JP6410459B2 - Image inspection method and image inspection apparatus - Google Patents

Description

  The present invention relates to an image inspection method and an image inspection apparatus for performing defect determination of an inspection object through image processing on image data obtained by imaging the appearance of the inspection object.

  2. Description of the Related Art Conventionally, in manufacturing various industrial products, an inspection technique for performing defect determination (good / bad determination) based on image processing on image data acquired by imaging an inspection object is known.

  Image processing in this type of image inspection (also referred to as optical inspection, appearance inspection, etc.) makes a pass / fail determination by extracting defects (scratches and deformation) from an image (grayscale image) acquired by imaging the inspection object. For example, while calculating a feature amount (luminance, gray value, edge strength, etc.) that expresses some characteristic of, for example, a pixel in the image data of the inspection object, a threshold value corresponding to the feature amount that can distinguish a defect is created. Then, a pixel having a feature amount outside this threshold is extracted as a defective portion.

  However, disturbance information such as the external shape of the inspection object and the reflection of illumination is present in the grayscale image obtained by actually imaging the inspection object. In order to maintain the defect extraction sensitivity while suppressing such erroneous extraction of disturbance information, it is important to set an appropriate threshold value for each pixel, for example.

  As a method for solving this problem, for example, the following Patent Document 1 discloses a technique for generating a threshold value for each pixel by using a statistic of gray value obtained by imaging an object to be inspected. In Patent Document 1, first, an average value and a standard deviation of shade values are calculated for each pixel having the same coordinate from a plurality of non-defective sample gray images prepared in advance, and a threshold value for each pixel is calculated using the calculated average value and standard deviation as parameters. Is calculated. At the time of inspection, the gray value of the gray image of the inspection object is compared with a threshold value for each pixel, and pixels that are outside the threshold value are extracted as defective portions. In Patent Literature 1, an attempt is made to set a threshold value according to the statistical behavior of the background gray value by such image processing.

  When the technique disclosed in Patent Document 1 uses feature values other than gray values and is generalized, the following inspection method can be considered. First, for example, feature quantities such as luminance and edge strength are calculated for each pixel as feature quantities that can be detected by emphasizing a defective portion from gray images of a plurality of non-defective samples. Next, a statistic calculation is performed on the feature amount calculated for each pixel having the same coordinates, and a predetermined statistic amount (for example, the above average value or standard deviation) is calculated. Next, a threshold for a feature amount that emphasizes a defective portion is generated for each pixel using these statistics as parameters. And the pixel of the image which imaged the actual to-be-inspected object is test | inspected using the produced | generated threshold value, and a defect site | part is detected.

  According to such image processing, it is possible to set a threshold value corresponding to the statistical behavior of the background feature amount for each pixel, and the threshold accuracy is improved. As a result, it is considered that defect extraction capability can be maintained in other pixels while erroneous background extraction is suppressed, and inspection accuracy can be improved.

JP 2005-265661 A

  However, in practice, the implementation of the technique disclosed in Patent Document 1 has the following difficulties. That is, due to the influence of individual differences and alignment tolerances, even for pixels with the same coordinates, the gray value differs for each gray image of a good sample and shows a different spatial distribution. For the same reason, feature values calculated from pixels having the same coordinates also have different values. That is, in order to give reliability to the statistical amount (or the threshold value generated based on the statistical amount) that emphasizes the defective portion, a grayscale image obtained by capturing a sufficiently large number of non-defective samples is required. However, for example, in order to generate a reliability threshold value for all pixel regions of an image obtained by imaging an inspection object, there is a problem that gray images of a very large number of non-defective samples must be prepared.

  An object of the present invention is to solve the above-described problem and use a relatively small amount of non-defective samples without generating a large amount of non-defective samples when generating a feature amount threshold value for performing defect determination from a plurality of non-defective samples. The feature amount threshold value can be generated.

To solve the above problems, the present invention, in the image inspecting method acquires image data of the object determining the defect of the inspection object, the pixels of the pixels in the same coordinates of the image data of a plurality of non-defective samples and values, based on the first feature amount acquired by the statistical amount calculation for the pixel values of a plurality of pixels located around the pixel of the same coordinate, and a threshold generation step of generating a feature value threshold for each pixel, the A correction step of acquiring image data of the object to be inspected, and correcting the feature amount threshold value generated for each pixel in the threshold value generation step based on a second feature amount acquired for each pixel; and the inspection object a first feature amount acquired from the image data of the defect extraction step of extracting a pixel exceeding the threshold value by comparing the corrected the feature amount threshold in said correction step as a defect site, the defect extraction step By Extracted pixels is characterized in that it comprises a determining defect determination step the presence or absence of a defect of the inspection object based on characteristics of the defect site configuration.

In the present invention, an image inspection apparatus which performs defect determination acquired by the inspection object image data of the object to be inspected, imaging means for obtaining image data of an imaged object, captured by the image pickup means Based on the first feature amount acquired by the statistic calculation for the pixel value of the pixel at the same coordinate in the image data of the plurality of non-defective samples and the pixel value of the plurality of pixels located around the pixel at the same coordinate , and a control unit for generating a feature value threshold of each, and a storage unit for storing the feature amount threshold value, the control device, from the image data of the object imaged by the imaging means, obtained for each pixel Based on the second feature value, the feature value threshold value for each pixel stored in the storage unit is corrected, the first feature value obtained from the image data of the inspection object, and the corrected feature value Threshold By comparing the extracted pixels exceeding the threshold value as a defective portion, the extracted pixels and judging the presence or absence of a defect of the inspection object based on characteristics of the defect site configuration.

  According to the above configuration, the statistical amount related to the first feature amount necessary for calculating the feature amount threshold necessary for defect determination is calculated from the pixels at the same coordinates and the pixels within the alignment tolerance of the images of the non-defective samples. It is calculated using the first feature amount. That is, the feature amount threshold value used for a specific pixel is calculated with reference to the first feature amount of the pixel within the alignment tolerance in addition to the pixel. As a result, it is possible to ensure the same degree of reliability as the value calculated from the images of a plurality of non-defective samples by the conventional method, for the statistical amount related to the first feature amount for the specific pixel of the image of the single non-defective sample. Therefore, it is possible to remarkably reduce the number of non-defective sample images necessary for generating a feature amount threshold sufficient to obtain the same level of reliability as in the past, and to reduce the number of required non-defective samples.

