JP3960127B2 - Appearance inspection device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention is installed, for example, in an area or line for producing an inspection object such as a beverage container, images the inspection object with a camera, digitally processes the image of the camera, and inspects an appearance defect of the inspection object. Visual inspection equipment,
In particular, defects such as smudges and missing prints in complicated patterns (patterns) formed by printing on the object to be inspected and having a specific shading distribution, specifically fine defects (that is, fine The present invention relates to an appearance inspection apparatus that can accurately detect defects having a large difference in density level from the periphery and light defects (that is, defects having a small difference in density level from the surrounding area but are not so fine).
[0002]
In the following drawings, the same reference numerals denote the same or corresponding parts.
[0003]
[Prior art]
Conventionally, in this type of visual inspection apparatus that performs digital processing of images captured by a camera of an object to be inspected with a specific pattern (pattern) to inspect defects in the pattern on the object to be inspected, missing prints, etc. In order to prevent false defect detection, manually set the area excluding the edge part with a large shading change in the picture according to the picture in advance, and set the picture area excluding this edge part. For the above image, a difference in gray level between a point of interest on the image and its periphery was detected to detect a defect.
[0004]
[Problems to be solved by the invention]
However, when a specific pattern to be inspected becomes complicated, it is extremely difficult to manually set an area other than the edge in the pattern. ThereIn the present inventionThe purpose is to solve this problem and automatically generate an edge mask image for extracting an inspection object image other than the edge from the image of the original pattern of the object to be inspected regardless of the complexity of the pattern. Is to provide.
[0005]
Next, if the image to be inspected has a dark area with a very low density level, such as a black pattern, and there is a defect in the dark area, for example, the black printed portion of the metallic beverage can as the inspection object is printed When there is a small spot-like bright spot of about 0.5 mmφ exposed, when the size of the bright spot is equal to or smaller than the light receiving area of one pixel of the image sensor, Due to the offset at the time of A / D conversion or the response characteristics of the video signal amplifier, the video signal voltage (density signal level) does not rise to the original density level of the bright spot, and the bright spot has a large area. There is a phenomenon that the response density level becomes smaller than the case.
[0006]
For this reason, if the same defect detection sensitivity is applied to the dark area as in the area excluding the dark area, the detection sensitivity of the dark area is significantly reduced, and if the detection sensitivity of the dark area is increased, the dark area is excluded. There is a problem that the detection sensitivity in the region becomes too high, and a non-defective product is erroneously detected as a defective product, thereby reducing the production efficiency.
In order to solve this problem, the dark area may be designated as an individual inspection area different from the area excluding the dark area, and the defect detection sensitivity corresponding to the individual inspection area may be set. In many cases, there is no clear boundary, and it is practically difficult to manually set the dark area. Also, since the density gradient of the image is relatively large near the boundary between the dark area and the non-dark area, there is a possibility that a defect is erroneously detected. Therefore, it is necessary to mask this boundary area in the same manner as the edge portion of the pattern.
[0007]
In addition, when the inspection target image is a color image, and the defect detection is performed by masking the inspection target image for each color component of RGB with the mask image of the corresponding color component, it is separated for each color component due to subtle variations in the print color. In particular, since the position of the edge portion and the density gradient in the pattern of the image fluctuate, there is a problem that the defect detection is mistaken near the edge. ThereIn other of the present inventionThe purpose is to solve this problem, to automatically extract a dark area from an inspection object image obtained by imaging a non-defective inspection object, to obtain a dark window image, and from a non-defective inspection object image as in the first invention. An appearance inspection apparatus that automatically generates an inspection mask image for extracting inspection target pixels in the dark area from the obtained edge mask image and the dark window image, and when the inspection target image is a color image, for each color component of RGB Using a common edge mask image consisting of the logical sum of individual edge mask images, create an inspection mask image used for image inspection for each RGB color component, and detect defect detection errors based on color fluctuations near the edges of the pattern An object of the present invention is to provide a visual inspection apparatus for preventing the above.
[0008]
Next, when the image to be inspected includes a horizontal striped pattern in which parallel thin lines are arranged, using the edge mask image causes a problem that the striped pattern portion is almost covered with the edge mask image and the defect cannot be detected. is there. ThereIn other of the present inventionThe object is to provide an appearance inspection apparatus that can solve this problem, extract a striped pattern area in an inspection target image without using an edge mask image, and perform reliable defect detection of the striped pattern area. is there.
