect, a pixel whose density gradient is equal to or greater than a predetermined value is extracted as a pixel of interest. Here, in the edge portion of the image that is the original boundary of the object to be inspected and the edge portion of a defect that may be included in the object to be inspected, the pixels constituting these edge portions have a density gradient of a predetermined magnitude. It is considered to be greater than the value. Thereby, pixels forming the edge portion can be extracted. Therefore, the number of pixels to be inspected can be reduced, and inspection time can be shortened.

The above-described aspect further includes a setting step of setting a mask of a predetermined size centered on the coordinates of a pixel corresponding to the pixel of interest in the restored image, and the determining step includes determining the edge code of the pixel of interest and the pixels within the mask. The method may include determining whether the pixel of interest is a defective pixel based on the edge code of the pixel of interest.

According to this aspect, it is determined whether the pixel of interest is a defective pixel based on the edge code of the pixel of interest and the edge code of the pixel within the mask. Thereby, even if there are variations in the object to be inspected, for example, when an edge code corresponding to the edge code of the pixel of interest is extracted from the mask, it is possible to determine that the pixel of interest is not a defective pixel. Therefore, detection of defective pixels due to variations in the inspection object can be suppressed.

In the above aspect, the difference image further includes a setting step of setting a mask of a predetermined size centered on the coordinates of a pixel corresponding to a predetermined pixel included in the restored image, and the determining step includes The method may include determining whether the corresponding pixel is a defective pixel based on the edge code and the edge code of the pixel within the mask.

According to this aspect, it is determined whether the pixel corresponding to the predetermined pixel is a defective pixel based on the edge code of the predetermined pixel included in the restored image and the edge code of the pixel within the mask. . Thereby, even if there are variations in the object to be inspected, for example, when an edge code corresponding to the edge code of a predetermined pixel is extracted from the mask, it can be determined that the corresponding pixel is not a defective pixel. Therefore, detection of defective pixels due to variations in the inspection object can be suppressed.

In the aspect described above, in the restored image, a first mask of a predetermined size centered on the coordinates of a corresponding pixel corresponding to a predetermined pixel included in the difference image is set, and in the difference image, a first mask of a predetermined size is set, and It further includes a setting step of setting a second mask of a predetermined size centered on the coordinates, and the determining step is based on the edge code of the pixel in the first mask and the edge code of the pixel in the second mask. , may include determining whether a predetermined pixel is a defective pixel.

According to this aspect, it is determined whether a predetermined pixel included in the difference image is a defective pixel based on the edge code of the pixel in the first mask and the edge code of the pixel in the second mask. be done. As a result, even if there are variations in the inspection target, for example, if the edge code corresponding to the edge code of a predetermined pixel included in the difference image is extracted from the first mask, or if the edge code corresponds to the edge code of the corresponding pixel, If the edge code is extracted from the second mask, it can be determined that the predetermined pixel is not a defective pixel. Therefore, detection of defective pixels due to variations in the inspection object can be suppressed.

In the aspect described above, the restored image includes a plurality of regions, and the restored image generation step may include generating a restored image from the test image for each region based on the plurality of learning images.

According to this aspect, the restored image includes a plurality of regions, and the restored image is generated from the test image for each region based on the plurality of learning images. Thereby, quantization errors and variations in the inspection object can be reduced for each region, and a restored image that is even closer to the inspection image can be obtained as a reference image.

Further, an inspection program according to another aspect of the present invention is an inspection program that is caused to be executed by a computer to inspect whether or not an object to be inspected has a defect based on an inspection image taken of the object to be inspected. a restored image generation step of generating a restored image from a test image based on a learning image of the test image; a difference image generation step of generating a difference image based on the test image and the restored image; The method includes a determination step of determining whether or not a pixel of the difference image is a defective pixel based on an edge code indicating a direction of gradient change and an edge code of a pixel included in the restored image.

According to this aspect, a restored image is generated from a test image based on a plurality of learning images. As a result, compared to the case where a learning image that is a non-defective image is used as a reference image, it is possible to eliminate quantization errors such as density (brightness) at the time of shooting, and the position and angle at the time of shooting, which may occur even in a non-defective image. A restored image RG close to the inspection image in which variations in the inspection object are reduced can be obtained as a reference image. Furthermore, it is determined whether or not a pixel in the differential image is a defective pixel based on the edge code of the pixel included in the differential image and the edge code of the pixel included in the restored image. Thereby, it is possible to separate the original edge pixels of the object to be inspected from the edge pixels of defects that may be included in the object to be inspected, and it is possible to determine that the original edge pixels of the object to be inspected are not defective pixels. Therefore, over-detection of defects can be suppressed.

According to the present invention, over-detection of defects can be suppressed.

FIG. 1 is a configuration diagram illustrating a schematic configuration of an inspection system according to an embodiment of the present invention. FIG. 2 is a flowchart illustrating a general operation of generating restoration parameters. FIG. 3 is a diagram illustrating divided regions of a learning image. FIG. 4 is a diagram illustrating an edge code. FIG. 5 is a flowchart illustrating a schematic operation of the inspection apparatus according to an embodiment of the present invention. FIG. 6 is a flowchart illustrating the restored image generation process shown in FIG. FIG. 7 is a flowchart illustrating the difference image generation process shown in FIG. 5. FIG. 8 is a conceptual diagram illustrating the concept of an example of the defective pixel determination process shown in FIG. 5. FIG. 9 is a flowchart illustrating an example of the defective pixel determination process shown in FIG. FIG. 10 is a flowchart illustrating another example of the defective pixel determination process shown in FIG. FIG. 11 is a conceptual diagram illustrating the concept of a first modification of the defective pixel determination process shown in FIG. 5. FIG. 12 is a conceptual diagram illustrating the concept of a second modification of the defective pixel determination process shown in FIG. 5.

Preferred embodiments of the present invention will be described with reference to the accompanying drawings. In addition, in each figure, those with the same reference numerals have the same or similar configurations.

First, an example of the configuration of an inspection apparatus according to an embodiment of the present invention will be described with reference to FIGS. 1 to 3. FIG. 1 is a configuration diagram illustrating a schematic configuration of an inspection system 100 according to an embodiment of the present invention. FIG. 2 is a flowchart illustrating a general operation of generating restoration parameters. FIG. 3 is a diagram illustrating divided regions of a learning image.

As shown in FIG. 1, the inspection system 100 includes an inspection device 50, an imaging device 60, an input device 70, and an output device 80. The inspection system 100 is configured, for example, to photograph an object to be inspected using an imaging device 60 and to inspect an image of the object to be inspected (hereinafter referred to as an "inspection image") using an inspection device 50.

