JPH0789054B2 - Product defect inspection method and equipment - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for inspecting a product for defects using image processing technology.

[0002]

[Prior Art] Published by Nippon Kogyo Publishing Co., Ltd. "Image Lab"
In the November 1990 issue, pp. 56-63, the product to be inspected is photographed by a CCD camera, and the grayscale value of, for example, 256 gradations per pixel with a resolution of, for example, 512 pixels in the horizontal direction and 448 pixels in the vertical direction. As a method, the quality of a product is inspected by storing the image in an image memory and then image-processing the stored image by a density morphology method of expansion and contraction of density.

The algorithm of this method is summarized as follows (1) to (5). (1) For the original image data of one screen stored in the image memory, the maximum gradation value (the brightest gradation value) of the predetermined number of pixels in the vertical and horizontal directions is determined, The expansion process of replacing the gradation value of each pixel with the obtained maximum gradation value for all pixels while sequentially shifting the determination pixel area. By this process, pixels that are brighter than the surroundings expand, while pixels that are darker expand. (2) Regarding the image data after the shadow contraction processing, the smallest gradation value in the judgment pixel area is calculated, and the gradation value of each pixel in the judgment pixel area is replaced with the calculated minimum gradation value. This is a contraction process that is performed for all pixels while sequentially shifting the determination pixel area. In this process, contrary to the above process (1), pixels darker than the surroundings expand while bright pixels contract. (3) Difference processing for calculating the gradation difference between the image data after the above (2) processing and the original image data stored in the image memory for each pixel. (4) Binarization processing in which each pixel data after the difference processing is binarized to determine the presence or absence of a defect for each pixel. (5) Defect determination processing in which the number of pixels that have been regarded as defects in the binarization processing is counted, and when the number of pixels is equal to or greater than a certain value, it is determined that there is a defect.

[0004]

However, this method is
Since it was proposed with the intention of removing the small dark defect portion existing inside the bright image and the small bright defect portion existing inside the dark image in the first place without destroying the original image pattern, the product For example, when a large concave-convex pattern is formed, a relatively small dark or bright defect existing in the concave or convex portion of the pattern (for example, a foreign substance having a refractive or light-shielding property against light)
Can be detected, but a large defect cannot be detected because it is difficult to distinguish from a concavo-convex pattern.

A first object of the present invention is, for example, in a product having a large concave-convex pattern formed, even if there is a large defect existing in the concave portion or the convex portion of the product, it can be quickly processed by a simple image processing method. Moreover, it is to enable accurate and easy detection. A second object of the present invention is to be able to detect even relatively small defects present in recesses or protrusions.

[0006]

In order to achieve the above first object, the defect inspection method according to the present invention irradiates a product with light from a light projector and receives transmitted light or reflected light with an image sensor, After storing the gradation value of each pixel from the image sensor in the image memory, the image analysis process is performed in the following procedure. (1) For the original image data of one screen stored in the image memory, the maximum gradation value in each of the large determination pixel regions M having a predetermined number of vertical and horizontal (vertical and horizontal directions) pixels is calculated, and the large determination is made. A shadow contraction process that replaces the gradation value of each pixel in the pixel area M with the obtained maximum gradation value for all pixels while sequentially shifting the large determination pixel area M. (2) Difference processing for calculating the gradation difference between the image data after the shadow contraction processing and the original image data stored in the image memory for each pixel. (3) For the pixel data after the difference processing, the gradation value of each pixel in each small judgment pixel area N having a predetermined number of vertical and horizontal pixels smaller than the number of pixels in the large judgment pixel area M is averaged gradation value. The smoothing process for replacing all the pixels by sequentially shifting the small determination pixel region N. (4) Binarization processing in which each pixel data after the smoothing processing is binarized to determine the presence or absence of a defect for each pixel. (5) Defect determination processing in which the number of pixels that have been regarded as defects in the binarization processing is counted, and when the number of pixels is equal to or greater than a certain value, it is determined that there is a defect.

Further, in order to achieve the second object, the image analysis processing is performed in the following procedure. (1) For the original image data of one screen stored in the image memory, the maximum gradation value is determined for each large determination pixel area M having a predetermined number of vertical and horizontal pixels, and each pixel in the large determination pixel area M is determined. The shadow contraction processing is performed for all the pixels while sequentially shifting the large determination pixel region M to replace the gradation value of 1 with the calculated maximum gradation value. (2) For the image data after the shadow contraction processing, the smallest gradation value in the large determination pixel area M is calculated, and the gradation value of each pixel in the large determination pixel area M is calculated as the minimum gradation value. The shadow dilation processing is performed for all pixels while sequentially shifting the large determination pixel area M. (3) First difference processing for calculating a gradation difference between the image data after the shadow expansion processing and the original image data stored in the image memory for each pixel, (4) Each after the difference processing A first binarization process for binarizing pixel data and determining the presence or absence of a defect for each pixel. (5) Counting the number of defective pixels in the binarization process, and determining that there is a defect when the number of pixels is a certain value or more.
Defect determination processing. (6) A second difference process for calculating, for each pixel, a gradation difference between the image data after the shadow contraction process and the original image data stored in the image memory. (7) For the pixel data after the difference processing, an intermediate gradation value is obtained for each of the small determination pixel areas N having a predetermined number of vertical and horizontal pixels smaller than the number of pixels of the large determination pixel area M, and the small determination pixel A smoothing process for replacing the gradation value of each pixel in the area N with the obtained intermediate gradation value for all pixels while sequentially moving the small determination pixel area N. (8) A second binarization process in which each pixel data after the smoothing process is binarized to determine the presence or absence of a defect for each pixel. (9) Counting the number of pixels that are regarded as defects in the binarization process, and determining that there is a defect when the number of pixels is a certain value or more.
Defect determination processing.