[0010]
[Means for Solving the Problems]
In order to solve the above-mentioned problems, the appearance inspection apparatus according to claim 1 is a pattern (31) formed by printing on an object to be inspected (work 20) having a metal as a base and having a predetermined light and shade distribution. ) (Via the camera 6), and A / D conversion (via the A / D conversion unit 1) of the picture of the picture obtained by this imaging is performed to obtain a digital image (development image 30). (Hereinafter, referred to as an image to be inspected) (and stored in the image memory 2), processing the density information of the image to be inspected, and detecting defects such as stains and missing prints on the image to be inspected. An appearance inspection apparatus (such as the image processing apparatus 01) that determines the quality of an inspection object,
ForecastInspected as good productsFirst cover obtained from the objectAn edge as a portion having a density gradient of a predetermined value or more on the inspection image is detected by two-dimensional differentiation processing (using a Sobel operator or the like), and all edge pixels obtained by this detection are detected on the inspection object. Depending on the shape etc.First coverFor each of one or a plurality of individual inspection areas (32-3, 32-4, etc.) set in the inspection image, N is an integer including 0 determined according to the individual inspection area. To generate an edge mask image (30M) that is inflated and to be inspected as a non-defective productFirst cover on the objectAfter obtaining a dark part binary image showing a pixel in the dark part region (100D) having a density level equal to or lower than a predetermined value on the inspection image by dividing it from other pixels, pixels in the dark part region on the dark part binary image are obtained. Is reduced by N pixels (where N is an integer including 0 determined according to the dark area), and a dark part window image (30D) is generated. An inspection mask image (30T) for masking pixels other than the inspection target pixel on the dark part window image is generated (and masked) using pixels excluding pixels corresponding to edge pixels on the edge mask image as inspection target pixels. Means (stored in the memory 3) (such as the microprocessor 5);Cover to be judged good or badObtained from inspectionSecond coverThe inspection image is masked by the inspection mask image so that the positions of the patterns corresponding to both images match, and the mask after the maskingSecond coverDefect detection means (defect detection circuit 4, defect count unit 7, microprocessor 5) that performs the defect detection with each region formed by connecting the inspection target pixels as one of the individual inspection regions for an inspection image. Etc.).
[0012]
Then billItem 2Appearance inspection device, claim1In the appearance inspection apparatus described above, when the image to be inspected is a color image, the edge mask image (individual edge mask image) generated with the one color component at the time of the defect detection performed for each RGB color component. 30M) is replaced with a new edge mask image (common edge mask image 30MK) that masks all of the pixels masked by any of the edge mask images for each color component.
[0013]
Next billItem 3Appearance inspection device, claim1 or 2In any one of the appearance inspection apparatuses, the defect detection unit moves the pixel of interest (Po) one pixel at a time along the scanning line to the density value (Po) of the pixel of interest and the scanning line of the pixel of interest. Before and after the alignment, an interval (span amount d) determined according to the individual inspection area is provided, and a number (m) of pixels determined according to the individual inspection area are connected in the scanning direction ( The difference between the average density values (Pfb and Pbb) of the pixel area (of the front background pixel Pf and the rear background pixel Pb) is obtained, and either of the previous and subsequent differences is set to a threshold value determined according to the individual inspection area. It is assumed that a first fine defect detecting means for determining that the pixel of interest is a defective pixel when it exceeds the threshold value is provided.
[0014]
Next billItem 4No appearance inspection device is claimed in claim 1Of 3In any one of the appearance inspection apparatuses, the defect detection unit moves the pixel of interest (Po) one pixel at a time along the scanning line to the density value (Po) of the pixel of interest and the scanning line of the pixel of interest. Before and after the alignment, m pixels × n pixels (where m and n are integers determined according to the individual inspection area), with an interval (span amount d) determined according to the individual inspection area. The difference between the two pixel areas (PGfb and PGbb) (of the front background pixel group PGf and the rear background pixel group PGb) is obtained, and any of the previous and subsequent differences is determined as the individual difference. It is assumed that there is provided a second fine defect detection unit that determines that the target pixel is a defective pixel when a threshold value determined in accordance with the inspection area is exceeded.
[0015]
Next billItem 5No appearance inspection device is claimed in claim 14In any one of the visual inspection apparatuses, the defect detection unit includes m pixels × n pixels (a span amount d) that is set in accordance with the individual inspection area before and after the scanning line. , M, n are integers determined according to the individual inspection area), and one of the two pixel areas (PGf, PGb, hereinafter referred to as a comparison pixel group) is a predetermined comparison pixel group. The representative pixel is a pixel of interest, and the pixel of interest is moved one pixel at a time along the scanning line, and the difference between the respective density average values of the comparison pixel groups before and after this is obtained. It is assumed that there is provided a light defect detection means for determining that the pixel of interest is a defective pixel when a threshold value determined accordingly is exceeded.
[0017]
Also as the mainBookThe effect of the invention is that a dark area on a non-defective inspection target image is extracted in advance by darkening the density by binarization, and an edge mask image obtained from the non-defective inspection target image is used. By generating an inspection mask image for extracting a pixel to be inspected and masking the pattern image of the original inspection object with this inspection mask image, the defective pixel can be obtained even in a dark area where the density level is low. When the inspection target image is a color image, when creating an inspection mask image for each RGB color component, instead of individual edge mask images for each RGB color component, individual edges are used. By using the edge mask image common to each color component consisting of the logical sum of the mask image, defect detection near the edge of the pattern based on the color variation, etc. Ri is intended to as to prevent.
[0019]
DETAILED DESCRIPTION OF THE INVENTION
(Embodiment 1)
First, a first embodiment of the first invention related to claims 1, 5 and 7 will be described with reference to FIGS.
