JP2011232303A - Image inspection method and image inspection device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the quality of a reference space created based on a group of normal products in an image inspection method and an image inspection device which inspect an image of a target product to be inspected, by comparing an inspection threshold determined based on the reference space with a Mahalanobis distance of the image of the target product.SOLUTION: It is evaluated whether a defectiveness determination image being an image of a target image to be inspected, which is determined to be defective as a result of inspection using an initial reference space (S31 to S35), is erroneously determined or not. A computer stores, into a storage device, an erroneously determined image being the defectiveness determination image evaluated to be erroneously determined (S37). It is determined whether the number N of erroneously determined images stored in the storage device reaches a prescribed number Nor not (S38). When it reaches the prescribed number N(S38 Yes), erroneously determined images are read out from the storage device (S39). The read-out erroneously determined images are added to an initial group of reference images to create a reference space again (S40).

Description

  The present invention relates to an image inspection method and an image inspection apparatus for inspecting the quality of an image showing the appearance of a product to be inspected in order to determine the presence or absence of defects in the product to be inspected.

  Conventionally, there is a case where an appearance inspection of a product to be inspected which is an inspection target is performed. Here, it is considered that the inspector automates the appearance inspection (visual inspection) that has been performed visually. In this case, when the inspector obtains a wide range of information from the appearance of the product to be inspected, and the inspector inspects these information comprehensively, the inspection standard value for automation is used. It can be difficult to decide. Therefore, conventionally, an inspection method utilizing the MT method (Mahalanobis-Taguchi method), which is a type of multivariate analysis (analysis method that aggregates a large amount of information into one scale), is known (for example, see Patent Document 1). .)

  This MT method is a method in which a large number of pieces of information are collected on one scale called Mahalanobis distance, and the quality is determined based on the magnitude of the Mahalanobis distance. That is, a feature amount (information) is extracted from an image of a normal product group in a normal state, and a reference Mahalanobis distance is calculated (also referred to as creating a reference space, and the reference space Mahalanobis distance is about 1 on average. Become). Next, the Mahalanobis distance of the image of the inspected product to be inspected is calculated. Then, as the Mahalanobis distance is further away from the reference Mahalanobis distance, it is determined that the inspected product is in an abnormal state.

JP-A-2005-252451

  However, in spite of the fact that the above-mentioned reference space has a great influence on the inspection accuracy, the reference space once created continues to be used as it is in the conventional methods such as the method disclosed in Patent Document 1. For this reason, if the reference space is not appropriate, there is a problem that an erroneous determination may be made that the product is a non-defective product even though it is a non-defective product, and there is a problem that there is a limit to the improvement in inspection accuracy.

  The present invention has been made in view of the above problems, and an object of the present invention is to provide an image inspection method and an image inspection apparatus capable of improving the quality of a reference space in order to realize further improvement in inspection accuracy. .

In order to solve the above problems, the present invention is an image inspection method for performing pass / fail inspection of an image of a product to be inspected processed by a computer,
A reference image acquisition step of acquiring images of a plurality of normal products having a normal appearance as reference images,
An extraction step for extracting a predetermined feature amount from each reference image acquired in the reference image acquisition step;
A creation step for creating a reference space composed of the Mahalanobis distance of each reference image based on the feature amount extracted in the extraction step;
Based on the Mahalanobis distance of the reference space created in the creation step, a threshold value determining step for determining a threshold value that serves as a criterion for quality determination of the image of the inspected product;
An inspection execution step of calculating pass / fail of the image of the inspected product by calculating the Mahalanobis distance of the image of the inspected product and comparing the Mahalanobis distance with the threshold value;
The defect judged to be a misjudgment as a result of evaluation as to whether or not the defect judgment image, which is an image of the inspected product judged as a defect in the inspection execution step, is a misjudgment. An erroneous determination image acquisition step of acquiring an erroneous determination image which is a determination image;
A re-creation step of re-creating the reference space by adding the erroneous determination image to the group of the reference images acquired in the reference image acquisition step.