The first and second binarization processes may be carried out only for a fixed inspection range within one screen.

In the defect inspection apparatus according to the present invention, the product is irradiated with the light from the projector, the transmitted light or the reflected light is received by the image sensor, and the gradation value of each pixel from the image sensor is stored in the image memory. After that, image analysis processing is performed by the following means using a computer. (1) For the original image data of one screen stored in the image memory, the maximum gradation value is determined for each large determination pixel area M having a predetermined number of vertical and horizontal pixels, and each pixel in the large determination pixel area M is determined. A shadow contraction processing unit that replaces the gradation value of the above with the obtained maximum gradation value for all pixels while sequentially shifting the large determination pixel region M. (2) For the image data after the shadow contraction processing, the smallest gradation value in the large determination pixel area M is calculated, and the gradation value of each pixel in the large determination pixel area M is calculated as the maximum gradation value. A shadow expansion processing unit that performs the replacement for all the pixels while sequentially shifting the large determination pixel region M. (3) First difference processing means for calculating, for each pixel, a gradation difference between the image data after the shadow expansion processing and the original image data stored in the image memory. (4) First binarization processing means for binarizing each pixel data after the difference processing and determining the presence or absence of a defect for each pixel. (5) Counting the number of defective pixels in the binarization process, and determining that there is a defect when the number of pixels is a certain value or more.
Defect determination processing means. (6) Second difference processing means for calculating, for each pixel, a gradation difference between the image data after the shadow contraction processing and the original image data stored in the image memory. (7) For the pixel data after the difference processing, the gradation value of each pixel in each of the small judgment pixel areas N having a predetermined number of vertical and horizontal pixels smaller than the number of pixels in the large judgment pixel area M is averaged gradation value. Smoothing processing means for performing replacement of all the pixels while sequentially shifting the small determination pixel region N. (8) Second binarization processing means for binarizing each pixel data after the smoothing processing and determining the presence or absence of a defect for each pixel. (9) Counting the number of pixels that are regarded as defects in the binarization process, and determining that there is a defect when the number of pixels is a certain value or more.
Defect determination processing means. (10) Exclusion signal output means for outputting a signal for excluding a product when one of the first and second defect determination processing means determines that there is a defect.

[0010]

Now, when the gradation value due to the shadow is described as the shade of light and shade, the original image data is subjected to the shadow contraction process, then the shadow expansion process is performed, and the shadow expansion process image data is subjected to the difference process with the original image data. The light shadow and the shadow due to the large and large defects disappear, but the shadow due to the small dark defect and the shadow due to the relatively light and small defect smaller than the pattern-like shadow remain, and as a defect due to the subsequent binarization processing. To be detected.

After the original image data is subjected to the shadow contraction processing, the shadow contraction processed image data is subjected to the difference processing from the original image data.
The light shadow due to the pattern disappears in the central part and remains in the peripheral part, and is further thinned by smoothing it, so that it will not be erroneously detected as a defect by the subsequent binarization process. On the other hand, the shadow due to the dark and large shade is likewise the central part disappears and the peripheral part remains due to the difference processing, and although it is slightly lightened by smoothing this, it still has sufficient darkness. It is detected as a defect by the binarization process. Further, the shadow due to the dark and small defect remains almost as it is in the difference processing and is detected as a defect by the binarization processing.

[0012]

EXAMPLES Examples of the present invention will be described in detail below. FIG. 1 shows an outline of a defect inspection apparatus according to the present invention applied for inspecting the bottom of a transparent glass container 1 for defects. In this device, an inspection hole 3 is provided at a predetermined position of an opaque table 2, and a transparent glass substrate 4 and a positioning plate 6 having a light transmitting hole 5 thereon are stacked in the inspection hole 3. The transparent glass container 1 to be inspected is placed on the positioning plate 6 in an upright state. The setting of the transparent glass container 1 at the inspection position in this way is detected by a sensor (not shown). In order to irradiate the transparent glass container 1 with diffused light from directly above, a projector 7 having a light diffusing plate 7a is supported downward on a stand 8. On the other hand, an L-shaped lens barrel 9 is installed below the inspection hole 3, and a CCD camera 12 having a reflecting mirror 10 at a corner bent at a right angle in the lens barrel 9 and a two-dimensional image sensor 11 at a tip portion thereof. Are arranged.