The inspection device 50 is for inspecting whether or not the object to be inspected has a defect based on an inspection image taken of the object to be inspected. Defects include, but are not limited to, visible scratches, cracks, chips, burrs, deposits, foreign matter, dents, uneven color, dirt, blurred print, misaligned print, etc. including things.

The inspection device 50 is, for example, an information processing device such as a computer. The inspection device 50 includes an input/output I/F (interface) 10, a storage section 20, and a control section 30. In addition, the inspection device 50 further includes a bus 40 configured to transmit signals and data between each section of the inspection device 50.

The input/output I/F 10 is an interface between the inspection device 50 and external equipment. The input/output I/F 10 is configured to exchange data and signals with external equipment. Further, the input/output I/F 10 is configured to control communication with external equipment. The input/output I/F 10 includes, for example, connection terminals of standardized interfaces such as USB (Universal Serial Bus), HDMI (registered trademark) (High-Definition Multimedia Interface), and IEEE1394. The input/output I/F 10 is connected to an imaging device 60, an input device 70, and an output device 80.

The storage unit 20 is configured to store programs, data, and the like. The storage unit 20 includes, for example, a hard disk drive, a solid state drive, and the like. The storage unit 20 stores in advance various programs executed by the control unit 30 and data necessary for executing the programs. The storage unit 20 also stores the inspection image 21 input from the imaging device 60 via the input/output I/F 10. Furthermore, the storage unit 20 stores in advance a restoration parameter file 22 for generating a restored image, which will be described later.

The control section 30 is configured to control the operation of each section of the inspection device 50, such as the input/output I/F 10 and the storage section 20. Further, the control unit 30 is configured to implement each function described below by executing a program stored in the storage unit 20 or the like. The control unit 30 includes, for example, a processor such as a CPU (Central Processing Unit), a memory such as a ROM (Read Only Memory), a RAM (Random Access Memory), and a buffer storage device such as a buffer. The control unit 30 includes, as its functional configuration, a restored image generation unit 31, a difference image generation unit 32, an extraction unit 33, a setting unit 34, and a determination unit 35, for example.

The restored image generation unit 31 is configured to generate a restored image from the inspection image 21 based on a plurality of learning images. Note that details of the restored image generation section 31 will be described later.

The difference image generation unit 32 is configured to generate a difference image based on the inspection image and the restored image. Note that details of the difference image generation section 32 will be described later.

The extraction unit 33 is configured to extract, as a pixel of interest, a pixel with a density gradient greater than or equal to a predetermined value in the difference image.

Here, in the edge part that is the original boundary of the inspection object in the image and the edge part of the defect that may be included in the inspection object, the pixels that make up these edge parts (hereinafter also referred to as "edge pixels") ), it is considered that the magnitude of the concentration gradient is greater than or equal to a predetermined value.

Therefore, edge pixels can be extracted by extracting pixels whose density gradient is equal to or greater than a predetermined value as pixels of interest. Therefore, the number of pixels to be inspected can be reduced, and inspection time can be shortened.

The setting unit 34 is configured to set a mask of a predetermined size centered on the coordinates of a predetermined pixel in the image. The setting unit 34 may set a mask for the restored image or may set a mask for the difference image. Alternatively, the setting unit 34 may set masks for both the restored image and the difference image.

The determination unit 35 determines whether a pixel in the differential image is a defective pixel based on an edge code indicating the direction of change in density gradient in the pixel included in the differential image and an edge code of the pixel included in the restored image. is configured to make a determination. More specifically, for example, the determining unit 35 is configured to determine whether the pixel of interest is a defective pixel based on the edge code of the pixel of interest and the edge code of the pixel included in the restored image. ing. Further, it may be configured to determine whether the pixel of interest is a defective pixel based on the edge code of the pixel of interest and the edge code of the pixel within the mask.

Each function of the control unit 30 can be realized by a program executed by a computer (microprocessor). Therefore, each function provided in the control unit 30 can be realized by hardware, software, or a combination of hardware and software, and is not limited to either case.

Further, when each function of the control unit 30 is realized by software or a combination of hardware and software, the processing can be executed by multitasking, multithreading, or both multitasking and multithreading. It is not limited to this case.

The imaging device 60 is configured to capture images and record them as data. The imaging device 60 is a digital camera, and includes, for example, optical components such as a lens, and electronic components such as an imaging element (light receiving element). Note that the optical system component may include a plurality of lenses. Further, the electronic component may include a light emitting device such as a flash. The imaging device 60 photographs the object to be inspected and outputs the photographed image to the inspection device 50 via the input/output I/F 10. The control unit 30 performs necessary processing on the image input from the imaging device 60 to generate a file of the inspection image 21, and stores the generated file of the inspection image 21 in the storage unit 20.

The input device 70 is configured so that information can be input by a user's operation. The input device 70 includes, for example, a keyboard, keypad, mouse, trackball, touch panel, microphone, and the like. It is possible to do so. For example, when a user operates a keyboard, keypad, mouse, trackball, touch panel, microphone, etc. (including voice operations using a microphone), the input device 70 outputs data or signals corresponding to the operation. It is output to the inspection device 50 via the input/output I/F 10. The control unit 30 generates data based on this data or signal, thereby making it possible to input information to the inspection device 50.

Output device 80 is configured to output information. The output device 80 includes, for example, a display device such as a liquid crystal display, an EL (Electro Luminescence) display, or a plasma display. For example, the output device 80 displays image data input from the inspection device 50 via the input/output I/F 10 on a display device, thereby making it possible to output information.

Here, a method for generating the restoration parameter file 22 from a plurality of learning images will be described. The restored image generation unit 31 is configured to generate a restored image from the inspection image 21 based on the generated restoration parameter file 22. As shown in FIG. 2, in the restoration parameter generation process S200, first, for example, the control unit 30 reads one of a plurality of learning images (S201). Each learning image is an image of a non-defective inspected object with no defects, that is, a non-defective image.

Next, the control unit 30 divides the learning image read in step S201 into a plurality of predetermined regions (S202).

The number of regions into which the learning image is divided is determined in advance, and the vertical and horizontal lengths (number of pixels) of each region are also determined in advance. For example, as shown in FIG. 3, the learning image LG is divided into four regions, and each region has the same length (number of pixels) in the X-axis direction and the same length (number of pixels) in the Y-axis direction. are set the same. However, the length in the X-axis direction (number of pixels) and the length in the Y-axis direction (number of pixels) may be the same or different. Each area is associated with an area number i, and for example, area numbers i from "1" to "4" are assigned counterclockwise from the upper left area of the learning image LG.