  According to this, in the erroneous determination image acquisition step, a defect determination image (error determination image) evaluated as an erroneous determination is acquired. In the re-creation step, the erroneous determination image is added to the original group of reference images, and the reference space is created again. Since it can be said that the re-created reference space takes into account the feature amount of the erroneously determined image, the quality of the reference space can be improved. Therefore, it is possible to further improve the inspection accuracy by inspecting the image of the inspected product using the reference space.

Further, in the image inspection method of the present invention, a storage step that is executed before the erroneous determination image acquisition step and that stores the erroneous determination image in a storage unit;
A quantity determination step for determining whether or not the quantity of the erroneous determination image stored in the storage means in the storage step has reached a predetermined amount,
The erroneous determination image acquisition step reads the erroneous determination image in a form of reading the erroneous determination image from the storage means when it is determined in the quantity determination step that the quantity of the erroneous determination image has reached the predetermined amount. It is a step to acquire.

  According to this, in the erroneous determination image acquisition step, when the number of erroneous determination images evaluated as erroneous determination reaches a predetermined amount, the erroneous determination image is read from the storage unit. As a result, since the reference space is created after the reading, it is not necessary to recreate the reference space every time it is evaluated as erroneous determination, and the calculation process can be made more efficient.

Further, the present invention is an image inspection apparatus for performing quality inspection of an image of a product to be inspected,
Reference image acquisition means for acquiring images of a plurality of normal products in a normal appearance as reference images,
Extraction means for extracting a predetermined feature amount from each reference image acquired by the reference image acquisition means;
Creating means for creating a reference space consisting of Mahalanobis distance of each reference image based on the feature amount extracted by the extracting means;
Based on the Mahalanobis distance of the reference space created by the creating means, a threshold value determining means for determining a threshold value that serves as a reference for determining the quality of the image of the inspected product;
An inspection execution means for determining the quality of the image of the inspected product by calculating the Mahalanobis distance of the image of the inspected product and comparing the Mahalanobis distance with the threshold value;
The defect judged to be a misjudgment as a result of the evaluation as to whether or not the defect judgment image, which is an image of the inspected product judged as a defect by the inspection execution means, is a misjudgment. An erroneous determination image acquisition means for acquiring an erroneous determination image which is a determination image;
Re-creation means for adding the erroneous determination image to the group of the reference images acquired by the reference image acquisition means and re-creating the reference space.

  Thereby, the quality of the reference space can be improved as in the image inspection method. Therefore, it is possible to further improve the inspection accuracy by inspecting the image of the inspected product using the reference space.

1 is a diagram illustrating an overall configuration of an image inspection apparatus 10. It is the flowchart which showed the procedure when creating a reference space. It is a figure for demonstrating the luminance waveform of the workpiece | work image. It is the figure which illustrated how to draw the horizontal line drawn on a luminance waveform, when calculating a differential value and an integral value. It is a figure explaining the method of calculating a differential value and an integral value from a luminance waveform. It is the flowchart which showed the procedure when determining the quality of a to-be-inspected image. It is the flowchart which showed the specific procedure when performing the quality determination of a to-be-inspected image while creating a reference | standard space.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a diagram showing an overall configuration of an image inspection apparatus 10 according to an embodiment of the present invention. The image inspection apparatus 10 photographs the appearance of the work 20 that is the inspection target, and performs an inspection based on the obtained image data. The workpiece 20 is, for example, a cut metal part, and requires inspection of defects such as scratches, burrs, and foreign matters on the surface of the workpiece 20. The image inspection apparatus 10 uses the workpiece 20 as a product to be inspected, and inspects the presence or absence of the defect.