The diffused light from the projector 7 enters from the inside of the transparent glass container 1 to the inner surface of the bottom 1a thereof and is transmitted to the outer surface thereof, and further is transmitted through the transparent glass substrate 4 as it is from the light transmitting hole 5 for inspection. The light passes through below the hole 3, reflects at a right angle on the reflecting mirror 10 and enters the CCD camera 12, and the two-dimensional image sensor 11 makes X (for example, 750) in the horizontal direction (horizontal direction) per frame in the vertical direction. The image is decomposed (vertically) into Y (for example, 500) pixels. Then, the image data is stored in the image memory in the image analysis processing device 13 by the computer with, for example, 256 gradation values for each pixel, and is subjected to image analysis processing in real time as described later. In this example, since it is intended to mainly detect defects of 0.5 mm or more, which are problems in quality assurance as a product of the transparent glass container 1, the optical system has a length of 1 m.
The image magnification is assumed to be 7 pixels per m.

Next, an example of the transparent glass container 1 to be inspected in this embodiment will be described. FIG. 2 shows a half section thereof, and FIG. 3 shows a bottom surface thereof. This glass container 1 has a large convex portion 14 radially integrally formed on the outer surface of the bottom portion 1a, and a narrow gap concave portion (or groove) 15 deeply depressed in a V shape between the convex portions 14 is formed. Large petal-shaped uneven patterns are formed when viewed from the whole. Each of the convex portions 14 projects from the circular central concave portion 16 at the center of the outer surface of the bottom so that each portion rises in a semicircular shape while gradually widening the width to both sides and gradually increasing the height.

Since the bottom portion 1a of the transparent glass container 1 is formed with such a large uneven pattern, when the light from the projector 7 is transmitted through the bottom portion 1a, it is largely refracted at the boundary of the unevenness, and the unevenness is caused. The pattern is directly captured by the CCD camera 12 as a shadow. In addition, the convex portion 14 or the concave portion 1
Refractive or reflective defects in 5 and 16, such as so-called falling powder, loading marks, transparent foreign substances such as bubbles and polka dots,
Alternatively, if there is rust, dirt, or an opaque foreign substance such as a slag, these defects are also photographed as shadows. Therefore, in the image analysis by the image analysis processing device 13, it is a shadow due to the uneven pattern or a shadow due to the defects. Is difficult to determine.

Therefore, the inventor of the present invention cannot prevent the shadow due to the uneven pattern of the bottom portion 1a of the transparent glass container 1 from appearing in the image of the CCD camera 12, or reduce the density even if it cannot be completely erased. As a result of repeated testing and research, I found that it is possible to prevent it from appearing as much as possible by the following.

Light from the projector 7 is transmitted from the inside of the transparent glass container 1 to the bottom thereof as shown in FIG. It is better to let light enter the inner surface of 1a and transmit it to the outer surface. This is because the concavo-convex pattern is formed on the outer surface of the bottom portion 1a and there is no concavity and convexity on the inner surface. Therefore, the shadow of the concavo-convex pattern becomes lighter when the inner surface is the light incident surface.

The degree of the shadow due to the uneven pattern mainly depends on the rising angle θ from the concave portions 15 and 16 to the convex portion 14 as shown in FIG. FIG. 5 is a graph obtained by simulating the relationship between the rising angle θ and the relative transmittance of light based on a Snell-Fresnel calculation formula regarding refraction / reflection. However, the positional relationship between the projector 7 and the transparent glass container 1 is as shown in FIG. As can be seen from this graph, the relative transmittance sharply decreases when the rising angle θ exceeds about 60 degrees. Therefore, when the rising angle θ is set to 65 degrees or less, preferably 60 degrees or less, a shadow due to the uneven pattern does not appear or even if it appears, it can be made thin.

Therefore, the concavo-convex pattern on the bottom 1a of the transparent glass container 1 should be such that the rising angle from the gap concave portion 15 to the convex portion 14 and the rising angle from the central concave portion 16 to the convex portion 14 are not more than 60 degrees. It is molded at the time of manufacture.

By the way, it is relatively easy to mold the convex portion 14 so as to have such a rising angle, when the width of the concave portion is wide, but the gap concave portion 15 of the transparent glass container 1 is formed. In the case of such a narrow space, it is difficult, and due to problems such as the attraction of the glass material by the mold and poor heat dissipation conditions, the gap recess 15 originally shaped as shown in FIG.
However, a small bulge 15a is formed in a part thereof as shown in FIG. 7, and this appears as a clear shadow in the captured image.