As mentioned above, although the learning image LG is a non-defective image, there are quantization errors such as density (brightness) at the time of photographing, and the inspection object such as the position and angle at the time of photographing, which can occur even in non-defective images. This includes variations in

Returning to FIG. 2, next, the control unit 30 sets the region number i to "1" (S203), and calculates the restoration parameter for the region with the region number i among the plurality of regions divided in step S203. (S204).

The restoration parameters calculated in step S204 depend on the algorithm of the machine learning model used when generating the restored image. For example, when using an autoencoder that performs reversible dimensional compression as an algorithm for a machine learning model, the control unit 30 calculates, for example, weights between neural layers as restoration parameters. Note that calculation of the restoration parameters is not limited to the case of using Autoencoder, and other machine learning model algorithms other than Autoencoder may be used.

Next, the control unit 30 writes the restoration parameters calculated in step S204 to the restoration parameter file 22, which is created for each area, for example, and stored in the storage unit 20 (S205). As a result, the restoration parameters are added to the restoration parameter file 22 for each region of the learning image LG.

Next, the control unit 30 adds "1" to the area number i (S206), and determines whether the area number i is larger than the maximum value of the area number i (S207). The maximum value of the region number i is the number of regions when the learning image LG is divided in step S202, and is "4" in the example shown in FIG. 3.

As a result of the determination in step S207, if the area number i is not greater than the maximum value of the area number i, that is, if the area number i is less than or equal to the maximum value, the process returns to step S204, and the control unit 30 determines that the area number i is the maximum value. The processes from step S204 to step S207 are repeated until the size becomes larger. Thereby, restoration parameters can be determined for all regions of the learning image LG.

On the other hand, if the result of the determination in step S207 is that the area number i is larger than the maximum value, the control unit 30 determines whether there is no next learning image LG (S208).

As a result of the determination in step S208, if there is a next learning image LG, the process returns to step S201, and the control unit 30 repeats the processes from step S201 to step S208 until there is no next learning image LG. Thereby, restoration parameters can be obtained for all regions of all learning images LG.

On the other hand, as a result of the determination in step S208, if there is no next learning image LG, the control unit 30 ends the restoration parameter generation process S200.

In this embodiment, an example was shown in which the control unit 30 of the inspection device 50 executes the restoration parameter generation process S200, but the present invention is not limited to this. For example, the restoration parameter generation process S200 may be executed by another device. In this case, the restoration parameter file 22 is transferred from the other device to the inspection device 50.

Next, with reference to FIG. 4, a method for determining the magnitude of the concentration gradient (hereinafter referred to as "edge strength") and the direction of change in the concentration gradient (hereinafter referred to as "edge code") will be described. FIG. 4 is a diagram illustrating an edge code.

For example, for a pixel E at coordinate position (x, y), the amount of change in density Ex(x, y) in the horizontal direction (X-axis direction) and the amount of change in density Ey(x, y) in the vertical direction (Y-axis direction) Find y) and. Then, the vector length IE (x, y) is calculated for the composite vector of the vectors indicated by Ex (x, y) and Ey (x, y). This IE(x,y) is the edge strength and is expressed as in the following equation (1).
IE (x, y) = [{Ex (x, y)} ² + {Ey (x, y)} ² ] ^1/2 … (1)

Further, the direction indicated by the above-mentioned composite vector corresponds to the direction of the density gradient in the pixel E. As shown in FIG. 4, for a pixel E, a vector C is set that is perpendicular to a vector F indicating the density gradient direction, and an angle indicating the direction of this vector C is defined as an edge code EC(x, y).

Note that the vector F is a direction from the bright side to the dark side, and the vector C corresponds to the direction obtained by rotating the vector F by 90 degrees clockwise. Furthermore, the edge code EC(x, y) is expressed based on the vector B going from the pixel E in the positive direction of the x-axis, and is expressed according to the values of Ex(x, y) and Ey(x, y). It is determined by one of the following equations (a) to (e).
(a) When Ex(x,y)>0 and Ey(x,y)≧0,
EC(x,y)=atan(Ey(x,y)/Ex(x,y))
(b) When Ex(x,y)>0 and Ey(x,y)<0,
EC(x,y)=360+atan(Ey(x,y)/Ex(x,y))
(c) When Ex(x,y)<0,
EC(x,y)=180+atan(Ey(x,y)/Ex(x,y))
(d) When Ex (x, y) = 0 and Ey (x, y) > 0,
EC(x,y)=0
(e) When Ex(x,y)=0 and Ey(x,y)<0,
EC(x,y)=180

In this embodiment, an example is shown in which the angle indicating the direction of the vector C is used as the edge code in FIG. 4, but the present invention is not limited to this. The edge code may be anything that indicates the direction of change in the density gradient. For example, the edge code may be an angle that indicates the direction of the vector F that indicates the direction of the density gradient.

Next, an example of the operation of the inspection apparatus according to an embodiment of the present invention will be described with reference to FIGS. 5 to 12. FIG. 5 is a flowchart illustrating a schematic operation of the inspection device 50 according to an embodiment of the present invention. FIG. 6 is a flowchart illustrating the restored image generation process S320 shown in FIG. FIG. 7 is a flowchart illustrating the difference image generation process S340 shown in FIG. FIG. 8 is a conceptual diagram illustrating an example of the concept of the defective pixel determination process S360 shown in FIG. FIG. 9 is a flowchart illustrating an example of the defective pixel determination process S360 shown in FIG. FIG. 10 is a flowchart illustrating another example of the defective pixel determination process S360 shown in FIG. FIG. 11 is a conceptual diagram illustrating the concept of a first modification of the defective pixel determination process S360 shown in FIG. FIG. 12 is a conceptual diagram illustrating the concept of a second modification of the defective pixel determination process S360 shown in FIG.

When activated by a user's operation, for example, the control unit 30 of the inspection device 50 executes inspection processing S300 shown in FIG. 5 .

<Inspection processing>
First, the restored image generation unit 31 reads the inspection image 21 from the storage unit 20 (S301). The test image 21 has the same vertical and horizontal lengths (number of pixels) as the learning image LG described above.

Next, the restored image generation unit 31 executes restored image generation processing S320. In the restored image generation process S320, the restored image generation unit 31 generates a restored image from the inspection image 21.