FIG. 2 is an explanatory diagram of an image pickup unit as an embodiment of the first invention. In the figure, reference numeral 20 denotes a work as an object to be inspected. In this example, 20 is a cylindrical container such as beer or juice, but a container such as a paper cup having a taber side surface may be used. Reference numeral 20a denotes a body portion as a cylindrical portion of the workpiece 20, and reference numeral 20b denotes a neck portion as an upper constriction portion following the body portion 20a of the workpiece 20.
[0020]
Reference numeral 31 denotes a pattern printed on the entire outer peripheral surface of the workpiece 20. The picture 31 includes a letter “A” and a black circle as well as a white background. In general, it is not a simple one as shown in this figure, but is usually a complex one composed of various characters and figures.
Reference numeral 14 denotes a motor that rotates the workpiece 20 at a constant speed with its central axis as the rotation axis 15, 11 is a light source, and its output light becomes illumination light 13 arranged in parallel on one plane via the line type fiber 12. Illuminate the outer peripheral surface of the line. Reference numeral 13 a denotes an illuminated line (illuminated line) on the outer peripheral surface of the workpiece 20.
[0021]
In this example, this illumination is performed so that the portion of the illuminated line 13a on the workpiece body 20a is parallel to the workpiece rotation axis 15.
Reference numeral 6 denotes a line camera that images the portion of the illuminated line 13a on the outer peripheral surface of the work 20 in a state where the work 20 is rotating at a constant speed as shown in the figure.
The rotational speed of the workpiece 20 is comprehensively determined based on the diameter of the workpiece 20, the amount of illumination light, the brightness of the camera lens, the number of processes as the apparatus, and the like. p. m.
[0022]
Video information on the illuminated line 13a picked up by the line camera 6 is sent from the camera 6 to the main body 01a of the image processing apparatus 01 shown in FIG. In the main unit 01a, digital video information on the illuminated line 13a is accumulated for one rotation of the work 20, and a two-dimensional image described below is obtained.
[0023]
FIG. 3 shows a developed image (that is, a print screen on the outer peripheral surface of the workpiece 20) 30 in which the image of the image taken by the line type camera 6 is developed into a two-dimensional image of one rotation of the workpiece 20. The entire developed image 30 is an image of the pattern 31. The image of the pattern 31 is formed with a predetermined shading distribution.
32 (32-1, 32-2) are individual inspection areas corresponding to the body portion 20a and the curved surface portion 20b of the workpiece 20 set for defect inspection in the developed image 30, respectively. The reason for dividing the defect inspection area in this way is that the curved surface portion 20b having a sagged shape is less likely to reflect the illumination light 13 (that is, darker) than the body portion 20a. This is because it is better to divide the parameters into the individual inspection areas 32-1 and 32-2.
[0024]
4 shows a mask image extracted and generated as described later from the developed image 30 of FIG. 3 for the workpiece 20 that has been determined to be non-defective in advance (hereinafter, this mask image is referred to as an edge mask image in order to distinguish the mask image more strictly). 30M).
The developed image 30 in FIG. 3 for the workpiece 20 to be inspected is masked by aligning the position of the pattern 31 with the edge mask image 30M in FIG. 4, so that the individual inspection areas 32-1 and 32-2 in the developed image 30 are masked. A print screen area corresponding to the area shown in white in FIG. 4 is extracted, and this is the area to be inspected in this example.
[0025]
FIG. 1 is a block diagram showing the configuration of an image processing apparatus as an embodiment of the first invention. The image processing apparatus 01 includes the image processing apparatus main body 01a and the camera 6. In the image processing apparatus main body 01a, 5 is a microprocessor for controlling the entire apparatus.
Reference numeral 1 denotes an A / D converter that converts an analog image signal output from the camera 6 while scanning the captured image in the vertical direction (Y-axis direction) in this example to a digital image signal, and 2 denotes the digital image signal. It is an image memory that stores a digital image that is input and corresponding to captured images of at least the entire circumference of the workpiece 20.
[0026]
Reference numerals 3, 4, and 7 denote a mask memory, a defect detection circuit, and a defect count section, which will be described later, which are the main parts of the first invention. A bus 8 couples the microprocessor 5 with the image memory 2, the mask memory 3, the defect count unit 7, and the like.
FIG. 7 is a flowchart showing a processing procedure of the image processing apparatus 01, and S1 to S12 are step numbers of this processing. Next, the processing contents of the first invention will be described along this flowchart with reference to other drawings as appropriate.
[0027]
In this embodiment, the difference processing described later is performed on the entire surface of the developed image (print screen) 30 (FIG. 3) as a digital image of the pattern 31 on the outer peripheral surface of the work 20 imaged by the line-type camera 6 to cause a print defect. However, if a portion having a large density change (density gradient) in the pattern 31 is not masked at the time of the difference, that portion will be erroneously detected. Therefore, first, processing for generating an edge mask image 30M for masking (FIG. 4) is performed.
[0028]
That is, the work 20 that has been determined to be non-defective in advance is imaged by the camera 6 (step S1), and a non-defective image 30 is acquired via the A / D conversion unit 1 and stored in the image memory 2.
Next, a sobel operator, which is a well-known two-dimensional differential processing means, is applied to the non-defective developed image 30 stored in the image memory 2 so that the density gradient in the pattern 31 is larger than a predetermined value. A certain edge is detected (step S2).