Therefore, as a result of repeated testing and research on this point, the present inventor was able to solve the above problem by devising the shape of the mold. That is, as shown in FIG. 8, in the mold 17, the convex portion 18 for forming the gap concave portion 15 is formed.
When the tip of the ridge was rounded, a small bulge 15a was sometimes formed, but when the tip of the convex portion 18 was sharpened as shown in FIG. 9, it disappeared, and the target shape shown in FIG. 6 could be formed.

As described above, also in the transparent glass container 1 itself to be inspected, the uneven pattern is a shadow and the CCD
If the image analysis processing device 13 is formed so as not to be easily imaged by the camera 12, the unevenness pattern and the defect can be very easily identified in the image analysis processing device 13, and the defect detection accuracy is improved.

Next, the image analysis and defect inspection processing by the image analysis processing device 13 will be described. FIG. 10 shows an outline of the processing. When the computer of the image analysis processing device 13 confirms from the signal from the sensor that the transparent glass container 1 is set at the inspection position (step S1),
The captured image of the CCD camera 12 is captured and stored in the image memory (step S2), and the first image analysis process (hereinafter referred to as "original image") of one frame of image data stored in the image memory is performed. , Which is referred to as T1 processing) (step S3). In this T1 process, a small defect is detected as described later, and each pixel is binarized as to whether or not it is a defective pixel. Then, the number of pixels as the defect is counted and it is determined whether or not the number is equal to or more than a predetermined number (step S4). When the number is equal to or more than the predetermined number, it is determined that there is a defect (step S5).

Next, a second image analysis process (hereinafter referred to as T2 process) is performed on the original image (step S).
6). In this T2 processing, a large defect is detected as described later, and each pixel is binarized as to whether or not it is a defective pixel. Then, the number of pixels as the defect is counted and it is determined whether or not the number is equal to or more than a predetermined number (step S7). When the number is equal to or more than the predetermined number, it is determined that there is a defect (step S8). In either of step S8 and step S5,
If there is a defect, an exclusion signal is output to exclude the transparent glass container 1 (step S9).

FIG. 11 shows the first image analysis process (T1 process).
3 is a flowchart showing a specific example of the algorithm of FIG.
When one pixel in the original image of one frame is set as a pixel of interest, a pixel region including, for example, three pixels in each of the left, right, up, and down directions with respect to the pixel of interest, that is, the vertical and horizontal 7
After imagining the large determination pixel area of × 7 = 49 pixels and obtaining the maximum gradation value in this (step S11), the gradation value of each pixel in the large determination pixel area is determined to be the maximum rank. The value is replaced with the key value (step S12). That is, when viewing the screen of one frame as the coordinates of the X and Y axes, for example, when the coordinates of the pixel of interest are (4, 4), 49 pixels of the coordinates of (1, 1) to (7, 7) Is the determination target, and the pixel at the coordinates (3, 3), for example, is the maximum gradation value “160”.
Then, the gradation values of 49 pixels in the coordinate area of (1, 1) to (7, 7) are all set to "160".

Then, the operations of steps S11 and S12 are performed for all the pixels of one frame while sequentially shifting the target pixel one by one (step S13). By such an operation, the dark part in the original image contracts, while the bright part expands. That is,
Shadows shrink due to defects. Regarding the pixels in the peripheral portion of the screen of one frame, the determination target area is less than 7 × 7 = 49 pixels, so an appropriate number of pixels less than 49 is set as the determination target area. The number is as large as 750 × 500 as described above, and the peripheral portion is actually outside the inspection target range for defect detection as described later, so that the processing at the peripheral portion does not adversely affect the defect detection.

Next, with respect to the image after the replacement with the maximum gradation value as described above, in the opposite manner to the above operation, 7 × 7 = 4 in height and width.
After the minimum gradation value is calculated in the large determination pixel area of 9 pixels (step S14), the gradation value of each pixel in the large determination pixel area is replaced with the calculated minimum gradation value (step S14).
15). Then, the operations of steps S14 and S15 are performed for all the pixels in one frame while sequentially shifting the target pixel one by one (step S1).
6). By such an operation, contrary to the above, the dark part expands, while the bright part contracts. That is, the shadow due to the defect expands.

Next, the tone values of the image after the minimum tone value replacement as described above and the original image are calculated for each pixel (step S17). By this difference processing, a small dark portion remains in the original image, and a large dark portion disappears. In other words, small defects that appear as small shadows remain, but large defects that appear as large shadows disappear. After that, the image after the difference processing is binarized for each pixel only within a specific inspection range, that is, the gradation value based on a predetermined threshold is compared for each pixel, and a defect is generated according to the size. It is determined whether or not the pixel is a pixel (step S18). Then, the number of defective pixels is counted (step S1).
9).