Next, the difference image generation unit 32 executes difference image generation processing S340. In difference image generation processing S340, the difference image generation unit 32 generates a difference image based on the inspection image 21 and the restored image.

Next, the extraction unit 33 sets the minimum value y1 to the Y coordinate (S302), and sets the minimum value x1 to the X coordinate (S303). The pixel at the coordinate position (x1, y1) whose X coordinate is the minimum value x1 and whose Y coordinate is the minimum value y1 is, for example, a pixel at the upper left corner of the image. Further, the pixel at the coordinate position (x2, y2) whose X coordinate is the maximum value x2 and whose Y coordinate is the maximum value y2 is, for example, a pixel at the lower right apex of the image.

Next, the extraction unit 33 calculates the edge intensity IE(x, y) of the pixel at the coordinate position (x, y) in the difference image generated in the difference image generation process S340, and calculates the edge intensity IE(x, y) of the pixel at the coordinate position (x, y). ) and a predetermined threshold value to determine whether the edge strength IE(x, y) is greater than or equal to the predetermined threshold value (S304).

As a result of the determination in step S304, if the edge strength IE (x, y) is greater than or equal to the predetermined threshold, the pixel at the coordinate position (x, y) in the difference image is extracted as the pixel of interest. Therefore, the setting unit 34 and the determining unit 35 execute defective pixel determining processing S360. In the defective pixel determination process S360, the determination unit 35 determines whether the pixel of interest in the difference image is a defective pixel.

As a result of the determination in step S304, if the edge intensity IE(x,y) is not greater than or equal to the predetermined threshold, that is, if it is less than the predetermined threshold, defective pixel determination processing S360 is not executed.

Next, the extraction unit 33 adds "1" to the X coordinate (S305), and determines whether the X coordinate is larger than the maximum value x2 of the X coordinate (S306).

As a result of the determination in step S306, if the X coordinate is not greater than the maximum value x2 of the X coordinate, that is, if the The processes from step S304 to step S306 are repeated until the value becomes larger.

On the other hand, as a result of the determination in step S306, if the X coordinate is larger than the maximum value x2 of the X coordinate, the extraction unit 33 adds "1" to the Y coordinate (S307), and the It is determined whether or not it is larger (S308).

As a result of the determination in step S306, if the Y coordinate is not greater than the maximum value y2 of the Y coordinate, that is, if the Y coordinate is less than or equal to the maximum value y2, the process returns to step S303, and the extraction unit 33 determines that the X coordinate is less than the maximum value x2. The processes from step S303 to step S308 are repeated until the value becomes larger.

On the other hand, as a result of the determination in step S308, if the Y coordinate is larger than the maximum value x2 of the X coordinate, the control unit 30 executes defect determination processing S309. In defect determination processing S309, the control unit 30 determines that the inspection target object has a defect, for example, when the difference image includes at least one defective pixel. Alternatively, if the total area of defective pixels in the differential image is greater than or equal to a predetermined value, or if the total number of defective pixels in the differential image is greater than or equal to a predetermined value, or if the total area of defective pixels in the differential image is It may be determined whether or not there is a defect in the object to be inspected by integrating the objects into one group and comparing the center of gravity, area, etc. of the group with a predetermined threshold.

After the defect determination process S309, the control unit 30 returns to step S301 and repeats the processes from step S301 to step S309 until, for example, the inspection apparatus 50 stops.

<Restored image generation process>
In the restored image generation process S320 shown in FIG. 6, the restored image generation unit 31 first divides the inspection image 21 read in step S301 into a plurality of predetermined regions (S321). The number of regions into which the test image 21 is divided is set to be the same as the learning image LG, and the vertical and horizontal lengths (number of pixels) of each region are also set to be the same as the learning image LG. Each area of the test image 21 is associated with an area number i, like the area of the learning image LG. For example, as in the example shown in FIG. Area number i of "4" is assigned.

Next, the restored image generation unit 31 sets the region number i to "1" (S322), and applies the region number i to the restoration parameter file 22 of the region number i among the plurality of regions divided in step S321. The image is restored using (S323).

Next, the restored image generation unit 31 adds "1" to the area number i (S324), and determines whether the area number i is larger than the maximum value of the area number i (S325). The maximum value of the area number i is the number of areas when the inspection image 21 is divided in step S321, and is, for example, "4".

As a result of the determination in step S325, if the area number i is not greater than the maximum value of the area number i, that is, if the area number i is less than or equal to the maximum value, the process returns to step S323, and the restored image generation unit 31 determines that the area number i is The processes from step S323 to step S325 are repeated until the value becomes larger than the maximum value. Thereby, the image can be restored for all areas of the restored image.

On the other hand, if the region number i is larger than the maximum value as a result of the determination in step S325, the restored image generation unit 31 generates a restored image by pasting together the regions restored in step S323 (S326).

After step S326, the restored image generation unit 31 ends the restored image generation process S320.

In this way, a restored image is generated from the inspection image 21 based on the plurality of learning images LG. As a result, compared to the case where the learning image LG, which is a non-defective image, is used as a reference image, it is possible to reduce quantization errors such as density (brightness) at the time of shooting, position and angle at the time of shooting, etc., which may occur even in a non-defective image. A restored image close to the inspection image 21 in which variations in the inspection object are reduced can be obtained as a reference image.

Further, the restored image includes a plurality of regions, and the restored image is generated from the inspection image 21 for each region based on the plurality of learning images LG. Thereby, quantization errors and variations in the inspection object can be reduced for each region, and a restored image closer to the inspection image 21 can be obtained as a reference image.

<Difference image generation processing>
In the difference image generation process S340 shown in FIG. 7, the difference image generation unit 32 first sets the minimum value x1 in the X coordinate (S341), and sets the minimum value y1 in the Y coordinate (S342). The pixel at the coordinate position (x1, y1) where the X coordinate is the minimum value x1 and the Y coordinate is the minimum value y1 is, for example, a pixel at the upper left corner of the image. Further, a pixel at a coordinate position (x2, y2) whose X coordinate has a maximum value x2 and whose Y coordinate has a maximum value y2, which will be described later, is, for example, a pixel at the lower right apex of the image.

Next, the difference image generation unit 32 calculates the density difference between the pixel at the coordinate position (x, y) in the inspection image 21 and the pixel at the coordinate position (x, y) in the restored image, that is, the corresponding pixel (S343 ). For example, if the density of a pixel in the inspection image 21 is ID (x, y) and the density of the corresponding pixel in the restored image is RD (x, y), then the density difference DD (x, y) can be calculated using the following equation (2). It is expressed as follows.
DD (x, y) = | ID (x, y) - RD (x, y) | ... (2)

Next, the difference image generation unit 32 compares the density difference DD(x, y) calculated in step S343 with a predetermined threshold, and determines that the density difference DD(x, y) is equal to or greater than the threshold. It is determined whether or not (S344).