  The image inspection apparatus 10 includes a camera 12, an illumination 13, an electronic calculator 14, a display device 15, and a storage device 16. The camera 12 captures the appearance of the workpiece 20 and obtains an inspection image that is image data of the workpiece 20. The camera 12 is specifically a CCD camera. As another example, a CMOS camera, an analog camera, or the like may be used as the camera 12. The illumination 13 adjusts the brightness of the imaging location of the workpiece 20 when imaging the appearance of the workpiece 20 by the camera 12. Illumination 13 is specifically LED ring illumination. As another example, fiber illumination, LED illumination, ring illumination, coaxial epi-illumination, or the like may be used as the illumination 13.

  The inspected image taken by the camera 12 is sent to the electronic computer 14. Based on the inspected image, the electronic computer 14 determines the presence / absence of a defect in the workpiece 20, that is, the quality determination, and causes the display device 15 to display the result of the quality determination. The storage device 16 as a storage means is, for example, a hard disk device of the electronic computer 14 and stores various information such as various images taken by the camera 12, reference space calculated by the electronic computer 14, and determination results. is there. Note that a memory such as a RAM or a flash memory may be used as the storage device 16.

  The camera 12 also captures normal products that are known to be non-defective (normal). As a result, a reference image serving as an inspection reference is obtained. Here, the reference image is an image of each of a plurality of normal products in a normal state, and is an image group used for creating a Mahalanobis reference space. The electronic computer 14 creates a Mahalanobis reference space to be used for pass / fail determination based on a plurality of reference images.

  Here, FIG. 2 is a flowchart showing a procedure for creating a reference space and setting a threshold value for determining pass / fail of the image to be inspected based on the reference space. The flowchart in FIG. 2 is for explaining the basic concept of the method for creating the reference space in the present invention, and the specific procedure is shown in FIG. 7 described later.

First, each of a plurality of normal product images in a normal state is taken by the camera 12 (step S11). Next, a feature amount (information to be a feature) is extracted from each photographed image (step S12). This feature amount represents the state of the workpiece 20 and is extracted according to a predetermined item. Here, feature amounts are extracted from N reference images according to K items such as a maximum luminance value, a minimum luminance value, a differential value of a luminance waveform, and an integrated value of a luminance waveform. Table 1 shows the result of extracting the feature amount. In Table 1, the subscript n (1, 2,..., N) of the feature quantity X _nk is a number for identifying the reference image (reference image No), and the subscript k (1, 2,..., K). Represents a number (item No.) for identifying an item.

  Here, a supplementary explanation will be given regarding the above-described feature amount. FIG. 3 is a diagram for explaining the luminance waveform. FIG. 3A shows a work image 401 of the work 20. FIG. 3B shows a graph of the luminance waveform of the image 401. Specifically, the horizontal axis of the graph of FIG. 3 indicates the pixel position in a predetermined line of the work image 401, and the vertical axis indicates each pixel. The luminance value of the pixel is shown. The curve on this graph is the luminance waveform. The luminance waveform is a waveform obtained by converting the luminance value of each pixel into a waveform for each row (line) of pixels, that is, a relationship between each pixel included in each row and the luminance value in each pixel. As an example, FIG. 3 shows a luminance waveform of a work image 401 having a luminance of 256 gradations and a pixel size n × n.

  As shown in FIG. 3 (b), the brightness of the work image 401 varies with respect to the pixels. Then, as described above, feature quantities such as the maximum luminance value, the minimum luminance value, the differential value of the luminance waveform, and the integrated value of the luminance waveform can be extracted from the luminance waveform. Here, the differential value of the luminance waveform and the integrated value of the luminance waveform will be described. 4 and 5 are diagrams for explaining the above. That is, as shown in FIG. 5, a straight line having an arbitrary luminance value, that is, a horizontal line L1 is drawn from the luminance waveform, and the number of intersections of the horizontal line L1 and the luminance waveform is obtained as a differential value.