Next, FIG. 12 shows a second image analysis process (T2
It is a flowchart which shows the specific example of the algorithm of (processing). The steps S21 to S23 in the T2 process are the same as the steps S11 to S13 in the T1 process described above, and the shadow due to the defect is contracted by substituting the maximum gradation value. In the next step S24, the tone values of the image after the maximum tone value replacement and the original image are obtained for each pixel, and then the following smoothing processing is performed on the image after the difference.

That is, in the image after the difference, when a certain pixel is set as a pixel of interest, a pixel region including, for example, one pixel in each of the left, right, up, and down directions with respect to this pixel, that is, 3 × 3 = 9 pixels in length and width After virtualizing the small determination pixel area of, and obtaining an intermediate gradation value in this (step S
25), the gradation value of each pixel in the small determination pixel area is replaced with the obtained intermediate gradation value (step S26). For example, if the coordinates of the pixel of interest are (4,4), (3,3)
9 pixels of coordinates (5, 5) are subject to the determination, and if, for example, a pixel of coordinates (3, 3) has an intermediate gradation value of “25”, (3, 3) to ( The gradation values of nine pixels in the coordinate areas 5 and 5 are all set to "25".

Then, the operations of steps S25 and S26 are carried out for all the pixels in one frame while sequentially shifting the target pixel one by one (step S27). By such an operation, only the dark dark part in the original image remains and the other parts disappear. That is, the dark large defect and the dark small defect remain, but the light large defect and the light small defect are assimilated into the bright portion. After this,
The image after the replacement of the intermediate gradation value is binarized for each pixel only within a specific inspection range, that is, the gradation value based on a predetermined threshold value is compared for each pixel, and a defect occurs according to the size. After determining whether or not the pixel is a pixel (step S28), the number of defective pixels is counted (step S29).

Next, the algorithms of the above-mentioned T1 processing and T2 processing will be illustrated and explained in an easy-to-understand manner based on an actual example.

Now, the boundary (rise) of the unevenness of the uneven pattern on the bottom of the transparent glass container 1 is made into a straight line as shown in FIG. 13A, and the shadow in the original image (hereinafter referred to as the pattern shadow) is Assuming that the line appears as a light straight line having a thickness of about 1 mm as shown in FIG. 7B, the operation from step S11 to step S13 in the T1 process (T
The same applies to the operations from step S21 to step S23 in the two processes), and the pattern shadow shrinks in size as shown in FIG.

When the operation from the next step S14 to step 16 in the T1 process is performed, the size of the pattern shadow is the same as in FIG. It is almost the same as (C). When this is subjected to the difference processing with the original image in step S17, the pattern shadow disappears as shown in FIG. Therefore, the pattern shadow is determined in step S1.
Since it has disappeared in this way in step 7, if it is binarized in the next step S18, it is naturally not judged as a defect.

Further, when the pattern shadow as shown in FIG. 7C is subjected to the difference processing with the original image in step S24 in the T2 processing, the central portion of the pattern shadow appearing as shown in FIG. It disappears like F) and the peripheral part remains. Next, when the operations of step S25 to step S27 are performed, the peripheral portion of the remaining pattern shadow becomes thin as shown in FIG.
Thereafter, in the T2 process, 2 is set based on the threshold value in step S28.
Since the value is digitized, the peripheral portion of the thinned pattern shadow disappears and is not judged as a defect.

On the other hand, the transparent glass container 1 has a defect as shown in FIG. 14 (A), and a part of the edge of the defect is a small and sufficiently dark shadow (1 mm or less) as shown in FIG. 14 (B). Hereinafter, the small small defect shadow will appear as a dark small defect shadow, and the dark small defect shadow disappears by the operation from step S11 to step S13 in the T1 process (the same as the operation from step S21 to step S23 in the T2 process). However, the whole image is slightly brightened.

When the operations from step S14 to step 16 in the T1 process are performed, the dark small defect shadows disappear and the entire image becomes slightly dark. When this is subjected to the difference processing with the original image in step S17, a dark small defect shadow appears again as shown in FIG.
It is detected as a defect at 18.

When the pattern shadow as shown in FIG. 6C is subjected to the difference processing with the original image in step S24 in the T2 processing, a dark small defect shadow appears again as shown in FIG. When the operations from step S25 to step S27 are performed, as shown in FIG. 7G, although it is slightly thin, it remains as a dark small defect shadow. Therefore, if this is binarized in step S28, it is detected as a defect.

On the other hand, as shown in FIG. 15B, a part of the edge of the defect as shown in FIG. 15A has a sufficiently thick shadow with a thickness exceeding 1 mm (hereinafter referred to as a large dark defect shadow). ) Appears by the operation from step S11 to step S13 in the T1 process (the same as the operation from step S21 to step S23 in the T2 process), the dark large defect shadow is as shown in FIG. Shrinks with the same thickness.

When the operations from the next step S14 to step 16 in the T1 process are carried out, the dark large defect shadow has the same size as that in FIG. It is almost the same as in FIG. When this is subjected to the difference processing with the original image in step S17, the dark large defect shadow disappears as shown in FIG. Therefore, since the dark large defect shadow has disappeared in this way in step S17, it cannot be detected even if it is binarized in the next step S18.