1 is a block diagram illustrating a configuration example of an image inspection apparatus capable of performing image processing according to the present invention. It is explanatory drawing which showed 1st Embodiment of the image processing by this invention by the functional block structure. It is the flowchart figure which showed the flow of the feature-value threshold value generation in 1st Embodiment of the image processing by this invention. It is the flowchart figure which showed the image test | inspection procedure in 1st Embodiment of the image processing by this invention. It is explanatory drawing which showed the image data processed in 1st Embodiment of the image processing by this invention. It is explanatory drawing which showed the shading value in the background vicinity especially in the image data processed in 1st Embodiment of the image processing by this invention. It is the flowchart figure which showed the flow of the feature-value threshold value generation in 2nd Embodiment of the image processing by this invention. It is explanatory drawing which showed 3rd Embodiment of the image processing by this invention by the functional block structure. It is the flowchart figure which showed the image test | inspection procedure in 3rd Embodiment of the image processing by this invention. It is explanatory drawing which showed the image data processed in 3rd Embodiment of the image processing by this invention. It is the flowchart figure which showed the flow of the feature-value threshold value generation of the image processing by this invention in detail. It is explanatory drawing which showed the mode of the feature-value threshold value generation of the image processing by this invention in detail. It is explanatory drawing which showed the mode of the feature-value threshold value generation of the image processing by this invention in detail. It is explanatory drawing which showed the mode of the feature-value threshold value generation of the image processing by this invention in detail.

  DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments for carrying out the invention will be described with reference to the accompanying drawings.

[First Embodiment]
<Overall configuration>
FIG. 1 shows a configuration example of an image inspection apparatus capable of performing image processing employing the present invention. Functional blocks (FIG. 2), which will be described later, of the arithmetic unit that executes image processing may be configured by dedicated hardware, but for example, arithmetic control means such as a CPU (Central Processing Unit) as shown in FIG. Thus, each hardware of the image processing apparatus can be configured. The image processing apparatus in FIG. 1 can be configured using, for example, the configuration of a PC (personal computer), and corresponds to the hardware configuration of the arithmetic unit 1300 (FIG. 2) as a whole.

  The image processing apparatus shown in FIG. 1 includes a CPU 2100 and a memory 2200 connected via an internal bus 2000. The memory 2200 includes a storage area such as a RAM that stores RAM, firmware, a boot loader, a basic IO program, and the like used as a work area of the CPU 2100.

  In FIG. 1, the storage device 2300 is particularly illustrated, but this storage device 2300 corresponds to a hardware configuration of a storage unit 1400 (FIG. 2) described later. The storage area of the storage device 2300 includes, for example, an image data storage unit 1401 that stores acquired image data, which will be described later, a feature amount threshold storage unit 1402 that stores the calculated feature amount threshold, and a pass / fail judgment reference storage unit 1403 that stores the threshold. Configure. The storage device 2300 can store an image inspection (image processing) program, which will be described later, in a data format that the CPU 2100 can execute. The storage device 2300 can be configured by a so-called external storage device (whether built-in or detachable) such as HDD and SSD, various flash memory devices, or any combination of this type of storage device. .

  An image inspection program to be described later is configured in a data format that can be executed by the CPU 2100, for example, stored in the storage device 2300, loaded into the memory 2200 as necessary, and executed by the CPU 2100. As a result, image processing of an arithmetic unit 1300 (FIG. 2) described later is executed.

  The image processing apparatus of FIG. 1 includes an external connection device 2500 connected via an external communication port 2400. The externally connected device 2500 includes, for example, a camera 1200 (FIG. 2) that captures a later-described inspection object as a grayscale image, and a position adjustment mechanism 1100 (same) that adjusts the position of the inspection object. The external communication port 2400 is configured by an arbitrary general-purpose interface necessary for connection of the external connection device 2500 such as a USB interface or a GPIB interface.

  Among the externally connected devices 2500, the camera 1200 has, for example, a digital camera configuration, and captures an image of an object to be inspected for image inspection and an image of a non-defective sample used for generating a threshold for image inspection. Used for. The position adjustment mechanism 1100 is composed of an arbitrary transport mechanism such as a conveyor belt or a robot arm, for example, and is configured to be able to position an object to be inspected at a predetermined position in an imageable area in front of the camera 1200.

  Note that the image processing apparatus hardware in FIG. 1 is also used in second and third embodiments described later. In particular, in the second embodiment, the position adjustment mechanism 1100 perturbs the same inspection object in the imageable area in front of the camera 1200 and, for example, performs multiple imaging on the same good sample under different imaging conditions. Used for

  In this embodiment (the same applies to the second and third examples described later), the non-defective sample refers to a non-defective product that has been manufactured or assembled in the same manner as the inspected object and subjected to visual inspection. It is determined that. In addition, the non-defective sample herein includes the same optical characteristics as the object to be inspected, such as a jig or an original instrument, when the image is taken, but also includes a material or work accuracy that does not necessarily match the object to be inspected. The image inspection of the present embodiment (the same applies to the second and third examples described later) can be preferably performed on any of the above-mentioned non-defective samples. In the following, for the sake of simplicity, the abbreviation “non-defective product” is also used to indicate the above “non-defective sample”. In addition, the abbreviation “non-defective image” is also used to indicate “a gray-scale image obtained by capturing non-defective products”.

  In addition, the image processing apparatus in FIG. 1 includes a display unit 1500 and an input unit 1600 configured from a keyboard, an arbitrary pointing device, and the like. The display unit 1500 is composed of a display device composed of an LCD panel or the like. The display unit 1500 and the input unit 1600 constitute user interface means for an operator (inspector).

  FIG. 2 is a block diagram showing an arithmetic function when performing image inspection (appearance inspection, optical inspection) using the image processing apparatus hardware shown in FIG. The configuration of FIG. 2 may be considered to indicate the configuration of an image inspection (appearance inspection, optical inspection) apparatus using the hardware of FIG. The configuration of FIG. 2 may be considered as a block configuration of an image inspection (appearance inspection, optical inspection) method implemented using the hardware of FIG. In FIG. 2, the members (mainly hardware) already described in FIG. 1 are also shown, but these are denoted by the same reference numerals.

  2 includes a position adjustment mechanism 1100 that adjusts the position of the inspection object 1900 and a camera 1200 that captures the inspection object as a grayscale image. The functional configuration of FIG. 2 includes an arithmetic unit 1300 that performs arithmetic processing on image data, and a storage unit 1400 that stores image data or control data. The arithmetic unit 1300 and the storage unit 1400 are realized in hardware by the configuration shown in FIG.

  In the present embodiment, the calculation unit 1300 and the storage unit 1400 include the following calculation processing blocks. The image data acquisition unit 1301 acquires a grayscale image captured by the camera 1200 as image data. Specifically, the image data acquisition unit 1301 performs processing for transferring image data in a predetermined format input from the camera 1200 to the image data storage unit 1401. The image data storage unit 1401 stores the image data acquired via the image data acquisition unit 1301.

  The positioning processing unit 1311 adjusts the position of the inspection object 1900 in conjunction with the position adjustment mechanism 1100. The positioning processing unit 1311 and the position adjusting mechanism 1100 are configured to perform perturbation control of the inspection object 1900 used in the second embodiment described later. The functional block of this perturbation control is shown as a position perturbation unit 1312 in FIG.