  This differential value can be used as an index of the number of changes in the feature amount of the image. In other words, it can be said that the larger the differential value, the more portions in the image where the feature amount varies. In the case of FIG. 5, a feature quantity indicating how much a fluctuating part is included in the work image 401 with reference to the luminance C1 indicated by the horizontal line L1 is extracted. In this way, by using the differential value as a feature amount, even in an image that includes many parts where the feature amount varies, it is possible to create a reference space that accurately reflects the feature amount. The quality of the image can be determined with high accuracy.

  Further, as shown in FIG. 5, a straight line having an arbitrary luminance value, that is, a horizontal line L2 is drawn with respect to the luminance waveform, and a total sum of intervals of the luminance waveform above the horizontal line L2 is obtained as an integrated value. This integral value can be used as an index of the proportion of the portion having a certain feature amount or more in the image. In other words, it can be said that the larger the integral value, the more parts in the image that have a certain or larger feature amount. In the case of FIG. 5, a feature amount indicating how much a part having a luminance C2 or higher indicated by the horizontal line L2 is included in the work image 401 is extracted. As described above, by using the integral value as a feature amount, even in an image including many parts having a feature amount greater than a certain amount, a reference space in which the feature amount is accurately reflected can be created. As a result, the quality of the image to be inspected can be determined with high accuracy.

  The horizontal line drawn on the luminance waveform can be set as shown in FIG. 4, for example. That is, in FIG. 4, for example, an average value of a plurality of luminance values included in a predetermined row of the work image 401 is μ, a standard deviation is σ, and μ ± σ, μ ± 0.8σ, μ ± 0.6σ,.・ Set horizontal lines at regular intervals as In this case, a differential value and an integral value for each horizontal line can be extracted as a feature amount.

Next, the Mahalanobis distance of each reference image is calculated, and the Mahalanobis distance serving as a reference for the inspection is calculated. That is, a reference space is created (step S13). Specifically, first, an average value μ _k and a standard deviation σ _k of feature amounts are calculated for each item in Table 1. Then, the calculated feature value x _nk is calculated by substituting the calculated average value μ _k and standard deviation σ _k into (Equation 1). Then, as shown in Table 2, a matrix of normalized feature values x _nk is used.

Next, the correlation coefficient r _pq between the item p and the item q is calculated by (Equation 2). Based on the correlation coefficient r _pq , a correlation coefficient matrix R having each correlation coefficient as an element is obtained as shown in (Equation 3). Then, as shown in (Expression 4), an inverse matrix A of the correlation coefficient matrix R is calculated.

Next, the Mahalanobis distance MD _n of the _nth reference image is calculated by (Equation 5). Eventually, the Mahalanobis distance of all reference images is calculated, and the creation of the reference space is completed.

At the end of step S13, the Mahalanobis distance of all the reference spaces, the average value μ _k of each item, the standard deviation σ _{k of} each item, and the inverse matrix A are stored in the storage device 16 as parameters of the reference space.

  Next, based on the Mahalanobis distance of each reference image, an inspection threshold value serving as a pass / fail judgment criterion is determined (step S14). That is, since the Mahalanobis distance of the reference image also has a distribution, for example, the inspection threshold is determined based on the maximum Mahalanobis distance among them. For example, since the maximum value of the Mahalanobis distance of the reference image calculated by Equation 5 is 1.6, the inspection threshold is set to 1.7 in order to make products up to a Mahalanobis distance of 1.6 good products.

  As described above, the electronic computer 14 creates the reference space and determines the inspection threshold based on the reference space. Next, the electronic computer 14 determines pass / fail of the inspected product to be inspected based on the inspection threshold. Here, FIG. 6 is a flowchart showing a procedure for determining pass / fail of an inspected image that is an image of the inspected product. Note that the flowchart of FIG. 6 is for explaining the basic concept of the procedure for determining pass / fail of the image to be inspected in the present invention, and the specific procedure is shown in FIG. 7 described later. ing.