[0029]
The processing of the Sobel operator is executed by a circuit that applies the Sobel operator to the image, or if the processing time is not a problem, the microprocessor 5 accesses the image memory 2 to calculate the operator for each pixel. And extract edges.
Then, the operation result of the Sobel operator is binarized to obtain an edge mask image 30M composed of edge pixels located at the edge portion (step S3).
[0030]
The edge mask image 30M may be left as it is when the position variation of the image is extremely small. However, when there is a position variation of the image based on the printing of the pattern 31 or the rotation of the workpiece 20, one or more edge mask images 30M are used. The edge mask image 30M is expanded inside and outside the edge by a predetermined number of pixels. Note that the magnitude of this expansion may be changed between the individual inspection areas 32-1 and 32-2. Then, the edge mask image 30M generated in this way is stored in the mask memory 3 (step S3a).
[0031]
Next, when masking the developed image 30 to be inspected with the edge mask image 30M, a reference position for superimposing the images 30M and 30 on each other is registered (step S4), and the inspection mode is entered (step S5). .
Note that the reference position is a partial image of, for example, about 20 pixels × 20 pixels in the developed image 30 (and thus the pattern 31), and appears only once in one round of the outer periphery of the work 20, that is, the developed image. A characteristic partial image position whose position is uniquely determined by 30 scans is selected.
[0032]
A reference position 33 indicated by a one-dot chain line in FIG. 8 shows an example of a reference position on the developed image 30. In step S4, coordinates indicating the relative position of the reference position 33 on the developed image 30 are shown. , And an image pattern in a square frame of a one-dot chain line are registered.
In the inspection mode, first, the outer peripheral surface (pattern printed surface) of the workpiece 20 as an inspection object is imaged in the same manner as in the case of the above-mentioned non-defective product, and the developed image 30 is stored in the image memory 2 (step S6).
[0033]
Next, the developed image 30 of the inspection object is scanned, the reference position 33 is detected by a pattern matching method, and the developed image 30 is transferred to the defect detection circuit 4. On the other hand, the edge mask image 30M in the mask memory 3 is also aligned with the reference position 33 (that is, the original developed image 30 of the edge mask image 30M and the developed image 30 of the inspection object are correctly matched). By transferring the defect image to the defect detection circuit 4 (in a positional relationship), the developed image 30 of the object to be inspected is converted into an edge image.TheMasking is performed with the mask image 30M (step S7).
[0034]
In the defect detection circuit 4, an image region generated by the mask process, excluding the edge mask image 30 M from the developed image 30 of the inspection object (that is, an image that does not include a partial image with a large density gradient that becomes an erroneous defect). The difference between smoothing and integration described below is taken for (region), and the defect is inspected (step S8).
In other words, defect detection is performed for each pixel in the scanning direction (specifically, the direction of raster scanning which is the Y-axis or X-axis direction, that is, either the vertical or horizontal direction). The difference between the gray level of the target pixel or the gray level average value of the predetermined pixel area with the target pixel as the representative pixel and the gray level average value of the predetermined peripheral pixel area before and after the pixel area is taken, and the difference is greater than or equal to the threshold value. At this time, the target pixel is detected as a defective pixel.
[0035]
Then, the defective pixel is counted by the defect counting unit 7 to check whether or not the number of defective pixels is equal to or larger than a predetermined threshold (step S9). As a result, if the number of defective pixels is equal to or greater than a predetermined threshold (branch Y), the inspection object (work 20) is defective (step S10), otherwise (step S9, branch N), The inspection object (work 20) is determined to be good (step S11).
[0036]
The number of defective pixels is counted separately for the individual inspection areas 32 (32-1, 32-2), and the defect set for each of the individual inspection areas 32-1, 32-2 is also determined for good / bad. This is performed according to a threshold value of the number of pixels. Therefore, if there is a defect determination in any of the individual inspection areas 32-1 and 32-2, the object to be inspected is a defective product, and a good determination is made in both the individual inspection areas 32-1 and 32-2. If so, the object to be inspected becomes a genuine good product.
[0037]
In the above-described difference inspection, the pixel itself that is added to the calculation does not match the mask pixel (that is, the masked pixel, which is the pixel included in the edge mask image 30M in the case of the first invention), and the target point and the peripheral pixel Is performed on condition that there is no mask pixel in a span amount d (described later) in which the interval is expressed by the number of pixels.
FIG. 5 shows a first smooth difference inspection for detecting a fine defect which is fine but has a large difference in gray level between the point of interest and the periphery. FIG. 6 shows a small difference in gray level between the point of interest and the periphery. It is each principle explanatory drawing of the integral difference inspection which detects the light defect which is not.
[0038]
For the fine defect, as shown in the explanatory diagram of the first smoothing difference inspection in FIG. 5, refer to the pixel value (density value) Po of the pixel of interest Po and the back of the pixel of interest Po together with scanning (see FIG. 5A). ), The difference from the pixel average value (density average value) Pbb of m pixels (referred to as the back background pixel Pb) separated by the span amount d, and the pixel value (density value) Po of the pixel of interest Po and the pixel of interest Any of the differences from the pixel average value (density average value) Pfb of m pixels (referred to as the front background pixel Pf) separated by the span amount d in front of Po (see FIG. 5, b) is a predetermined 2 When the threshold value is exceeded, the pixel of interest Po is set as a defective pixel.