However, when the dark large defect shadow as shown in FIG. 7C is subjected to the difference processing with the original image in step S24 in the T2 process, the central portion of the dark large defect shadow appearing as shown in FIG. Disappears as shown in FIG. 6F, and the peripheral portion remains. Next, when the operations from step S25 to step S27 are performed, the peripheral portion of the remaining dark large defect shadow is shown in FIG.
Although it becomes a little thin as shown in, it still remains as a dark defect shadow. Therefore, if this is binarized based on the threshold value in step S28, it can be detected as a defect.

Next, the transparent glass container 1 has a defect as shown in FIG. 16 (A), and a part of the edge of the defect is darker than the pattern shadow but relatively light as shown in FIG. 16 (B). Moreover, a small shadow of 1 mm or less (hereinafter referred to as a light small defect shadow)
, The operation of steps S11 to S13 in the T1 process (T2
The same applies to the operations from step S21 to step S23 in the processing), the light small defect shadow disappears, and the entire image becomes slightly brighter.

When the operations from step S14 to step 16 in the T1 process are performed, the light small defect shadow remains disappeared, and the entire image becomes slightly dark. When this is subjected to the difference processing with the original image in step S17, a light small defect shadow appears slightly darker than the pattern shadow as shown in FIG. 8E, and therefore it can be detected as a defect in step S18.

When the pattern shadow as shown in FIG. 6C is subjected to the difference processing with the original image at step S24 in the T2 processing, a light small defect shadow appears again as shown in FIG. When the operations from step S25 to step S27 are performed, as shown in (G) of the same figure, a slight thin defect shadow remains. If this is binarized in step S28, it is possible to detect when there is a gradation difference from the pattern shadow,
If there is almost no difference, it cannot be detected as a defect.

As described above, whether the defect detection of large and small by the T1 process and the T2 process can be arranged is as follows. (1) Detection of defects with dark shadows of 1 mm or less is possible with either T1 processing or T2 processing. (2) The detection of a defect of a light shadow of 1 mm or less is possible in the T1 process and not possible (sometimes possible) in the T2 process. (3) Detection of dark defects exceeding 1 mm is not possible with T1 processing, but is possible with T2 processing. (4) The detection of faint defects exceeding 1 mm is impossible with either T1 treatment or T2 treatment, but the probability of such defects occurring is very small considering the manufacturing method of the transparent glass container. Not detected.

FIGS. 17 to 20 show the above-described processing in the case of each of the pattern shadow, the dark small defect shadow, the dark large defect shadow, and the light small defect shadow, the gradation value of the gradation of the image and the gradation by the difference processing. The difference is symbolized and illustrated.

In the case of the pattern shadow of FIG. 17, since the gradation differences after the difference processing in step S17 are all less than 10 gradations in the T1 processing, the threshold value in the binarization processing in step S18 is set to 20 with a margin, for example. Then, the pattern shadow will not be erroneously detected as a defect. In addition, after the smoothing operation from step S25 to step S27 in the T2 process, the maximum gradation difference may be close to 20.
If the threshold value of the binarization process in step S28 is set to about 30 in consideration of variations in sensitivity due to differences in the pixels of the camera 12, the pattern shadow will not be erroneously detected as a defect.

In the case of the dark small defect shadow of FIG. 18, since the gradation difference of the defect portion after the difference processing in step S17 is 30 to 80 in the T1 processing, the threshold value in the binarization processing in step S18 is set to the above value. If the number is 20, the defect can be sufficiently detected. Further, after the smoothing operation from step S25 to step S27 in the T2 process, since the gradation difference of the defect portion is 40 to 60, the threshold value of the binarization process at step S28 is set to 30 as described above.
If it is a degree, it can be sufficiently detected as a defect.

In the case of the dark large defect shadow shown in FIG. 19, since the gradation difference of the defect portion after the difference processing in step S17 is less than 10 in the T1 processing, the threshold value in the binarization processing in step S18 is set as described above. If it is set to 20, the disadvantage is that detection is impossible. On the other hand, after the smoothing operation from step S25 to step S27 in the T2 process, since the gradation difference of the defect portion is 40 to 70, the threshold value of the binarization process at step S28 is set to 30 as described above. If it is a degree, it can be sufficiently detected as a defect.

In the case of the light small defect shadow of FIG. 20, since the gradation difference of the defect portion after the difference processing in step S17 is 15 to 25 in the T1 processing, the threshold value in the binarization processing in step S18 is set to the above value. If the number is 20, the defect can be detected. However, after the smoothing operation from step S25 to step S27 in the T2 process, the gradation difference of the defect portion is 15 to 25, so that the threshold value of the binarization process at step S28 is about 30 Is impossible to detect.