  First, an image of the workpiece 20 (inspected product) to be inspected is taken by the camera 12 (step S21). Next, a feature amount (information that becomes a feature) is extracted from each of the taken images to be inspected (step S22). This feature amount is extracted according to the item used when creating the reference space.

Next, the Mahalanobis distance of the image to be inspected is calculated through the same procedure as that for creating the reference space (step S23). However, for the average value μ _k and standard deviation σ _{k of} the feature quantity used when standardizing the feature quantity, and the inverse matrix A of the correlation coefficient matrix R, the values calculated when creating the reference space are used. To do.

  Next, the pass / fail determination of the image to be inspected is performed by comparing the Mahalanobis distance MD of the image to be inspected with the previously set inspection threshold (step S24). When the Mahalanobis distance MD is larger than the inspection threshold (Yes at Step S24), it is determined that there is an abnormality in the inspected image, that is, the workpiece 20 is a defective product (Step S25). On the other hand, when the Mahalanobis distance MD is less than or equal to the inspection threshold (step S24, No), it is determined that the inspected image is normal, that is, the workpiece 20 is a non-defective product (step S26).

  Thus, in this embodiment, the pass / fail determination of the image to be inspected is performed by comparing the inspection threshold determined based on the reference space with the Mahalanobis distance of the image to be inspected. It is not always appropriate. For example, the normal product group for creating the reference space also has a distribution in the feature amount, and if the reference space is created only with normal products having the mode value feature amount in the distribution, it is determined that the product is excessively defective. May end up. In addition, the reference space that was appropriate at the time of creation may become unsuitable afterwards due to changes in the criteria required for inspection. That is, if the reference space once created is used as it is, an erroneous determination may be made in which it is determined as a defective product even though it is a non-defective product. Therefore, in the present invention, the reference space once created can be reviewed. Here, FIG. 7 is a flowchart showing a specific procedure for creating a reference space and determining pass / fail of an image to be inspected.

  First, each of a plurality of normal product images in a normal state is taken by the camera 12 (step S31). This step S31 is the same as the process of step S11 described above. Step S31 corresponds to the “reference image acquisition step” of the present invention, and the electronic computer 14 that executes step S31 corresponds to the “reference image acquisition means” of the present invention.

  Next, a feature amount (information serving as a feature) is extracted from each captured reference image (step S32). This step S32 is the same as the process of step S12 described above. Step S32 corresponds to the “extraction step” of the present invention, and the electronic computer 14 executing step S32 corresponds to the “extraction means” of the present invention.

  Next, the Mahalanobis distance of each reference image is calculated. That is, a reference space is created (step S33). This step S33 is the same as the process of step S13 described above. Step S33 corresponds to the “creation step” of the present invention, and the electronic computer 14 executing step S33 corresponds to the “creation unit” of the present invention.

  Next, based on the Mahalanobis distance of each reference image, an inspection threshold value serving as a pass / fail criterion is determined (step S34). This step S34 is the same as the process of step S14 described above. Step S34 corresponds to the “threshold determination step” of the present invention, and the electronic computer 14 executing step S34 corresponds to the “threshold determination means” of the present invention.

  Next, the inspection image is inspected by the same processing as that shown in FIG. 6 (step S35). That is, in this step S35, the process of the flowchart of FIG. 6 is executed. Step S35 corresponds to the “inspection execution step” of the present invention, and the electronic computer 14 executing step S35 corresponds to the “inspection execution means” of the present invention. Various data such as a reference image, a reference space, and an image to be inspected are stored in the storage device 16.