As a result of the determination in step S344, if the density difference DD(x, y) is greater than or equal to the predetermined threshold, the difference image generation unit 32 generates the density difference DD(x, y) calculated in step S343. The density is set to the pixel at the coordinate position (x, y) in the difference image, that is, the density of the corresponding pixel (S345).

On the other hand, as a result of the determination in step S344, if the density difference DD(x, y) is not greater than or equal to the predetermined threshold value, that is, if it is less than the threshold value, the difference image generation unit 32 generates a minimum value, for example, "0". is set to the density of the corresponding pixel in the difference image to be generated (S346).

After step S345 or step S346, the difference image generation unit 32 adds "1" to the X coordinate (S347), and determines whether the X coordinate is larger than the maximum value x2 of the X coordinate (S348).

As a result of the determination in step S348, if the X coordinate is not larger than the maximum value x2 of the X coordinate, that is, if the X coordinate is less than or equal to the maximum value The processes from step S343 to step S348 are repeated until the value becomes larger than x2.

On the other hand, as a result of the determination in step S348, if the X coordinate is larger than the maximum value x2 of the X coordinate, the difference image generation unit 32 adds "1" to the Y coordinate (S349), and the Y coordinate It is determined whether it is larger than y2 (S350).

As a result of the determination in step S350, if the Y coordinate is not larger than the maximum value y2 of the Y coordinate, that is, if the Y coordinate is less than or equal to the maximum value y2, the process returns to step S342, and the difference image generation unit 32 determines that the X coordinate is the maximum value y2. The processes from step S342 to step S350 are repeated until the value becomes larger than x2.

On the other hand, as a result of the determination in step S350, if the Y coordinate is larger than the maximum value x2 of the X coordinate, the difference image generation unit 32 ends the difference image generation process S340.

Here, the concept of an example of defective pixel determination processing will be explained. As shown in FIG. 8, a pixel PIX of the difference image DG is focused (hereinafter referred to as "pixel of interest PIX"), and a mask MA of a predetermined size is set at a position in the restored image RG corresponding to the pixel of interest PIX. Then, the edge code of each pixel in the mask MA is sequentially compared with the edge code of the pixel of interest PIX.

When an edge pixel on the original edge of the inspection object is selected as the pixel of interest PIX, if the difference image DG and restored image RG are perfectly aligned, the center pixel ma0 in the mask MA corresponds to the pixel of interest PIX. In this case, it can be considered that the edge codes of the pixel of interest PIX and the pixel ma0 indicate similar values.

However, in reality, there is a possibility that the edge code of the pixel ma0 does not correspond to the pixel of interest PIX due to errors occurring during alignment or variations in the size of the object to be inspected. The size of the mask MA is adjusted in consideration of the error that occurs during this alignment and the amount of variation in the size of the object to be inspected. That is, when the pixel of interest PIX is a pixel that indicates the original edge of the object to be inspected, even if pixel ma0 is not a pixel that indicates the edge of the restored image RG, but corresponds to the inside or outside of the edge, It can be considered that there is a pixel indicating an edge at any position within the mask MA, and the edge code of that pixel is approximate to the edge code of the pixel of interest PIX. Therefore, when an edge code that approximates the edge code of the pixel of interest PIX is extracted from the mask MA, the pixel of interest PIX can be considered to be a pixel that indicates the original edge of the object to be inspected.

On the other hand, even if the pixel of interest PIX indicates an edge, if there is no edge code that approximates the edge code of the pixel of interest PIX in the mask MA, the pixel of interest PIX indicates an edge other than the original edge of the object to be inspected. In other words, it can be considered as a pixel indicating the edge of a defect.

As will be described later, among the edge pixels of the difference image DG, those whose density difference with respect to the pixel at the corresponding coordinate position in the restored image RG, pixel ma0 in the example shown in FIG. , the edge code comparison process described above is performed. Edge pixels of the difference image DG, which have a large density difference with the restored image RG, are areas where the original edge of the object to be inspected is misaligned with respect to the edge of the restored image RG, or are caused by burrs, defects, etc. It can be considered that it corresponds to the part that protrudes from the original edge. Therefore, by performing edge code comparison processing on edge pixels for which the density difference with the restored image RG is greater than or equal to a predetermined threshold, it is possible to distinguish between the original edge of the object to be inspected and the defective edge of the object to be inspected. can be separated with high precision.

<Defective pixel determination process>
Here, details of an example of the defective pixel determination process S360 will be described. In the defective pixel determination process S360 shown in FIG. 9, to obtain the minimum value of the difference between the edge code of the pixel of interest PIX shown in FIG. 8 and the edge code of the pixel in the mask MA (hereinafter referred to as "minimum difference MIND"). Then, the setting unit 34 sets an initial value to the minimum difference MIND (S361). As described later, the edge code difference (hereinafter also referred to as "edge code difference") takes a value from "0" to "180", so the setting unit 34 sets "180" as the minimum value as the initial value. Set the difference MIND.

Next, the setting unit 34 sets the minimum value to the scan counter j in the Y-axis direction (S362), and sets the minimum value to the scan counter k in the X-axis direction (S363). The scan counter j and the scan counter k are values indicating relative positions with respect to the coordinate position of the center pixel ma0 in the mask MA shown in FIG. In the example of mask MA shown in FIG. 8, the minimum value of scan counter j is "-2" and the minimum value of scan counter k is "-2". Each pixel in the mask MA can be manipulated by incrementing the value of the scan counter j and the value of the scan counter k by 1, respectively. Hereinafter, the pixel at the coordinate position (x+k, y+j) with respect to the coordinate position (x, y) of the center pixel ma0 will be referred to as a scanning pixel.

Note that the maximum value of a scanning counter j, which will be described later, is "2", and the maximum value of a scanning counter k, which will be described later, is "2". Further, the minimum and maximum values of the scanning counter j and the minimum and maximum values of the scanning counter k are respectively registered in advance in response to a user's setting operation, and can be changed as appropriate.

Next, the determination unit 35 calculates the edge strength IE (x+k, y+j) of the scanned pixel of the restored image RG, compares the edge strength IE (x+k, y+j) with a predetermined threshold, and determines the edge strength. It is determined whether IE (x+k, y+j) is greater than or equal to a predetermined threshold (S364).