  The first feature quantity selection receiving unit 1321 inputs a user (worker or inspector) selection operation via the input unit 1600. In the image processing according to the present embodiment, as a feature value of image data used in image inspection by threshold generation or threshold comparison, brightness, edge strength, high frequency (or low frequency) components, etc. can be considered in addition to the gray value. In the present embodiment, the user (worker or inspector) is configured to be able to specify which of these feature quantities is to be used via the feature quantity selection receiving unit 1321.

  The feature value threshold value calculation unit 1322 generates a threshold value for the first feature value designated via the first feature value selection reception unit 1321 for each pixel from a plurality of non-defective gray images. The feature value threshold value generated by the feature value threshold value calculation unit 1322 is stored in the feature value threshold value storage unit 1402.

  Further, in the image inspection of the inspection object, the first feature amount calculation unit 1331 is captured from the grayscale image of the inspection object captured by the camera 1200 and stored in the image data storage unit 1401 via the image data acquisition unit 1301. The first feature amount is calculated for each pixel.

  Then, the comparison extraction unit 1332 includes a feature amount threshold storage unit based on the first feature amount calculated from the gray image of the inspection object by the first feature amount calculation unit 1331 and image processing of a plurality of non-defective gray images. The feature amount threshold value stored in 1402 is compared for each pixel. As described above, the feature value threshold value is generated in advance by the feature value threshold value calculation unit 1322 based on image processing of a plurality of non-defective gray images, and is stored in the feature value threshold value storage unit 1402. Thereby, the comparison extraction unit 1332 can extract pixels that are outside the threshold as defective portions.

  The defect feature amount calculation unit 1333 calculates a defect feature amount from the defect portion extracted by the comparison extraction unit 1332. For example, the defect feature quantity calculation unit 1333 performs a labeling process and the like, and performs a process of analyzing the continuity of the pixel group corresponding to the defect part extracted by the comparison extraction unit 1332. Thereby, the defect feature amount calculation unit 1333 calculates, for example, a defect feature amount representing the defect portion, with a defect candidate pixel group that can be identified as a single defect portion as a unit. This feature quantity is basically the same as the feature quantity stored in the above-described feature quantity threshold value storage unit 1402 or the comparison extraction unit 1332 or the target of computation (in addition to the gray value, the luminance and edge strength). Etc.) are used.

  On the other hand, the pass / fail judgment reference storage unit 1403 stores the types of defect feature amounts that can be compared with the defect feature amounts calculated by the defect feature amount calculation unit 1333 and the threshold values for the types.

  Then, the determination unit 1334 compares the defect feature amount calculated by the feature amount threshold calculation unit 1322 with the corresponding type of defect feature amount threshold (quality determination criterion) stored in the quality determination criterion storage unit 1403. As a result, the determination unit 1334 can determine whether the inspection object 1900 is good or bad. The result of the quality determination can be output using the display unit 1500, for example. The display format of the pass / fail judgment result on the display unit 1500 is arbitrary. For example, the display unit 1500 may visually display a detected defective portion with a color or density different from that of other portions, or may output a threshold value for calculation or a feature amount of a pixel (group) by numerical display. It is done.

  Hereinafter, the operation of the present embodiment configured as described above will be described. In the present embodiment, in the above description, only the first feature amount (or feature amount threshold value) appears, but this is not limited to the first feature amount but the second feature amount (or feature amount). This is for the purpose of distinguishing from the configuration of a later-described third embodiment configured to use (threshold).

  FIG. 3 shows a flow of processing for generating a (first) feature amount threshold value that acts on detection of a defect feature amount as preparation for image inspection.

  In step S301 in FIG. 3, first, a plurality of non-defective products are imaged by the camera 1200. At this time, it is assumed that the non-defective product, which is an inspection object, has already been aligned by the position adjustment mechanism 1100 and the positioning processing unit 1311. A plurality of non-defective gray images are acquired by the image data acquisition unit 1301 and stored in the image data storage unit 1401. Note that the plurality of non-defective products used at this time are selected in advance by, for example, visual inspection by a skilled inspector.

  Here, FIG. 5 shows an image captured by the camera 1200 in a one-dimensional format (for example, along a certain scanning line). 5A and 5C, images captured by the camera 1200 are displayed in grayscale. 5B and 5D display images captured by the camera 1200 with the horizontal axis in the coordinate (space) direction on the scales corresponding to (A) and (C). Value sectional view).

  5A and 5B are examples of non-defective gray images captured in the generation of the (first) feature amount threshold value in FIG. Further, FIGS. 5C and 5D described later are examples of gray images of the inspection object captured in the image inspection.

  FIG. 5A is an example of a non-defective gray image, and there is a background 510a which is disturbance information such as the outer shape of the inspection object and the reflection of illumination (hereinafter, the outer shape of the inspection object for simplification). Disturbance information such as lighting reflections and lighting are collectively referred to as “background”). A pixel at coordinates 501 is a pixel located in the background 510a, and pixels at coordinates 502, 503, and 504 are pixels located near the background 510a. The pixel at the coordinate 505 is a pixel for which the background 510a does not exist. FIG. 5 (B) 520 represents a gray value cross section of FIG. 5 (A).

  Next, in step S302 in FIG. 3, the first feature quantity selection receiving unit 1321 receives selection information input by the user (operator or inspector) through the input unit 1600 as the first feature quantity that emphasizes the defective part. Accept. As the first feature value selected here, a feature value suitable for emphasizing a defective portion may be arbitrarily set. For example, a high frequency component calculated using an FIR filter may be used. Hereinafter, for simplification, a case where a gray value is selected as the first feature amount will be described.

  Next, in step S303, the feature value threshold value calculation unit 1322 calculates the feature value threshold value to be stored in the image data storage unit 1401 for each pixel from the plurality of non-defective gray images acquired in step S301 in advance. The feature amount threshold value calculated here is stored in the feature amount threshold storage unit 1402. The image processing for generating the feature amount threshold value in step S303 will be described in detail later.

  In step S304, the defect feature quantity type and threshold value for determining whether or not the extracted defect part is within the reference are set and stored in the pass / fail judgment reference storage unit 1403. The type and threshold value of the defect feature amount stored here is, for example, that a defective portion having an area of 20 pixels or more is regarded as defective, but is not limited thereto. You may set arbitrarily what is suitable for every to-be-inspected object. Thus, the preparation for starting the image inspection is completed.

<Detailed Description of Step S303>
FIG. 11 shows a flow of image processing for generating a feature amount threshold value in step S303 of FIG.

  In the image processing of FIG. 11, first, in step S3031, a first feature amount is calculated for each pixel from a plurality of non-defective gray images. Here, as described above, the gray value is used as the first feature amount.