  Here, it is assumed that there is a defect determination image that is an image to be inspected that is determined as a defective product by the processing in steps S31 to S35. In this case, for the reason described above, the determination may be an erroneous determination. Therefore, it is evaluated whether or not an erroneous determination is made for the product to be inspected (work 20) related to the defect determination image. Specifically, for example, a visual inspector visually determines the quality of the inspected product. For example, when it is determined that the surface of the workpiece 20 is damaged and is determined to be a defective product, the visual inspector determines whether or not the scratch corresponds to a scratch that is determined to be a defective product based on the length of the scratch. Based on evaluation. Further, for example, the evaluation may be automatically performed by a computer by means different from the MT method. In this case, for example, the surface roughness of the surface of the workpiece 20 related to the defect determination image can be automatically measured by a computer, and the evaluation is performed based on the surface roughness.

  It should be noted that the timing for performing this evaluation can be set at regular intervals, for example, every time step S35 is performed, or when the number of times step S35 is performed is a predetermined number. Further, in the case of inspecting a plurality of workpieces 20 (inspected products) at the time of one inspection, the evaluation may be performed when inspection of all the workpieces 20 is completed, that is, at each inspection. .

  Returning to the description of FIG. 7, it is determined whether or not an erroneous determination has occurred (step S36), assuming that the above-described evaluation has been made after the inspection in step S35. Specifically, for example, when the evaluation is made by a visual inspector, it is determined whether or not an erroneous determination has occurred based on an input operation by the visual inspector. That is, the visual inspector can perform an input operation to the electronic computer 14 that the erroneous determination has occurred, and the electronic computer 14 determines that the erroneous determination has occurred when the input has been made. For example, when the above evaluation is automatically performed by a computer, it is determined whether or not an erroneous determination has occurred based on the evaluation result by the computer.

  When it is determined in step S36 that an erroneous determination has occurred (Yes in step S36), an erroneous determination image that is a defect determination image related to the erroneous determination is stored in the storage device 16 (step S37). If the erroneous determination image is already stored in the storage device 16 as the image to be inspected, the erroneous determination image is moved from the initial storage area to the predetermined erroneous determination image storage area. Step S37 corresponds to the “storage step” of the present invention.

Then, the quantity N misjudgment image stored in the storage device 16 determines whether it has reached the predetermined amount N ₀ (step S38). Step S38 corresponds to the “quantity determination step” of the present invention. Here, when the number N of misjudgment images has not yet reached the predetermined amount N ₀ (step S38, No), the process returns to step S35. In this case, the reference space is not yet reviewed, and the original reference space (original inspection threshold) is used as it is, and the image inspection is continued for other images to be inspected. In this case, when an erroneous determination occurs again (Yes in step S36), an erroneous determination image related to the erroneous determination is stored in the storage device 16 (step S37), and thus the erroneous determination stored in the storage device 16 is performed. Images will be accumulated.

Then, in step S38, the if the quantity N of erroneous determination image has reached a predetermined amount _{N 0} (step S38, the Yes), it reads all of the erroneous determination image stored in the storage device 16 (step S39). Note that step S39 corresponds to the “false determination image acquisition step” of the present invention, and the electronic computer 14 that executes step S39 corresponds to the “false determination image acquisition unit” of the present invention.

  Next, as a review of the reference image, all the read erroneous determination images are added to the group of reference images acquired in step S31, and a reference space is created again (step S40). The creation method of the reference space is the same as the creation method in step S33. Step S40 corresponds to the “re-creation step” of the present invention, and the electronic computer 14 executing step S40 corresponds to the “re-creation unit” of the present invention.

  Next, the process returns to step S34, and a new inspection threshold is determined based on the re-created reference space (step S34). Thereafter, the quality inspection of the image to be inspected is performed based on the new inspection threshold (step S35). As described above, as a result of recreating the reference space, in this example, it was possible to reduce erroneous determination by 50% or more compared to before recreating the reference space.

  On the other hand, if no erroneous determination occurs in step S36 (No in step S36), the process returns to step S35. In this case, the original reference space is used as it is, and the quality inspection of the inspected image is continued (step S35). In addition, when the reference space is recreated and no erroneous determination occurs thereafter (step S36, No), the recreated reference space is continuously used as it is, and the quality of the image to be inspected is determined. The inspection is continued (step S35).