As a result of the determination in step S364, if the edge strength IE (x+k, y+j) is greater than or equal to the predetermined threshold, the determination unit 35 determines the edge code difference between the pixel of interest PIX of the difference image DG and the scanned pixel of the restored image RG. is calculated (S365). For example, if the edge code of the pixel of interest PIX in the difference image DG is EC(x, y), and the edge code of the scanning pixel in the restored image RG is EC(x+k, y+j), the edge code difference ECD(x, y) is It is expressed as the following equation (3).
ECD (x, y) = f (EC (x, y) - EC (x + k, y + j)) ... (3)

Note that in equation (3), if (EC(x, y) - EC(x+k, y+j)) = θ, the value indicated by the function f(θ) is determined by the following equation (p ) is determined by either equation (s).
(p) When 0°≦θ≦180°,
f(θ)=θ
(q) When -180°≦θ≦0°,
f(θ)=-θ
(r) When 180°≦θ≦360°,
f(θ)=360°−θ
(s) When -360°≦θ≦-180°,
f(θ)=360°+θ

After step S365, the determination unit 35 compares the edge code difference ECD (x, y) calculated in step S365 with the minimum difference MIND, and determines whether the edge code difference ECD (x, y) is smaller than the minimum difference MIND. (S366).

As a result of the determination in step S366, if the edge code difference ECD (x, y) is smaller than the minimum difference MIND, the determination unit 35 updates the minimum difference MIND with the edge code difference ECD (x, y) calculated in step S365. (Rewrite) (S367).

As a result of the determination in step S364, if the edge strength IE (x+k, y+j) is not greater than or equal to the predetermined threshold value, that is, less than the threshold value, as a result of the determination in step S366, the edge code difference ECD(x, y) is not smaller than the minimum difference MIND, that is, is greater than or equal to the minimum difference MIND, or after step S367, the setting unit 34 adds "1" to the scan counter k in the X-axis direction (S368), and It is determined whether k is greater than the maximum value of the scanning counter k (S369).

As a result of the determination in step S369, if the scan counter k is not greater than the maximum value of the scan counter k, that is, if the scan counter k is less than or equal to the maximum value, the process returns to step S364, and the determination unit 35 and setting unit 34 The processes from step S364 to step S369 are repeated until k becomes larger than the maximum value.

On the other hand, as a result of the determination in step S369, if the scan counter k is larger than the maximum value of the scan counter k, the setting unit 34 adds "1" to the scan counter j in the Y-axis direction (S370), and the scan counter j is It is determined whether the scan counter j is larger than the maximum value (S371).

As a result of the determination in step S371, if the scan counter j is not greater than the maximum value of the scan counter j, that is, if the scan counter j is less than or equal to the maximum value, the process returns to step S363, and the setting unit 34 and the determination unit 35 The processes from step S363 to step S371 are repeated until j becomes larger than the maximum value.

On the other hand, if the scan counter j is larger than the maximum value of the scan counter j as a result of the determination in step S371, the determination unit 35 compares the minimum difference MIND with a predetermined threshold, and determines that the minimum difference MIND is within the predetermined threshold. It is determined whether the value is greater than or equal to the value (S372). At this time, the minimum difference MIND stores the minimum value of the edge code differences ECD (x, y) found within the mask MA.

As a result of the determination in step S372, if the minimum difference MIND is greater than or equal to the threshold value, it can be considered that the edge code corresponding to the edge code of the pixel of interest PIX in the difference image DG has not been extracted from the mask MA. Therefore, the determination unit 35 determines the pixel of interest PIX of the difference image DG to be a defective pixel (S373).

On the other hand, as a result of the determination in step S372, if the minimum difference MIND is not greater than or equal to the threshold value, that is, less than the threshold value, the edge code corresponding to the edge code of the pixel of interest PIX in the difference image DG is extracted from the mask MA. It can be considered that Therefore, the determining unit 35 determines that the pixel of interest PIX of the difference image DG is not a defective pixel, that is, a non-defective pixel (S374).

After step S373 or step S374, the determination unit 35 ends the defective pixel determination process S360.

In this way, it is determined whether a pixel in the difference image DG is a defective pixel based on the edge code of the pixel included in the difference image DG and the edge code of the pixel included in the restored image RG. Thereby, it is possible to separate the original edge pixels of the object to be inspected from the edge pixels of defects that may be included in the object to be inspected, and it is possible to determine that the original edge pixels of the object to be inspected are not defective pixels. Therefore, over-detection of defects can be suppressed.

Further, based on the edge code of the pixel of interest PIX and the edge code of the scanning pixel within the mask MA, it is determined whether the pixel of interest PIX is a defective pixel. As a result, even if there are variations in the object to be inspected, for example, if the edge code corresponding to the edge code of the pixel of interest PIX is extracted from the mask MA, it is possible to determine that the pixel of interest PIX is not a defective pixel. . Therefore, detection of defective pixels due to variations in the inspection object can be suppressed.

In this embodiment, in FIG. 9, defective pixel determination processing S360 calculates the minimum difference MIND between the edge code of the pixel of interest PIX and the edge code of the pixel within the mask MA, and sets the minimum difference MIND and a predetermined threshold value. Although an example of comparing the two has been shown, the present invention is not limited to this example.

For example, instead of determining the minimum edge code difference MIND, each time the edge code difference ECD(x,y) is calculated for a scanned pixel within the mask MA, it may be compared with a predetermined threshold.

That is, as shown in FIG. 10, in defective pixel determination processing S380, the determination unit 35 calculates the edge code difference ECD (x, y) between the pixel of interest PIX of the difference image DG and the scanned pixel of the restored image RG ( S384).

Next, the determination unit 35 compares the edge code difference ECD (x, y) calculated in step S384 with a predetermined threshold, and determines that the edge code difference ECD (x, y) is equal to or greater than the predetermined threshold. It is determined whether there is one (S385).

As a result of the determination in step S385, if the edge code difference ECD (x, y) is not equal to or greater than the threshold value, that is, if it is less than the threshold value, the determination unit 35 determines that the pixel of interest PIX of the difference image DG is a non-defective pixel. It is determined (S386), and the defective pixel determination process S380 is ended.

On the other hand, if there is no scanning pixel whose edge code difference ECD (x, y) is less than the threshold value in the mask MA, the determination unit 35 determines the pixel of interest PIX in the difference image DG to be a defective pixel. (S391), and the defective pixel determination process S380 is ended. According to this defective pixel determination process S380, the scanning can be ended when the edge code corresponding to the pixel of interest PIX of the difference image DG is found, so that the inspection time can be shortened.