  Next, in step S3032, a parameter used when calculating a feature amount threshold value is acquired for each pixel from the first feature amount calculated from a plurality of non-defective images. At this time, in the plurality of non-defective images, not only the first feature amount calculated from the pixel having the same coordinate as the target pixel but also the first feature amount calculated from the pixels within the alignment range existing around the target feature amount. Consider. This alignment range is set according to the positioning accuracy of the apparatus and the resolution of the camera.

  Next, a statistic calculation is performed based on a preset formula using the parameters acquired for each pixel in step S3032, and a feature amount threshold value is calculated for each pixel (step S3033).

  As described above, the first value obtained for each pixel by the statistic calculation on the pixel value of the pixel at the same coordinate of the image data of the non-defective sample and the pixel value of the pixel within the positioning tolerance range of the pixel at the same coordinate. A feature amount threshold value is generated for each pixel based on the feature amount statistics. Hereinafter, in the above step S3033, some examples of calculation formulas used for the statistic calculation for generating the feature amount threshold are exemplified.

<Example 1 of calculation formula for calculating the feature amount threshold>
The feature amount threshold P can be calculated by the following equation (1) using, for example, the average value M and the standard deviation σ as parameters.

式（１）の演算によって、画素ごとの第１の特徴量に対する上限閾値を設定することができる。このときｋは０以上の整数であれば適当な数でよいが、統計上、ｋ≧３とすることが望ましい。

The upper limit threshold for the first feature value for each pixel can be set by the calculation of Expression (1). At this time, k may be an appropriate number as long as it is an integer of 0 or more, but it is desirable that k ≧ 3 in terms of statistics.

  Through the above calculation, for example, a threshold value 530 as shown in FIG. 5B can be acquired for each pixel from the image in FIG. This calculation will be described in the following specific example 1.

<Specific example 1>
FIG. 12 illustrates a specific calculation example. 12A is an image example of a non-defective product (A), and FIG. 12B is an image example of another good product (B), both of which are pixel values (for example, gray values) corresponding to the first feature amount. ) Is shown in the form of a grid in which 1 桝 corresponds to one pixel. In the following, for simplification, an example in which the feature amount threshold value P is calculated for each pixel from the two non-defective images in FIGS. 12A and 12B will be described. Here, the first feature amount is a pixel value (for example, a gray value). Further, the alignment range of the object to be inspected is, for example, a range of one pixel at a chessboard distance from the target pixel. It should be noted that the range of one pixel at this chessboard distance is a range that surrounds the pixel of interest with a distance of one pixel in the grid in the figure, and a total of nine pixels including a diagonal distance of one pixel.

  As a calculation formula for calculating the threshold value P, for example, the following formula (2) using only the average value M as a parameter may be used.

  In FIGS. 12A and 12B, calculation for calculating the threshold value P is considered when the pixel of interest is the pixel of interest 131a and 131b in the fourth row and the fifth column from the upper left.

  In FIG. 12A, the nine pixel values of the pixel of interest 131a and the pixel 132a within the alignment range surrounding it are (0, 1, 1, 1) from left to right and from top to bottom. , 2, 2, 2, 3, 3).

  In FIG. 12B, the pixel values of the target pixel 131b and the pixel 132b within the alignment range are (2, 2, 1, 3, 3, 2, 3, 3, 2) in order from the upper left. Yes.

  Since the average value of the pixel values of the 18 pixels in FIGS. 12A and 12B is 2, the threshold value P of the pixel of interest 131c is 2 from Expression (2). The result of performing this operation for all pixels is as shown in FIG. FIG. 12C shows the value of the feature amount threshold value P calculated for each pixel by the above equation (2).

<Example 2 of calculation formula for calculating the feature amount threshold value>
Note that the formula for calculating the feature amount threshold P is not limited to the above formulas (1) and (2). For example, the average value M and the standard deviation σ may be used as parameters, and may be calculated by the following equation (3).

この式（３）によれば、画素ごとの第１の特徴量に対する上下限閾値を設定することができる。

According to this equation (3), the upper and lower thresholds for the first feature value for each pixel can be set.

<Example 3 of calculation formula for calculating the feature amount threshold value>
Alternatively, the feature amount threshold value P may be calculated as in the following equation (4), for example, using the maximum value as a parameter.

この式（４）において、Ａは、位置合わせ範囲内の画素から算出された第１の特徴量の集合である。これは式（１）においてｋを大きくした場合は最大値で近似できる性質を利用したものであり、暗視野検査など上限閾値のみを設定する場合に有効である。

In Expression (4), A is a set of first feature values calculated from pixels within the alignment range. This utilizes the property that can be approximated by the maximum value when k is increased in Equation (1), and is effective when only the upper threshold is set, such as in dark field inspection.

<Specific example 2>
A specific calculation example when the feature amount threshold value P is generated using the above equation (4) will be described with reference to FIG. 13A shows an image example of the non-defective product A, and FIG. 13B shows another image example of the non-defective product B. Each of the pixel values (for example, the gray value) corresponding to the first feature amount is 1 桝. Is shown in a grid format corresponding to one pixel. In the following, for simplification, an example in which the first feature amount threshold value P is calculated for each pixel from the two non-defective images in FIGS. 13A and 13B will be described. Here, the first feature amount is a pixel value (for example, a gray value). In addition, the alignment range of the object to be inspected is, for example, a range of one pixel from the target pixel to the city distance. Note that the range of one pixel at this urban distance is a range of five cross-shaped pixels that do not include one pixel in the oblique direction, and is a range that surrounds the target pixel at a distance of one pixel in the grid in the figure.

  In FIG. 13, a calculation for calculating the threshold value P is considered when the pixel of interest is the pixel of interest 141a in the fourth row and the fifth column from the upper left.

  In FIG. 13A, the pixel values of the target pixel 141a and the in-alignment range pixel 142a are (1, 1, 2, 2, 3) from top to bottom and from left to right.

  In FIG. 12B, pixel values of the target pixel 141b and the in-alignment range pixel 142b are (0, 0, 0, 1, 1) from top to bottom and from left to right. Yes.

  Since the maximum value of the pixel values of the 10 pixels in FIGS. 13A and 13B is 3, the threshold value P of the pixel of interest 141c is 3 from Equation (4). The result of performing this operation for all pixels is as shown in FIG.

<Another Form of Example 3 of Calculation Formula for Calculating Feature Threshold Value>
When the feature amount threshold P is set to the maximum value of the first feature amount according to the equation (4), it can be calculated by the maximum value synthesis process and the morphology process for each pixel. The maximum value composition processing is processing for comparing pixel values at the same coordinates between a plurality of images and outputting the maximum value.

  Morphology processing is one of image processing methods in which a value based on a pixel of interest and a comparison target pixel in the vicinity thereof is output. The positional relationship between the target pixel and the comparison target pixel in the vicinity thereof is called a structuring element. For example, if the maximum value in the structuring element is output, multi-value expansion processing is performed, and if the minimum value is output, multi-value contraction processing is performed. By processing such pixel values, the calculation of Expression (4) can be realized with a small calculation cost (or calculation resource).