As described above, in the present embodiment, the concavo-convex pattern itself of the transparent glass container 1, that is, the rising angle from the concave portion to the convex portion is set so that the image captured by the CCD camera 12 is less likely to have a shadow. In order not to scatter even if the shadow appears, the light from the projector 7 is made to enter from the inside of the transparent glass container 1 to the inner surface of the bottom portion 1a and transmitted to the outer surface. , T2 again
The processing mainly detects large defects, and even if a large uneven pattern is formed on the bottom of the transparent glass container 1, large and small defects can be detected with high accuracy.

In the described embodiment, the transparent glass container 1 is used as an inspection object, the light from the projector 7 is transmitted to the bottom portion 1a of the transparent glass container 1, and an image of the transmitted light is taken by the CCD camera 12 to inspect for defects. However, it is possible to inspect for defects other than the bottom portion in the same manner. It is also possible to take an image by inspecting by reflected light instead of transmitted light. In that case, not only a transparent product but also an opaque product having an uneven pattern is subjected to the above-mentioned T1 process and T2 process. The presence or absence of large and small defects can be inspected by.

[0053]

The effects of each claim of the present invention are as follows. (Claim 1) For example, in a product having a large concavo-convex pattern, even a large defect existing in the concave portion or the convex portion can be detected quickly, accurately and easily by a simple image processing method. .

(Claims 2 and 4) Not only large defects but also small defects can be detected similarly.

(Claim 3) Since the binarization process is performed only for a certain inspection range in one screen, the invalid part (periphery of the screen) in the shadow contraction process, shadow expansion process and difference process which are the preceding operations. Part) can be eliminated and defect detection can be performed without any trouble.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram of an example of a defect inspection apparatus of the present invention.

FIG. 2 is a diagram showing a cross section in the left half of a transparent glass container which is an example of an inspection target, and a diagram showing the outer surface in the right half.

FIG. 3 is a bottom view of the transparent glass container.

FIG. 4 is a diagram illustrating a rising angle from a concave portion to a convex portion in the concavo-convex pattern of the transparent glass container.

FIG. 5 is a graph showing the relationship between the rising angle and the relative transmittance.

FIG. 6 is an enlarged cross-sectional view showing a normal shape of the recess.

FIG. 7 is an enlarged cross-sectional view showing an abnormal shape in which a part of the recess is raised.

FIG. 8 is an enlarged cross-sectional view of a part of a conventional mold for molding the uneven pattern.

FIG. 9 is an enlarged cross-sectional view of a portion of the improved mold.

FIG. 10 is a flowchart showing an outline of image analysis and defect inspection processing by the image analysis processing apparatus.

FIG. 11 is a flowchart showing a specific example of an algorithm of a first image analysis process in the above.

FIG. 12 is a flowchart showing a specific example of an algorithm of second image analysis processing.

FIG. 13 is a schematic diagram illustrating a transformation process of a pattern shadow by the first and second image analysis processes.

FIG. 14 is a schematic diagram illustrating the transformation process of a dark small defect shadow.

FIG. 15 is a schematic diagram illustrating a metamorphosing process of a dark large defect shadow.

FIG. 16 is a schematic diagram illustrating a metamorphosis process of a light small defect shadow.

FIG. 17 is an explanatory diagram illustrating the gradation change process of the pattern shadow by symbolizing the gradation value of the light and shade of the image and the gradation difference by the difference processing.

FIG. 18 is an explanatory diagram illustrating a gradation change process of a dark small defect shadow similarly.

FIG. 19 is an explanatory diagram illustrating a gradation change process of a dark large defect shadow similarly.

FIG. 20 is an explanatory diagram illustrating a gradation change process of a light small defect shadow similarly.

[Explanation of symbols]

 1 transparent glass container 7 light projector 11 image sensor 12 CCD camera 13 image analysis processing device

Claims (4)