In addition, in this embodiment, in FIG. 8, a mask MA of a predetermined size is set at a position of the restored image RG corresponding to the pixel of interest PIX of the difference image DG, and each pixel within the mask MA Although an example has been shown in which the edge code of the target pixel PIX is sequentially compared with the edge code of the pixel of interest PIX, the present invention is not limited to this.

(First modification)
For example, as shown in FIG. 11, focusing on a pixel of the restored image RG corresponding to the pixel of interest PIX of the difference image DG (hereinafter referred to as "corresponding pixel of interest PIX'"), the difference image corresponding to the corresponding pixel of interest PIX' is A mask MA of a predetermined size may be set at the position of DG. In this case, the central pixel in the mask MA corresponds to the corresponding pixel of interest PIX', that is, the pixel of interest PIX. Then, the edge code of each pixel in the mask MA is sequentially compared with the edge code of the corresponding pixel of interest PIX', and if an edge code that approximates the edge code of the corresponding pixel of interest PIX' is extracted from the mask MA, the edge code of each pixel of interest is extracted from the mask MA. The edge code of the pixel PIX is considered to indicate the edge of the object to be inspected, and the pixel of interest PIX is determined to be a non-defective pixel. On the other hand, if there is no edge code that approximates the edge code of the corresponding pixel of interest PIX' in the mask MA, the edge code of the pixel of interest PIX is considered to indicate a defective edge, and the pixel of interest PIX is considered to be a defective pixel. It will be judged.

In this way, based on the edge code of the corresponding pixel of interest PIX' included in the restored image RG and the edge code of the pixel in the mask MA, the pixel of interest PIX corresponding to the corresponding pixel of interest PIX' included in the difference image DG is , it is determined whether the pixel is a defective pixel or not. As a result, even if there are variations in the object to be inspected, for example, if the edge code corresponding to the edge code of the corresponding pixel of interest PIX' is extracted from the mask MA, it is possible to determine that the pixel of interest PIX is not a defective pixel. Can be done. Therefore, detection of defective pixels due to variations in the inspection object can be suppressed.

(Second modification)
Alternatively, as shown in FIG. 12, focusing on the pixel of interest PIX of the difference image DG, the first mask MA1 of a predetermined size is set at the position of the difference image DG corresponding to the pixel of interest PIX, and the first mask MA1 corresponding to the pixel of interest PIX is set. The second mask MA2 of a predetermined size may be set at the position of the difference image DG corresponding to the corresponding pixel of interest PIX' of the restored image RG. In this case, the central pixel in the first mask MA1 becomes the corresponding pixel of interest PIX', and the central pixel in the second mask MA2 becomes the corresponding pixel of interest PIX. Then, the edge code of each pixel in the first mask MA1 is sequentially compared with the edge code of the pixel of interest PIX, and if an edge code that approximates the edge code of the pixel of interest PIX is extracted from the first mask MA1, The edge code of the pixel PIX is considered to indicate the edge of the object to be inspected, and the pixel of interest PIX is determined to be a non-defective pixel. Further, when the edge code of each pixel in the second mask MA2 is compared with the edge code of the corresponding pixel of interest PIX' in order, and an edge code that approximates the edge code of the corresponding pixel of interest PIX' is extracted from the second mask MA2. Also, the edge code of the pixel of interest PIX is considered to indicate the edge of the object to be inspected, and the pixel of interest PIX is determined to be a non-defective pixel. On the other hand, if there is no edge code that approximates the edge code of the pixel of interest PIX in the first mask MA1, and there is no edge code that approximates the edge code of the corresponding pixel of interest PIX' in the second mask MA2, The edge code of pixel PIX is considered to indicate a defective edge, and pixel PIX of interest is determined to be a defective pixel.

In this way, based on the edge code of the pixel in the first mask MA1 and the edge code of the pixel in the second mask MA2, it is determined whether the pixel of interest PIX included in the difference image DG is a defective pixel. It will be judged. As a result, even if there are variations in the inspection object, for example, if the edge code corresponding to the edge code of the pixel of interest PIX is extracted from the first mask MA1, or the corresponding pixel of interest PIX' included in the restored image RG, When the edge code corresponding to the edge code is extracted from the second mask MA2, it can be determined that the pixel of interest PIX is not a defective pixel. Therefore, detection of defective pixels due to variations in the inspection object can be suppressed.

Exemplary embodiments of the invention have been described above. According to the inspection device 50, inspection method, and inspection program according to an embodiment of the present invention, a restored image RG is generated from the inspection image 21 based on a plurality of learning images LG. As a result, compared to the case where the learning image LG, which is a non-defective image, is used as a reference image, it is possible to reduce quantization errors such as density (brightness) at the time of shooting, position and angle at the time of shooting, etc., which may occur even in a non-defective image. A restored image RG close to the inspection image 21 in which variations in the inspection object are reduced can be obtained as a reference image. Further, based on the edge code of the pixel included in the difference image DG and the edge code of the pixel included in the restored image RG, it is determined whether the pixel of the difference image DG is a defective pixel. Thereby, it is possible to separate the original edge pixels of the object to be inspected from the edge pixels of defects that may be included in the object to be inspected, and it is possible to determine that the original edge pixels of the object to be inspected are not defective pixels. Therefore, over-detection of defects can be suppressed.

The embodiments described above are intended to facilitate understanding of the present invention, and are not intended to be interpreted as limiting the present invention. Each element included in the embodiment, as well as its arrangement, material, conditions, shape, size, etc., are not limited to those illustrated, and can be changed as appropriate. Further, it is possible to partially replace or combine the structures shown in different embodiments.