<Specific example 3>
14 uses the same two non-defective images (FIGS. 14A and 14B) as in FIGS. 13A and 13B, and the feature amount threshold value by the maximum value synthesis process and the morphology process for each pixel. A specific calculation example when P is generated will be described. 13A shows an image example of the non-defective product A, and FIG. 13B shows another image example of the non-defective product B. Each of the pixel values (for example, the gray value) corresponding to the first feature amount is 1 桝. Is shown in a grid format corresponding to one pixel. Here, as in the case of FIG. 13, the first feature value threshold P is obtained from the two non-defective images in FIGS. 14A and 14B (the same as those in FIGS. 13A and 13B). An example of calculating for each pixel is shown. Here, the first feature amount is a pixel value (for example, a gray value). Similarly to the case of FIG. 13, the alignment range is a range of one pixel from the target pixel to the city distance, and the calculation formula for calculating the threshold value P is the above formula (4).

  Here, first, an image shown in FIG. 14C is created by synthesizing the maximum values of the good image (A) and the good image (B) for each pixel. For example, when attention is paid to the pixel in the fourth row and the fifth column from the upper left, since the maximum value of the pixel values of the pixels 151a and 151b is 2, 2 is output as the pixel value of the pixel 151c.

  Next, morphology processing is performed on the image of FIG. The structuring element at this time is set according to the alignment range. In this case, as shown in FIG. 14E, the comparison target pixel 154e is set for the target pixel 153e. The comparison target pixel 154e is four pixels in the range of one pixel at a city distance with respect to the target pixel 153e. Here, for example, when the pixel 151c is the target pixel of the structuring element, the surrounding range 152c is a pixel to be compared. When the morphological process is performed using this structuring element and the maximum value is output, a threshold distribution as shown in FIG. 14D is obtained, and the threshold value “3” is output to the pixel 151d. The calculation result of FIG. 14D is equivalent to that of FIG.

  In the above, for the sake of simplification, an example in which one-stage morphological process is performed using one structuring element has been shown, but the contents of the morphological process need not be a single multi-value expansion process. Further, a plurality of structured elements and a plurality of morphological processes under different conditions may be combined according to the alignment range.

<Reduction of the number of non-defective images necessary for generating the feature threshold value>
Now, the first feature value calculated from the pixels within the alignment range can be considered as an approximate value of the first feature value calculated from the pixel of the same coordinate in another non-defective product image. This will be described with reference to FIG. In the following, the alignment range is set to ± 1 pixel for simplification.

  FIG. 6A shows a grayscale cross section of a good product image by taking a one-dimensional format on one scanning line in the spatial direction (horizontal axis), for example. Here, coordinates 503 near the center of the horizontal axis and coordinates 502 and 504 corresponding to the positions of the neighboring pixels are shown. 6B is a gray value cross section obtained by shifting one pixel in the right direction in FIG. 6A, and FIG. 6C is a gray value obtained by shifting one pixel in the left direction in FIG. 6A. It is a cross section. Here, the gray value calculated from the pixels of the coordinates 502 and 504 located within the alignment tolerance of the coordinate 503 is considered as an approximate value of the gray value calculated from the pixel of the coordinate 503 in another good quality gray image. Can do.

  This is because the non-defective product in FIG. 6A may be imaged as shown in FIG. 6B or FIG. 6C due to the alignment tolerance. That is, by using the first feature amount calculated from the pixels within the alignment tolerance as an approximate value, a plurality of values are acquired as the first feature amount that can be calculated from a certain pixel from a single good image. be able to. As a result, the statistics for each pixel relating to the first feature amount can be given the same level of reliability as in the past. For example, the number of grayscale images that must be read to generate a threshold value for each pixel is more than in the conventional method. Can be significantly reduced.

  FIG. 4 shows a flow of a control procedure of image inspection using the feature amount threshold generated as described above. In the process of FIG. 4, it is assumed that the feature amount threshold value storage unit 1402 stores the feature amount threshold value generated in advance by the generation processing described with reference to FIGS. 3 and 11 to 14.

  In step S401 of FIG. 4, after alignment, the inspection object is imaged by the camera 1200, and a grayscale image of the inspection object is acquired via the image data acquisition unit 1301. In step S402, the first feature amount calculation unit 1331 calculates a first feature amount for each pixel from the grayscale image of the inspection object.

  Next, in step S403, the threshold value stored in the feature value threshold value storage unit 1402 by the comparison extraction unit 1332 and the first feature value of the inspected object calculated by the first feature value calculation unit 1331 for each pixel. Compare. As a result, pixels that are outside the threshold are extracted as defective portions (defective pixels).

  Here, FIG. 5C shows an example of a grayscale image of the inspection object. In this image, a background 510c and a defective portion 511c exist. FIG. 5D is a one-dimensional view of the gray value cross section of the inspection object.

  FIGS. 5C and 5D show the feature value threshold value 530 (calculated in step S303 with respect to the cross-section 540 of the gray value calculated for each pixel as the first feature value in step S402 in FIG. It may be considered that the display is displayed in a superimposed manner. In this example, the portion of the background 510d in FIG. 5D is not extracted because it is within the threshold, but the defective portion 511d is outside the threshold and can be extracted as a defective portion. Specifically, the pixel at the coordinate 505 corresponding to the defective portion 511d is determined to be a defective pixel by comparing the feature amount and the feature amount threshold value 530 (step S403).

  In step S404 and subsequent steps in FIG. 4, the defective part extracted as described above is analyzed using the pass / fail judgment criteria stored in the pass / fail judgment reference storage unit 1403 in advance, and the inspection object (the whole) is non-defective. Or, it is determined whether it is a defective product.

  In step S <b> 404, the defect feature quantity calculation unit 1333 calculates the type of defect feature quantity stored in the pass / fail judgment reference storage unit 1403 for each defect part extracted by the comparison extraction unit 1332. For example, a defective part can be considered as a set of pixels determined to be a defective part, and a defect feature amount representative of the defective part can be obtained and used for comparison with a quality criterion. As the defect feature amount, for example, the area of the defect portion can be used. The area of the defective portion is obtained, for example, by performing a labeling process on a pixel determined as a defective pixel by threshold determination, identifying a spatially continuous defective portion, and determining the number of pixels in the defective portion as a unit. be able to.

  In step S405, the determination unit 1334 compares the defect feature amount calculated by the defect feature amount calculation unit 1333 for each defective portion with the reference stored in the pass / fail determination reference storage unit 1403 to determine the presence or absence of a fatal defect. . Then, if it is within the reference, the inspection object is determined to be a non-defective product (step S406a), and if there is a defective part outside the reference, it is determined to be a defective product (step S406b).