[Claims] 1. A large uneven pattern or the like forms the light from the projector.
Irradiating a formed product and receiving its transmitted light or reflected light by an image sensor, and storing the gradation value of each pixel from the image sensor in an image memory for image analysis processing. good is, the presence or absence of a defect present in the large concave and convex pattern or the like
A method of inspecting, (1) above for the original image data of one screen stored in the image memory, obtains a maximum gradation value therein for each large determination pixel region M of the vertical and horizontal predetermined number of pixels, the large A shadow contraction process that replaces the gradation value of each pixel in the judgment pixel area M with the obtained maximum gradation value for all pixels while sequentially shifting the large judgment pixel area M, and (2) the shadow A difference process for calculating a tone difference between each pixel for the image data after the contraction process and the original image data stored in the image memory; (3) for the pixel data after the difference process, the large determination pixel region For each small judgment pixel area N having a predetermined number of vertical and horizontal pixels smaller than the number of M pixels, an intermediate gradation value is obtained, and the gradation value of each pixel in the small judgment pixel area N is obtained. Substituting the value sequentially shifts the small determination pixel area N Smoothing processing performed on all pixels while (4) binarizing processing for binarizing each pixel data after the smoothing processing to determine the presence or absence of a defect for each pixel, and (5) binarizing processing 2. A defect inspection method for a product, comprising: a defect determination process of counting the number of pixels determined to be defects and determining that there is a defect when the number of pixels is a certain value or more. 2. The light from the projector is shaped into a large uneven pattern.
Irradiating a formed product and receiving its transmitted light or reflected light by an image sensor, and storing the gradation value of each pixel from the image sensor in an image memory for image analysis processing. good is, the presence or absence of a defect present in the large concave and convex pattern or the like
A method of inspecting, (1) above for the original image data of one screen stored in the image memory, obtains a maximum gradation value therein for each large determination pixel region M of the vertical and horizontal predetermined number of pixels, the large A shadow contraction process that replaces the gradation value of each pixel in the judgment pixel area M with the obtained maximum gradation value for all pixels while sequentially shifting the large judgment pixel area M, and (2) the shadow Regarding the image data after the contraction processing, the smallest gradation value in the large determination pixel area M is obtained, and the gradation value of each pixel in the large determination pixel area M is replaced with the obtained maximum gradation value. , A shadow expansion process performed on all pixels while sequentially moving the large determination pixel region M, and (3) a gradation difference between the image data after the shadow expansion process and the original image data stored in the image memory for each pixel. The first difference processing which is calculated for each (4) A first binarization process for binarizing each pixel data after the difference process to determine the presence / absence of a defect for each pixel, and (5) counting the number of pixels that are defective in the binarization process, If the number of pixels exceeds a certain value, it is determined that there is a defect.
Defect determination processing of (6), second difference processing of calculating a gradation difference between the image data after the shadow contraction processing and the original image data stored in the image memory for each pixel, (7) For the pixel data after the difference processing, an intermediate gradation value is obtained for each of the small determination pixel areas N having a predetermined number of vertical and horizontal pixels smaller than the number of pixels of the large determination pixel area M, and in the small determination pixel area N. The smoothing process of replacing the gradation value of each pixel with the obtained intermediate gradation value for all pixels while sequentially moving the small determination pixel region N, and (8) each pixel data after the smoothing process Second binarization processing for binarizing each pixel to determine the presence / absence of a defect for each pixel, and (9) counting the number of pixels identified as defects in the binarization processing, and the number of pixels is a certain value or more. If there is a second defect
A defect inspection method for a product, comprising: 3. The defect inspection method for a product according to claim 2, wherein the first and second binarization processing is performed only for a certain inspection range in one screen. 4. The light from the projector is shaped into a large uneven pattern.
Irradiating a formed product and receiving its transmitted light or reflected light by an image sensor, and storing the gradation value of each pixel from the image sensor in an image memory for image analysis processing. good is, the presence or absence of a defect present in the large concave and convex pattern or the like
A defect inspection apparatus for a product to be inspected , characterized by comprising the following means. (1) For the original image data of one screen stored in the image memory, the maximum gradation value is determined for each large determination pixel area M having a predetermined number of vertical and horizontal pixels, and each of the large determination pixel areas M is determined. A shadow contraction processing unit that replaces the gradation value of a pixel with the obtained maximum gradation value for all pixels while sequentially shifting the large determination pixel region M. (2) For the image data after the shadow contraction processing, the smallest gradation value in the large determination pixel area M is calculated, and the gradation value of each pixel in the large determination pixel area M is calculated as the minimum gradation value. A shadow expansion processing unit that performs the replacement for all the pixels while sequentially shifting the large determination pixel region M. (3) First difference processing means for calculating, for each pixel, a gradation difference between the image data after the shadow expansion processing and the original image data stored in the image memory. (4) First binarization processing means for binarizing each pixel data after the difference processing and determining the presence or absence of a defect for each pixel. (5) Counting the number of defective pixels in the binarization process, and determining that there is a defect when the number of pixels is a certain value or more.
Defect determination processing means. (6) Second difference processing means for calculating, for each pixel, a gradation difference between the image data after the shadow contraction processing and the original image data stored in the image memory. (7) For the pixel data after the difference processing, an intermediate gradation value is obtained for each of the small determination pixel areas N having a predetermined number of vertical and horizontal pixels smaller than the number of pixels of the large determination pixel area M, and the small determination pixel Smoothing processing means for replacing the gradation value of each pixel in the area N with the obtained intermediate gradation value for all the pixels while sequentially shifting the small determination pixel area N. (8) Second binarization processing means for binarizing each pixel data after the smoothing processing and determining the presence or absence of a defect for each pixel. (9) Counting the number of pixels that are regarded as defects in the binarization process, and determining that there is a defect when the number of pixels is a certain value or more.
Defect determination processing means. (10) Exclusion signal output means for outputting a signal for excluding a product when one of the first and second defect determination processing means determines that there is a defect.