(Appendix)
1. An inspection device 50 that inspects whether an object to be inspected has a defect based on an inspection image 21 taken of the object to be inspected,
a restored image generation unit 31 that generates a restored image RG from the inspection image 21 based on the plurality of learning images LG;
a difference image generation unit 32 that generates a difference image DG based on the inspection image 21 and the restored image RG;
It is determined whether a pixel in the difference image DG is a defective pixel based on the edge code indicating the direction of change in density gradient in the pixel included in the difference image DG and the edge code of the pixel included in the restored image RG. A determination unit 35,
Inspection device 50.
7. An inspection method for inspecting whether or not an object to be inspected has a defect based on an inspection image 21 taken of the object to be inspected, comprising:
a restored image generation step of generating a restored image RG from the inspection image 21 based on the plurality of learning images LG;
a difference image generation step of generating a difference image DG based on the inspection image 21 and the restored image RG;
It is determined whether a pixel in the difference image DG is a defective pixel based on the edge code indicating the direction of change in density gradient in the pixel included in the difference image DG and the edge code of the pixel included in the restored image RG. a determining step;
Inspection method.
13. An inspection program executed by a computer for inspecting whether or not an object to be inspected has a defect based on an inspection image 21 taken of the object to be inspected,
a restored image generation step of generating a restored image RG from the inspection image 21 based on the plurality of learning images LG;
a difference image generation step of generating a difference image DG based on the inspection image 21 and the restored image RG;
It is determined whether a pixel in the difference image DG is a defective pixel based on the edge code indicating the direction of change in density gradient in the pixel included in the difference image DG and the edge code of the pixel included in the restored image RG. a determining step;
Inspection program.

DESCRIPTION OF SYMBOLS 10... Input/output I/F, 20... Storage part, 21... Inspection image, 22... Restoration parameter file, 30... Control part 31... Restoration image generation part, 32... Difference image generation part, 33... Extraction part, 34... Setting Part, 35... Judgment unit, 40... Bus, 50... Inspection device, 60... Imaging device, 70... Input device, 80... Output device, 100... Inspection system, B... Vector, C... Vector, DD... Density difference, DG ...Difference image, E...Pixel, EC...Edge code, ECD...Edge code difference, Ex, Ey...Amount of change, f...Function, F...Vector, i...Area number, IE...Edge intensity, j, k...Scanning counter , LG... Learning image, MA... Mask, ma0... Pixel, MA1... First mask, MA2... Second mask, MIND... Minimum difference, PIX... Pixel of interest, PIX'... Corresponding pixel of interest, RG... Restored image, S200... Restoration parameter generation processing, S300...Inspection processing, S309...Defect determination processing, S320...Restored image generation processing, S340...Difference image generation processing, S360...Defective pixel determination processing, S380...Defective pixel determination processing, x1, y1...Minimum value , x2, y2...maximum value.

Claims (13)

An inspection device that inspects whether or not an object to be inspected has a defect based on an inspection image taken of the object to be inspected, comprising:
a restored image generation unit that generates a restored image from the test image based on a plurality of learning images;
a difference image generation unit that generates a difference image based on the inspection image and the restored image;
Based on an edge code indicating a direction of change in density gradient in a pixel included in the differential image and the edge code of a pixel included in the restored image, it is determined whether the pixel in the differential image is a defective pixel. a determination unit that determines;
Inspection equipment. further comprising an extraction unit that extracts, as a pixel of interest, a pixel whose density gradient is equal to or greater than a predetermined value in the difference image;
The determination unit determines whether the pixel of interest is a defective pixel based on the edge code of the pixel of interest and the edge code of a pixel included in the restored image.
The inspection device according to claim 1. Further comprising a setting unit for setting a mask of a predetermined size centered on the coordinates of a pixel corresponding to the pixel of interest in the restored image,
The determination unit determines whether the pixel of interest is a defective pixel based on the edge code of the pixel of interest and the edge code of a pixel within the mask.
The inspection device according to claim 2. Further comprising a setting unit for setting a mask of a predetermined size centered on the coordinates of a pixel corresponding to a predetermined pixel included in the restored image in the difference image,
The determination unit determines whether the corresponding pixel is a defective pixel based on the edge code of the predetermined pixel and the edge code of the pixel within the mask.
The inspection device according to claim 1. In the restored image, a first mask of a predetermined size centered on the coordinates of a corresponding pixel corresponding to a predetermined pixel included in the difference image is set; further comprising a setting section for setting a second mask of a predetermined size centered on the coordinates,
The determination unit determines whether the predetermined pixel is a defective pixel based on the edge code of the pixel in the first mask and the edge code of the pixel in the second mask.
The inspection device according to claim 1. The restored image includes a plurality of regions,
The restored image generation unit generates the restored image from the test image for each region based on the plurality of learning images.
An inspection device according to any one of claims 1 to 5. An inspection method for inspecting whether or not the object to be inspected has a defect based on an inspection image taken of the object to be inspected, the method comprising:
a restored image generation step of generating a restored image from the test image based on a plurality of learning images;
a difference image generation step of generating a difference image based on the inspection image and the restored image;
Based on an edge code indicating a direction of change in density gradient in a pixel included in the differential image and the edge code of a pixel included in the restored image, it is determined whether the pixel in the differential image is a defective pixel. a determination step of determining,
Inspection method. further comprising an extraction step of extracting a pixel with a density gradient greater than or equal to a predetermined value in the difference image as a pixel of interest;
The determining step includes determining whether the pixel of interest is a defective pixel based on the edge code of the pixel of interest and the edge code of a pixel included in the restored image.
The inspection method according to claim 7. further comprising a setting step of setting a mask of a predetermined size centered on the coordinates of a pixel corresponding to the pixel of interest in the restored image;
The determining step includes determining whether the pixel of interest is a defective pixel based on the edge code of the pixel of interest and the edge code of a pixel within the mask.
The inspection method according to claim 8. further comprising a setting step of setting a mask of a predetermined size centered on the coordinates of a pixel corresponding to a predetermined pixel included in the restored image in the difference image;
The determining step includes determining whether the corresponding pixel is a defective pixel based on the edge code of the predetermined pixel and the edge code of the pixel within the mask.
The inspection method according to claim 7. In the restored image, a first mask of a predetermined size centered on the coordinates of a corresponding pixel corresponding to a predetermined pixel included in the difference image is set; further comprising a setting step of setting a second mask of a predetermined size centered at
The determining step includes determining whether the predetermined pixel is a defective pixel based on the edge code of the pixel in the first mask and the edge code of the pixel in the second mask. including,
The inspection method according to claim 7. The restored image includes a plurality of regions,
The restored image generation step includes generating the restored image from the test image for each region based on the plurality of learning images.
The inspection method according to any one of claims 7 to 11. An inspection program executed by a computer that inspects whether or not the object to be inspected has a defect based on an inspection image taken of the object to be inspected,
a restored image generation step of generating a restored image from the test image based on a plurality of learning images;
a difference image generation step of generating a difference image based on the inspection image and the restored image;
Based on an edge code indicating a direction of change in density gradient in a pixel included in the differential image and the edge code of a pixel included in the restored image, it is determined whether the pixel in the differential image is a defective pixel. a determination step of determining,
Inspection program.

Images