[0039]
Note that the hatched pixels in the rear background pixel Pb and the front background pixel Pf in FIG. 5 represent pixels representing the positions of the m pixels in the averaging region. The number m of background pixels, the span amount d, and the binarization threshold in the smooth difference inspection are determined separately for the individual inspection area 32 (32-1, 32-2).
As for the light defect, as shown in the explanatory diagram of the integral difference inspection in FIG. 6, there are two such that the representative pixels of the pixel group of n pixels × m pixels (referred to as a comparison pixel group) are spaced by the span amount d. The difference between the average pixel values obtained by integrating the pixel values of the comparison pixel group, that is, the comparison pixel group PGf in the front and the comparison pixel group PGb in the rear, on the scanning line Compared with a predetermined binarization threshold while moving, and when the difference exceeds the binarization threshold, pay attention to one of the representative pixels of the comparison pixel groups PGf and PGb, for example, the representative pixel of the rear comparison pixel group PGb The pixel Po is a defective pixel.
[0040]
Also in this integral difference inspection, the number of pixels m and n (in this example, both are selected in the range of 1, 2 and 4 pixels for ease of calculation), the span amount d, and the binarization threshold are the individual It is determined separately for the inspection area 32 (32-1, 32-2).
In this manner, the processing after step S6 is repeated for each object to be inspected, the quality is determined, and the inspection mode is terminated when necessary (step S12).
[0041]
By the way, the camera 6 used in the image processing apparatus 01 may be a monochrome camera or a color camera. In the case of a color camera, an edge mask image 30M for each RGB color from the three developed images 30 that are separated for each RGB color. The developed image 30 of the inspected object is also acquired for each of the RGB colors and inspected for each color, and when the developed image 30 of any color is determined to be a non-defective product, Make it a genuine good product.
[0042]
(Embodiment 2)
Next, an embodiment of the second invention mainly related to claims 2, 3 and 6 will be described mainly with reference to FIGS. In this embodiment, the configuration of the image processing apparatus 01 of FIG. 1 and the imaging unit of FIG. 2 (except for the configuration of the pattern 31 and the individual inspection area 32 of the imaging target workpiece 20 of FIG. 2) are applicable.
[0043]
FIG. 9 shows a workpiece 20 which is a cylindrical container having a dark area as one embodiment of the second invention, and a black pattern 42 or the like is formed on the entire circumference of the container bottom side as the lower part of the trunk portion 20a of the workpiece 20. It is assumed that there is a black printing unit 100 made of other patterns having gradation.
FIG. 10 shows an example of the density distribution on the vertical broken line connecting the points YA, YP and YB on the body 20a of the workpiece 20 of FIG. Note that the black printing portion 100 is between the points YA and YP.
[0044]
In other words, the black printing portion 100 is approximately 30 levels or less in 256 gradation levels. In this example, the density between the points YA and YB changes with gradation, and there is no clear boundary line between the background 102 and the black printing unit 100. The pattern 41 in FIG. 9 is lighter than the black printing unit 100 and darker than the background 102, and the pattern 42 is darkest.
[0045]
When printing peeling 101 occurs in the black printing unit 100, the metal ground of the beverage can is exposed, and the exposed portion is mirror-like, so that the illumination light is reflected and becomes a spot-like bright spot. If such a bright spot has a large area, the density signal level reaches a saturation level, but if the bright spot is small, it responds only to about plus 15 to 20 levels with respect to the dark part level.
[0046]
For this reason, in the second invention, a dark part region such as the black printing unit 100 is extracted by the method described below, and the dark part region is defined as a separate region (32-3 in this example) of the individual inspection region 32. An image defect such as the print peeling 101 in the area is detected.
FIGS. 11 to 14 show images used for the image defect detection processing in the dark area when the workpiece 20 of FIG. 9 is the inspection object, and FIG. 15 shows the second smoothing difference inspection applied in this embodiment. Show the principle.
[0047]
FIG. 16 is a flowchart showing a basic processing procedure of the image processing apparatus 01 of FIG. 1 as one embodiment of the second invention, and this figure corresponds to FIG. In FIG. 16, steps S3b to S3d are added to FIG. 7, and steps S7 and S8 are replaced with S7A and S8A, respectively.
Next, the procedure of FIG. 16 will be described mainly with respect to the points different from FIG. 7 while adding the description of FIGS.
[0048]
First, as in the case of FIG. 7, a non-defective inspection object (in this example, the work 20 in FIG. 9) is imaged by the camera 6 to obtain a developed image 30 as shown in FIG. 11, and the image memory 2 in FIG. (Step S1).
Next, as in the case of FIG. 7, for example, a Sobel operator is applied to the developed image 30 of FIG. 11 (step S <b> 2), and then fixed binarization is performed to obtain an edge mask image 30 </ b> M as shown in FIG. Step S3).