  The image inspection result can be displayed (printed) and output using a display unit 1500 (or a printer (not shown), for example). In this case, for example, information indicating that the inspection object is either a non-defective product or a defective product is output according to the determination results of steps S405, S406a, and S406b. Similarly, the defective portion detected in step S403 can be displayed (printed) and output using, for example, the display unit 1500 (or a printer (not shown)). In that case, a technique can be used in which a defective part (pixel) is visually displayed with a color or density different from that of other parts, or a threshold value for calculation or a feature value of a pixel (group) is output by numerical display. .

  As described above, according to the present embodiment, the first feature amount calculated from the pixels within the alignment tolerance of the same non-defective image is, for example, the first feature amount calculated from the pixel at the same coordinate of another non-defective image. A statistical amount related to the first feature value is calculated for each pixel by using the approximate value of the feature value. And the feature-value threshold value for comparing with the 1st feature-value calculated | required from the actual to-be-inspected object can be produced | generated using the statistics regarding this 1st feature-value. That is, a plurality of values that can be calculated from the pixel for the first feature amount can be acquired from one good product image, and a feature amount threshold value can be generated based on the acquired value.

  For example, when generating a feature amount threshold value to be generated for a certain pixel, it is necessary to use a statistical amount of a plurality of non-defective images in the related art as in Patent Document 1. In contrast, in the present embodiment, by using the first feature amount calculated from the pixels within the alignment tolerance, the same effect as that obtained by collecting statistics of a plurality of non-defective images can be obtained. That is, according to the present embodiment, by using the first feature amount calculated from the pixels within the alignment tolerance, the first feature amount of a certain pixel, or the feature amount threshold generated based on the first feature amount, is the same as the conventional feature amount. A degree of reliability can be imparted.

  For this reason, it is possible to remarkably reduce the number of non-defective gray images necessary for generating a feature amount threshold value for each pixel having the same level of reliability as in the past. That is, according to the present embodiment, when generating a threshold value for performing defect determination for each pixel from a plurality of non-defective samples for image inspection, a large amount of non-defective samples is not required, and only a relatively small amount of non-defective samples is obtained. Threshold generation can be performed.

  In addition, the image inspection method of this embodiment thru | or each below-mentioned embodiment can be implemented in production lines, such as an industrial product, for example. For example, an apparatus for performing an image inspection method according to the present embodiment or each of the embodiments described later can be disposed in the middle or at the end of a production line for industrial products or the like. Then, the production line can be controlled so that only inspected objects determined as having no defect by the image inspection method of the present embodiment or each of the embodiments described later are sent to a downstream assembly process or shipping process.

[Second Embodiment]
In the first embodiment, for example, (n) non-defective images are acquired from a plurality of (for example, n) non-defective products, and a feature amount threshold value is generated from the acquired (n) images. However, it is also conceivable to obtain a plurality (n) of images necessary for generating the feature amount threshold value from the same non-defective product. In the present embodiment, a configuration example in which the number of non-defective products to be prepared for image inspection is reduced by acquiring a plurality (n) of non-defective images from the same non-defective product through the perturbation control by the position perturbation unit 1312. explain.

  The hardware of this embodiment may be the same as that of FIG. 1 shown in the first embodiment. Further, in this embodiment, the block configuration of the calculation function when performing image inspection (appearance inspection, optical inspection) using the image processing apparatus hardware shown in FIG. 1 is the same as that in FIG. 2 shown in the first embodiment. It can be the same.

  However, in the present embodiment, the position perturbation unit 1312 that is not used in the first embodiment performs position perturbation of the inspection object, and generates a threshold value for each pixel. The difference from the first embodiment is that, after alignment of the inspection object 1900, imaging is performed by the camera 1200 while the inspection object 1900 is displaced to an arbitrary position within the tolerance range.

  FIG. 7 shows a flow of processing for generating a (first) feature amount threshold value that acts on detection of a defect feature amount as preparation for image inspection in the present embodiment. The feature amount threshold value generation process of FIG. 7 corresponds to FIG. 3 of the first embodiment.

  The difference from the process of FIG. 3 in the first embodiment is that when acquiring a plurality of non-defective product image data in image inspection preparation, a plurality of non-defective products are not actually prepared and read, but the same non-defective product is read once or twice. The point is that two or more images are taken from the same non-defective product by being perturbed several times.

  That is, in step S311 of FIG. 7, the same non-defective product is perturbed and displaced by 1 to several degrees at an arbitrary position within the alignment tolerance by the position perturbation unit 1312. Get an image. For example, the alignment tolerance is set within a range of one pixel (or several pixels). Then, after the initial alignment and imaging, if the position perturbation unit 1312 perturbs by an appropriate distance (one to several pixels) along the orthogonal coordinate axis within the range, up, down, left and right (east, west, north, and south), the same Four more images can be acquired from non-defective products.

  Processing steps other than step S311 in FIG. 7 can be performed in the same manner as in FIG. 3, as indicated by the same step numbers as in FIG. For example, in particular, in the feature amount threshold value generation in step S303, all the calculation methods described in the expressions (1) to (4) and FIGS. 11 to 14 in the first embodiment can be used. In that case, a plurality of (n sheets) images acquired from the same non-defective product via the perturbation control of the position perturbation unit 1312 as described above are described in the first embodiment (a plurality of non-defective products respectively acquired from different non-defective products). The above calculation method may be used instead of the image. Then, the actual image inspection of the inspection object may use the processing procedure shown in the flowchart of FIG.

  According to the present embodiment, a plurality of non-defective images are acquired from the same non-defective product via the perturbation control of the position perturbation unit 1312, and a (first) feature amount threshold value that acts on detection of a defect feature amount is generated. Can do. Any of the methods described in the first embodiment can be used for the feature amount threshold generation processing.

  For example, in the present embodiment, for example, n non-defective images can be acquired from the same non-defective product through the perturbation control by the position perturbation unit 1312 for the feature amount threshold. On the other hand, if it is assumed that the arithmetic processing of the first embodiment prepares n non-defective products for n non-defective images, according to the present embodiment, only 1 / n non-defective products are prepared. Thus, it is possible to generate a feature amount threshold value having substantially the same reliability as that of the first embodiment.

  In this embodiment, the inspection object is actually perturbed within the alignment tolerance through the perturbation control by the position perturbation unit 1312, and an image is acquired from the inspection object at that position. For this reason, even when the same non-defective product is used, a plurality of non-defective images that reflect changes in the illumination direction and the positional relationship with the camera that occur when the position of the object to be inspected are actually obtained are acquired. It is possible to generate a (first) feature amount threshold value that affects detection.

  For this reason, according to the present embodiment, it is possible to calculate, for each pixel, a statistic including the influence of the change in the first feature amount calculated from the pixel at the same coordinate due to the displacement of the imaging position. As a result, it is possible to set a feature amount threshold corresponding to the displacement of the imaging position, and an effect of improving the threshold setting accuracy can be obtained.