As described above, in the present embodiment, the erroneous determination image is added to the group of reference images, and the reference space is recreated. Therefore, a high-quality reference space is created in consideration of the feature amount of the erroneous determination image. it can. Therefore, it can lead to the further improvement of inspection accuracy. The reference space is erroneous since the quantity N of the determination image is recreated at the timing reaches a predetermined amount N _0, there is no need to recreate the reference space for each erroneous determines, it can be efficiently calculation process.

The image inspection method and the image inspection apparatus of the present invention are not limited to the above-described embodiment, and can be variously modified without departing from the scope of the claims. For example, in the above embodiment, the reference space is recreated at the timing when the number N of erroneously determined images reaches the predetermined amount N ₀ , but it may be when the inspected image that has been inspected reaches a predetermined amount. In addition, the reference space may be recreated every time an erroneous determination is made.

DESCRIPTION OF SYMBOLS 10 Image inspection apparatus 12 Camera 13 Illumination 14 Electronic computer 15 Display apparatus 16 Memory | storage device 20 Workpiece | work

Claims (3)

An image inspection method for performing pass / fail inspection of an image of a product to be inspected processed by a computer,
A reference image acquisition step of acquiring images of a plurality of normal products having a normal appearance as reference images,
An extraction step for extracting a predetermined feature amount from each reference image acquired in the reference image acquisition step;
A creation step for creating a reference space composed of the Mahalanobis distance of each reference image based on the feature amount extracted in the extraction step;
Based on the Mahalanobis distance of the reference space created in the creation step, a threshold value determining step for determining a threshold value that serves as a criterion for quality determination of the image of the inspected product;
An inspection execution step of calculating pass / fail of the image of the inspected product by calculating the Mahalanobis distance of the image of the inspected product and comparing the Mahalanobis distance with the threshold value;
The defect judged to be a misjudgment as a result of evaluation as to whether or not the defect judgment image, which is an image of the inspected product judged as a defect in the inspection execution step, is a misjudgment. An erroneous determination image acquisition step of acquiring an erroneous determination image which is a determination image;
An image inspection method comprising: a re-creation step of re-creating the reference space by adding the erroneous determination image to the group of the reference images acquired in the reference image acquisition step. A storage step of storing the erroneous determination image in a storage means, which is executed before the erroneous determination image acquisition step;
A quantity determination step for determining whether or not the quantity of the erroneous determination image stored in the storage means in the storage step has reached a predetermined amount,
The erroneous determination image acquisition step reads the erroneous determination image in a form of reading the erroneous determination image from the storage means when it is determined in the quantity determination step that the quantity of the erroneous determination image has reached the predetermined amount. The image inspection method according to claim 1, wherein the image inspection method is an acquisition step. An image inspection apparatus for inspecting the quality of an image of a product to be inspected,
Reference image acquisition means for acquiring images of a plurality of normal products in a normal appearance as reference images,
Extraction means for extracting a predetermined feature amount from each reference image acquired by the reference image acquisition means;
Creating means for creating a reference space consisting of Mahalanobis distance of each reference image based on the feature amount extracted by the extracting means;
Based on the Mahalanobis distance of the reference space created by the creating means, a threshold value determining means for determining a threshold value that serves as a reference for determining the quality of the image of the inspected product;
An inspection execution means for determining the quality of the image of the inspected product by calculating the Mahalanobis distance of the image of the inspected product and comparing the Mahalanobis distance with the threshold value;
The defect judged to be a misjudgment as a result of the evaluation as to whether or not the defect judgment image, which is an image of the inspected product judged as a defect by the inspection execution means, is a misjudgment. An erroneous determination image acquisition means for acquiring an erroneous determination image which is a determination image;
An image inspection apparatus comprising: a re-creation unit that re-creates the reference space by adding the erroneous determination image to the group of the reference images acquired by the reference image acquisition unit.