[0049]
In FIG. 13, 41E and 42E as the portions represented by black lines are the edges of the patterns 41 and 42, respectively. Each pixel of the edges 41E and 42E is represented by data 1, and pixels other than the edges are represented by data 0. Has been. Further, as in the case of FIG. 7, the edge mask image 30M is expanded to the inside and the outside of the edge according to the degree of position variation of the image (step S3a).
[0050]
Next, the dark part area | region in the black printing part 100 of the workpiece | work 20 is extracted from the expansion | deployment image 30 of the good workpiece | work 20 of FIG. For this purpose, the developed image 30 is fixed and binarized by setting the density threshold THD to about 30, and a dark window image 30D as shown in FIG. 12 is obtained (step S3b).
In the dark part window image 30D, the extracted dark part region 100D has data of 1 for each pixel, and the data of each pixel has 0 except for the dark part region 100D. A point YP in FIG. 9 represents the position of the pixel corresponding to the density threshold value THD, as shown in FIG.
[0051]
Next, the dark portion window image 30D is reduced by a predetermined amount according to the degree of gradation on the black printing portion 100 (S3c). This is because the dark area 100D (data) is used in order to prevent erroneous detection of defective pixels in the vicinity of the boundary between the dark area 100D (data: 1) and the other area (data: 0), which is the target area for image defect detection. 1) it is necessary to narrow the position.
[0052]
Next, even in the dark part region 100D, there may be a case where an edge of the pattern (in this example, the edge 42E of the pattern 42) exists. Therefore, the polarity of the dark part window image 30D that has undergone the reduction processing in step S3c is reversed. 13 (ie, the pixel area to be masked is data 1 and the non-mask area (inspection target pixel area) is data 0, with the same logical polarity as the mask image), and the edge mask of FIG. 13 that has undergone the expansion process in step S3a A logical sum image with the image 30M is obtained, and an inspection mask image 30T (that is, a mask image used for actual inspection) 30T as shown in FIG. 14 is obtained (step S3d), and this inspection mask image 30T is stored in the mask memory 3 of FIG. To store.
[0053]
Similarly to the case of FIG. 7, the reference position on the image when the developed image 30 of the inspection object is masked with the inspection mask image 30T is registered, and the inspection mode is entered (steps S4 and S5).
In the inspection mask image 30T, the pixel of data 1 is masked and defect detection is not performed, and the pixel of data 0 is detected defect. In other words, in the present example, out of the pixels in the dark area 100D shown in FIG. 12, the pixels excluding the pixels of the edge 42E of the pattern 42 in the dark area 100D become the defect detection target pixels.
[0054]
Next, in the inspection mode, as in the case of FIG. 7, after imaging the workpiece 20 of the inspection object (step S6), the defect detection circuit 4 of FIG. 1 displays the developed image 30 of the inspection object obtained by this imaging. Then, the inspection mask image 30T stored in the mask memory 3 is masked so that the reference positions of both the images 30 and 30T coincide with each other (step S7A).
[0055]
Then, the detection target pixel area on the dark area 100D extracted with this mask is set as a separate individual inspection area (32-3 in this example), the detection sensitivity is increased exclusively for the dark area, and defective pixels are detected by differential inspection ( Step S8A).
Note that, if necessary, the defect detection parameters such as detection sensitivity are changed with the pixel area to be inspected surrounded by the edge 42E in the dark area 100D as another individual inspection area (32-4 in the example of FIG. 14). There is also.
[0056]
5 and 6 may generally be applied to the difference inspection, but in this embodiment, a fine defect is detected by using the second smoothing difference inspection method shown in FIG.
The principle of FIG. 15 is basically the same as that of FIG. 5 (first smoothing difference inspection), but the difference from FIG. 5 is that the front and rear background pixels to be compared with the pixel of interest Po are arranged in a line. This is a point in which a pixel group of m pixels × n pixel regions is formed from m pixels.
[0057]
In the smoothing difference inspection in FIG. 15, the pixel value (density value) Po of the pixel of interest Po and the pixel behind the pixel of interest Po (see FIG. 15A) along with scanning are n pixels × m pixels separated by the span amount d. The difference from the pixel average value (density average value) PGbb of the back background pixel group PGb, the pixel value (density value) Po of the pixel of interest Po, and the front of the pixel of interest Po (see FIG. 15, b)), span The difference from the pixel average value (density average value) PGfb of the front background pixel group PGf of n pixels × m pixels separated by the amount d is predetermined binarized while moving the pixel of interest Po on the scanning line one pixel at a time. Compared with the threshold value, when any of the differences exceeds a predetermined binarization threshold value, the pixel of interest Po is determined as a defective pixel.
[0058]
Note that the hatched pixels in the rear background pixel group PGb and the front background pixel group PGf in FIG. 15 represent pixels that represent the position of an area of n pixels × m pixels. In addition, the number m and n of pixels in this difference inspection (in this example, both are selected in the range of 1, 2 and 4 pixels for ease of calculation), the span amount d, and the binarization threshold are the individual inspection regions ( In this example, it is defined for 32-3).
[0059]
The second smoothing difference inspection method of FIG. 15 can generally be applied to the first invention described above and the third invention described later.