  For example, the control of the present embodiment is subject to an inspection of a shape that is easily affected by changes in the illumination direction or the positional relationship with the camera caused by the displacement of the inspection object (for example, having a large number of deep irregularities on the imaged surface) There is a possibility that it can be suitably performed in image inspection of objects.

[Third Embodiment]
In the first and second embodiments, a first feature amount is acquired from a plurality of non-defective images, and a feature amount threshold value is generated for each pixel by calculation using the acquired first feature amount. It is also conceivable that this feature value threshold value is obtained separately from the image data of the object to be inspected, and the second feature value is used to correct the feature value to a more appropriate value for the purpose of the image inspection. It is done. In the present embodiment, such a feature amount threshold value correction technique will be described.

  The hardware of this embodiment may be the same as that of FIG. 1 shown in the first embodiment. FIG. 8 shows a block configuration of a calculation function when performing image inspection (appearance inspection, optical inspection) using the image processing apparatus hardware shown in FIG. 1 in the present embodiment. However, although the position perturbation unit 1312 used in the second embodiment is illustrated in FIG. 8 in the same manner as in FIG. 2 of the first embodiment, in this embodiment, this position is shown in this embodiment for ease of explanation. The image inspection control performed without using the perturbation control by the perturbation unit 1312 will be described.

  In FIG. 8, the difference from the first and second embodiments is that a second feature amount calculation unit 1323 that calculates a second feature amount for each pixel from the grayscale image of the object to be inspected, and a second feature amount. And a feature amount threshold value correcting unit 1324 that corrects the feature amount threshold value. Other functional blocks may be the same as those shown in FIG. 2 as the same reference numerals are used.

  Hereinafter, the control of this embodiment will be described with reference to FIGS. 9 and 10. FIG. 9 shows an image inspection procedure of the inspection object in the present embodiment in the same flowchart format as FIG. FIG. 10 schematically shows the structure of image data processed in the present embodiment.

  The difference between the functional configuration of FIG. 8 and the control procedure of FIG. 9 from the operations of the first and second embodiments is that the second feature amount calculation unit 1323 first uses the second feature from the grayscale image of the inspection object. The amount is calculated for each pixel (step S411). Second, the feature value threshold correction unit 1324 corrects the first feature value threshold value using the second feature value (step S412). Other processing steps can be performed in the same manner as in FIG. 4 as indicated by the same step number.

  In the first and second embodiments, in order to prevent erroneous background extraction, which is disturbance information such as the outer shape of the object to be inspected and the reflection of illumination, there is a tendency that the feature amount threshold is set excessively in the pixels near the background. is there. As a result, the threshold setting accuracy is low in pixels near the background, and the defect extraction sensitivity may be low.

  For example, FIG. 10A shows an example of a grayscale image of an inspection object. In the image of FIG. 10A, a background 110a and a defective portion 111a exist. Here, based on the case where the edge strength is selected as the first feature amount, the above-described problem relating to the defect extraction sensitivity will be considered.

  FIG. 10B displays the feature amount threshold 130 calculated in step S303 of the first embodiment, for example, overlaid on the cross-section of the edge strength calculated in step S402 from the image of FIG. 10A. Is. In FIG. 10B, reference numeral 120 indicates the waveform of the edge intensity signal acquired as the first feature amount from the inspection object.

  As shown in FIG. 10B, the waveform of the feature amount threshold 130 has a threshold higher than the peak of the background 110b with respect to the pixels of the neighboring coordinates 102, 103, and 104 of the background 110a. Therefore, with such a feature amount threshold value 130 (waveform thereof), erroneous extraction of the background 110b can be suppressed, but it is difficult to extract the defective portion 111b.

  On the other hand, according to the present embodiment, first, the second feature amount is calculated for each pixel from the grayscale image of the inspection object (step S411 in FIG. 9). As the second feature amount, in the present embodiment, for example, information on a low frequency component of a spatial frequency is used. However, the second feature amount can be obtained from the image data of the inspection object, and the feature amount has such a property that a pixel having a background has a large value and a pixel having a defective portion has a small value. For example, any other feature amount can be used.

  In this embodiment, the low-frequency component information to be acquired as the second feature amount from the grayscale image of the inspection object is obtained by, for example, passing a pixel value (grayscale value, edge strength, etc.) in the scanning line direction through a low-pass filter. can get. The pass characteristics of the low-pass filter may be appropriately determined in advance according to the characteristics of the background caused by the characteristics of the object to be inspected, the camera, and the illumination system. In addition, the low frequency component information to be acquired as the second feature amount can be acquired for each pixel from the coordinate position corresponding to the original pixel in the image signal after passing through the low-pass filter.

  FIG. 10C shows a cross section of a low-frequency component calculated from the grayscale image of the inspection object shown in FIG. In FIG. 10A, the defective portion 111a has a smaller area than the background 110a. For this reason, the pixel of the coordinate 104 located near the background 110a and having the defective portion 111a has a lower frequency component than the pixels of the coordinates 102 and 103 having the background 110a as shown in FIG. 10C.

  Using the low-frequency component information as shown in FIG. 10C, the feature amount threshold 130 in FIG. 10B can be corrected to the waveform feature amount threshold 140 as shown in FIG. 10D. In FIG. 10D, reference numeral 120 denotes the waveform of the edge intensity signal acquired as the first feature amount from the inspection object, as in FIG.

  In FIG. 10D, the two edge intensity peak portions of the background 110d are covered below the waveform of the corrected feature amount threshold value 140. On the other hand, in the edge intensity peak portion of the defect portion 111d, the value of the feature amount threshold value 140 is corrected to be lower than that in FIG. 10B based on the low-frequency component information, and thus the defect portion 111d is correctly extracted as a defect. Can do.

  Image inspection preparation of this embodiment is performed as shown in FIG. 3 described in the first embodiment. In the process of FIG. 3, as described above, a feature amount threshold value is acquired for each pixel from the non-defective image in step S303 and stored in the feature amount threshold value storage unit 1402.

  On the other hand, as described above with reference to FIG. 10, the second feature value for correcting the feature value threshold value needs to be acquired from the image of the inspection object actually inspected. Therefore, in the image inspection process of FIG. 9, in step S411, for example, low frequency component information is acquired as the second feature amount from the grayscale image data of the inspection object.

  In step S412, the feature value threshold value stored for each pixel in the feature value threshold value storage unit 1402 is corrected using the second feature value. This correction is performed on the stored feature amount threshold value for each pixel. For the correction calculation, for example, a calculation such as Expression (5) can be used.

  In Expression (5), Ta is a threshold value after correction, Tb is a threshold value before correction, and a correction amount by which a portion of (L × s1) is multiplied by Tb. L is the second feature amount acquired as described above, and is a pixel value corresponding to the low-frequency component of the pixel in this embodiment. Further, s1 is an arbitrary constant, and in the present embodiment, it is preferable to use the reciprocal of the gradation number of the grayscale image in order to scale the low-frequency component L from 0 to 1. The other processing steps in FIG. 9 are the same as the processing steps having the same step numbers described in FIG. 4, and thus detailed description thereof is omitted here.