Next, when the image of the inspection object is a color image, basically, the processing of FIG. 16 may be performed on the image for each color component obtained by decomposing the inspection object into RGB color components. Then, there is a subtle variation in print color, and when color separation is performed into RGB, there is a variation in density level for a specific color component.
[0060]
For this reason, a pixel to be masked in an edge mask image with a certain color component is not masked with an edge mask image with another color component, but becomes a pixel to be inspected, and defect detection may be erroneously performed.
FIG. 17 is a flowchart showing a color image processing procedure of the image processing apparatus 01 that prevents such erroneous detection of a defect, and corresponds to FIG. Here, steps S101 to S103, S103a to S103d, S104 to S106, S107A, S108A, and S109 to S112 in FIG. Corresponds to ~ S12.
[0061]
However, in FIG. 17, with respect to FIG. 16, branch steps BR1 to BR3 for confirming whether or not the processing for each of the RGB color components has been completed in the course of the series of processing of FIG. S103A is added.
Here, the workpiece 20 in FIG. 9 is an inspected object printed in color, and FIG. 17 will be described mainly with respect to the differences from FIG. 16. First, the non-defective workpiece 20 is decomposed into RGB color components and imaged (steps). S101) From the developed image 30 of one color component, an edge mask image for each color component (herein referred to as “individual edge mask image”) 30M that has been expanded in accordance with image fluctuations as in FIG. Repeat for three color components (steps S102 to S103a, BR1).
[0062]
Then, in the next step S103A which is the eye of the processing of FIG. 17, it is composed of the logical sum of the three individual edge mask images 30M for each RGB color component described above (that is, masked by the individual edge mask image 30M for each color component). A common edge mask image 30MK is obtained as an edge mask image (where all pixels to be masked are mask pixels).
[0063]
Next, as in FIG. 16, the developed image 30 of one color component of the non-defective work 20 is fixed and binarized, and a dark part window image 30D for the one color component reduced according to the image variation is created. The logic of the dark part window image 30D is inverted to obtain a logical sum with the common edge mask image 30MK, and the inspection mask image 30T of the one color component is created (steps S103b to S103d). Then, the processing in steps S103b to S103d is repeated for each color component, and inspection mask images 30T for three RGB color components are obtained (step BR2) and stored in the mask memory 3.
[0064]
Next, the reference positions of the developed image 30 and the inspection mask image 30T for each color component are registered in the same manner as in FIG. 16, and the inspection mode is entered (steps S104 and S105).
In the inspection mode, after the workpiece 20 of the inspection object is color-separated into RGB and imaged (step S106), the defect detection circuit 4 displays the developed image 30 for each color component of the inspection object obtained by this imaging. 16, the developed image 30 of each one-color component is masked with the inspection mask image 30 </ b> T of the one-color component so that the reference positions of both the images 30 and 30 </ b> T coincide with each other. A difference inspection of (non-masked pixels) is performed (steps S107A and S108A). Then, the number of defective pixels is checked through the defect count unit 7 (S109).
[0065]
Then, the processes in steps S107A to S109 are repeated for the three color components of RGB (step BR3), and in the process of repetition, the number of defective pixels equal to or greater than the threshold is detected in any of the individual color components. The inspection object is determined to be defective (step S110), and the inspection objects determined to be good for all the color components are finally determined as non-defective products (steps S111 and S112).
[0066]
(Embodiment 3)
Next, an embodiment of the third invention related mainly to claim 4 will be described mainly with reference to FIGS. In this embodiment, the configuration of the image processing apparatus 01 of FIG. 1 and the imaging unit of FIG. 2 (except for the configuration of the pattern 31 and the individual inspection area 32 of the imaging target workpiece 20 of FIG. 2) are applicable.
[0067]
FIG. 18 shows the appearance of a work 20 of a container having a striped pattern area as one embodiment of the third invention. In the example of FIG. 18, it is assumed that there is a striped pattern printing unit 200 in which a horizontal striped pattern 43 is printed below the cylindrical body 20 a of the workpiece 20 having the same shape as shown in FIG. It is assumed that the printing defect occurs as an image defect generated in the printing unit 200. Moreover, the direction of the straight line which forms the stripe of the horizontal stripe pattern 43 is assumed to face the circumferential direction of the cylindrical body 20a in this example.
[0068]
FIG. 19 shows a developed image 30 obtained by imaging the work 20 of FIG. An edge mask image is created from the developed image 30 of the inspection object including the horizontal striped pattern 43 as shown in FIG. 19 as in the first invention, and when the developed image 30 is masked, the straight stripes forming the horizontal striped pattern 43 are formed. The entire portion is masked, and an image defect such as the print peeling 201 in FIG. 18 cannot be detected.
[0069]
For this reason, in the embodiment of the third invention, a separate individual inspection area 32 (this example) is used without using an edge mask for the existence area of only the straight lines forming the stripes of the horizontal stripe pattern 43 in the stripe pattern printing unit 200. In this individual inspection area 32-5, the defect detection scanning direction is set in the same direction as the straight line forming the stripes, and the m pixels × n pixels shown in FIG. The image defect such as a printing defect is detected by performing the light defect detection process according to.