  As described above, in the image inspection processing, for example, low-frequency component information is acquired as the second feature amount from the image data of the object to be inspected, and is acquired for each pixel from the non-defective product according to the value and stored. The feature amount threshold value can be corrected. As a result, it is possible to prevent erroneous extraction of the background, which is disturbance information such as the outer shape of the object to be inspected and the reflection of illumination, and to perform reliable defect (good or bad) determination.

  As shown in FIG. 10D, the two edge intensity peak portions of the background 110d are covered below the corrected feature amount threshold value 140 waveform by the processing of the present embodiment, and the non-defective feature portions are not erroneously extracted. Can be detected as On the other hand, at the edge intensity peak portion of the defective portion 111d, the value of the feature amount threshold value 140 is corrected to be low, and the defective portion 111d can be correctly extracted as a defect.

  As described above, according to the present embodiment, by using the second feature amount calculated from the object to be inspected, the feature amount threshold set for the pixels near the background is set to a value different from that of the pixels in which the background exists. It can be corrected. Thereby, according to this embodiment, the threshold setting accuracy in the pixels near the background is improved, the defect extraction sensitivity is improved, and more accurate image inspection is possible.

  Although the three embodiments have been described above, the image inspection technique according to the present invention can be applied to an arbitrary image inspection apparatus as an image inspection method for imaging an arbitrary industrial product and determining a defect (good or bad) from the captured image. Can be implemented. It is executed by the CPU 2100 of the image inspection (processing) apparatus that performs image processing and defect determination in each of the above embodiments. Therefore, it can be configured to be achieved by supplying a recording medium recording a software program for realizing the above-described functions to the image inspection apparatus, and reading and executing the program stored in the recording medium by the CPU 2100. In this case, the program itself read from the recording medium realizes the functions of the above-described embodiment, and the program itself or the recording medium recording the program constitutes the present invention.

  In each of the above embodiments, the computer-readable recording medium has been described as the storage device 2300 or other external storage device (not shown). However, the present invention is not limited to such a form. Absent. The program for carrying out the present invention may be recorded on any recording medium as long as it is a computer-readable recording medium. For example, as a recording medium for supplying the program, it goes without saying that any external storage device such as a removable HDD, SSD, optical disk, or various flash memory devices can be used.

1200 ... Camera, 1400 ... Storage unit, 1401 ... Image data storage unit, 1403 ... Pass / fail judgment reference storage unit, 1500 ... Display unit, 1600 ... Input unit, 1900 ... Inspected object, 2000 ... Internal bus, 2100 ... CPU, 2200 ... Memory, 2300 ... Storage device, 2400 ... External communication port, 2500 ... Externally connected device

Claims (16)

In the image inspection method for acquiring the image data of the inspection object and determining the defect of the inspection object,
Based on the first feature amount acquired by the statistic calculation for the pixel value of the pixel at the same coordinate in the image data of the plurality of non-defective samples and the pixel value of the plurality of pixels located around the pixel at the same coordinate , A threshold generation step of generating a feature amount threshold for each ;
A correction step of acquiring image data of the inspection object and correcting the feature amount threshold generated for each pixel in the threshold generation step based on a second feature amount acquired for each pixel;
A defect extraction step of comparing the first feature amount acquired from the image data of the inspection object and the feature amount threshold value corrected in the correction step to extract a pixel exceeding the threshold value as a defect site;
A defect determination step of determining the presence / absence of a defect in the inspection object based on a characteristic of a defective portion formed by the pixel extracted by the defect extraction step. 2. The image inspection method according to claim 1, wherein in the threshold generation step, the same non-defective sample is perturbed within a positioning tolerance range when acquiring the image data, and the acquired same non-defective sample is obtained every time the perturbation occurs. An image inspection method using the image data as image data of the plurality of non-defective samples. 2. The image inspection method according to claim 1, wherein a plurality of pixels positioned around the pixel having the same coordinate in the threshold generation step are pixels within a positioning tolerance range when acquiring the image data. An image inspection method characterized by the above.   4. The image inspection method according to claim 1, wherein a feature quantity of a type selected according to an inspector's selection operation is used as the first feature quantity. 5. . 5. The image inspection method according to claim 1, wherein the statistic calculation is performed around a pixel value of a pixel at the same coordinate in the image data of the plurality of non-defective samples and a pixel around the pixel at the same coordinate. An image inspection method, which is multi-value expansion processing or multi-value contraction processing for pixel values of a plurality of pixels. 6. The image inspection method according to claim 1, wherein the statistic calculation is performed around a pixel value of a pixel at the same coordinate in the image data of the plurality of non-defective samples and a pixel around the pixel at the same coordinate. An image inspection method comprising: a maximum value synthesis process and a morphology process for pixel values of a plurality of pixels.   The image inspection method according to any one of claims 1 to 6, wherein the first feature amount is a gray value of a pixel.   The image inspection method according to claim 1, wherein the first feature amount is information on edge strength calculated for each pixel. 7. The image inspection method according to claim 1 , wherein the second feature amount is information of a low-frequency component calculated for each pixel. 8.   The image inspection program for making a control apparatus perform each process of the image inspection method of any one of Claim 1 to 9.   A computer-readable recording medium storing the image inspection program according to claim 10.   An inspection process for inspecting an inspection object by the image inspection method according to claim 1, wherein the inspection object determined as having no defects in the defect determination process is a downstream assembly process or shipping. A production method characterized by being sent to a process. In the image inspection apparatus for acquiring the image data of the inspection object and determining the defect of the inspection object,
Imaging means for obtaining image data of the imaging object;
The first obtained by a statistic calculation on the pixel value of the pixel of the same coordinate of the image data of the plurality of non-defective samples imaged by the imaging means and the pixel value of the pixel located around the pixel of the same coordinate based on the feature quantity, and a control unit for generating a feature amount threshold for each pixel,
A storage unit for storing the feature amount threshold,
Wherein the control device, from the image data of the object imaged by the imaging means, on the basis of the second feature amount acquired for each pixel, corrects the feature amount threshold value for each pixel stored in the storage unit Then, the first feature amount acquired from the image data of the object to be inspected is compared with the corrected feature amount threshold value, and a pixel exceeding the threshold value is extracted as a defective portion, and the defect formed by the extracted pixel An image inspection apparatus for determining the presence or absence of a defect of the inspection object based on a characteristic of a part. 14. The image inspection apparatus according to claim 13, wherein the control device acquires a gray value of a pixel as the first feature amount. The image inspection apparatus according to claim 13, wherein the control device acquires edge intensity information calculated for each pixel as the first feature amount. 16. The image inspection apparatus according to claim 13, wherein the control device acquires information on a low frequency component calculated for each pixel as the second feature amount. apparatus.

Images