[0070]
FIG. 20 is a flowchart showing the processing operation of the image processing apparatus 01 of FIG. 1 as one embodiment of the third invention, that is, in this case, the horizontal stripe pattern when the work 20 of FIG. Fig. 4 shows a flowchart of image inspection.
Note that steps S201, S204 to S207A, and S209 to S212 in FIG. 20 are the same processes as steps S1, S4 to S7A, and S9 to S12 in FIG. 16, respectively, and steps SP1, SP2, and S208B in FIG. It is a unique process.
[0071]
Therefore, the specific processing will be mainly described below. In FIG. 20, first, a non-defective product of the workpiece 20 shown in FIG. 18 is imaged (step S <b> 201), and then the position of the existence area of only the straight lines forming the stripes of the horizontal stripe pattern 43 printed on the workpiece 20 is image processing apparatus. 01 is set, and this existence area is set as an individual inspection area 32-5 (step SP1).
[0072]
Next, an inspection mask image for masking pixels other than the individual inspection area 32-5 on the non-defective image is created (step SP2) and stored in the mask memory 3.
Further, in the inspection mode, the defect detection circuit 4 masks an image (in this case, the developed image 30 in FIG. 19) obtained by imaging the inspected product by matching the reference positions of both images with the inspection mask image. (Steps S206 and S207A).
[0073]
Then, a non-mask pixel on the inspection target image obtained by imaging the inspected product, that is, a pixel in the individual inspection area 32-5 is differentially inspected to detect defective pixels. In this differential inspection, in this embodiment, the scanning direction AR of defect detection is set to the direction of a straight line that forms the stripes of the horizontal stripe pattern 43, and the light defect detection processing by m pixels × n pixels shown in FIG. This is performed (step S208B).
[0074]
For defect inspection of other individual inspection areas other than the individual inspection area 32-5, as in the case of the first or second embodiment, an edge mask image obtained from the developed image 30 in FIG. 19 is used. it can. This is because the inspection sensitivity of the horizontal stripe pattern portion is lowered by the mask, and no defect is detected in the horizontal stripe pattern portion, so that it is not necessary to particularly distinguish the horizontal stripe pattern region.
[0075]
【The invention's effect】
According to the first aspect of the present invention, the edge portion having a large density change in the image of the pattern on the object to be inspected determined in advance is detected by two-dimensional differentiation processing, and the edge pixel obtained by this detection is detected as the pattern image. For each individual inspection area set in accordance with the shape of the object to be inspected, a predetermined integer pixel including 0 is expanded to generate an edge mask image,
The defect inspection image area in the pattern image of the inspection object is extracted by masking the pattern image of the original inspection object with this edge mask image so that it is independent of the complexity of the pattern. Thus, the defect inspection area is automatically determined, and the labor and time for defect inspection can be reduced as compared with the case where the defect inspection area is manually set according to the conventional pattern.
[0076]
In addition, the density value or density average of the point of interest in the area other than the edge in the pattern as the defect inspection image area extracted as described above, without erroneously capturing defects due to a large change in shading at the edge in the pattern Since the defect portion is detected by comparing the level difference between the value and the average density value of the background before and after it with a threshold value, it is possible to accurately detect a fine defect or a light defect regardless of the pattern.
[0077]
Next, according to the second aspect, the dark part is obtained from the dark part window image obtained by extracting the dark part area of the inspection target image by binarizing the density and the edge mask image obtained from the inspection target image in the same manner as the first aspect. An inspection mask image for extracting inspection target pixels in an area is automatically generated, and when the inspection target image is a color image, common to each color component consisting of a logical sum of individual edge mask images for each RGB color component Using the edge mask image of, the inspection mask image for each RGB color component was created.
Even in the dark area where the density level is small, defective pixels can be detected reliably, and fine print peeling in the black print area can also be detected as defective. Also in color image processing, it is possible to reliably detect defective pixels near the edge of a pattern while suppressing the influence of RGB color fluctuations.
[0078]
According to the third invention, a striped pattern region having a predetermined shape including parallel straight lines is extracted from the inspection target image without using an edge mask image, and the direction of the straight line is used as a scanning direction for defect detection. Since the difference inspection is performed on these pixels, the defective pixels in the striped pattern area can be reliably detected without being covered with the edge mask image, and fine print peeling in the striped pattern area can also be detected as defective.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a system configuration as an embodiment of the first to third inventions.
FIG. 2 is an explanatory diagram of an image pickup unit that similarly picks up an image on an inspection object.
3 is a diagram showing a developed image of a pattern on the inspection object in FIG. 2;
4 shows a mask image corresponding to FIG. 3;
FIG. 5 is a principle diagram of a first smoothing difference inspection in the first to third inventions.
FIG. 6 is a principle diagram of integral difference inspection in the first to third inventions.
FIG. 7 is a flowchart for explaining the operation as one embodiment of the first invention;
8 is a diagram showing an example of a reference position on the developed image of FIG.
FIG. 9 is an external view of an inspection object as one embodiment of the second invention.
10 is a diagram showing an example of a density distribution on the inspection object in FIG. 9;
11 is a diagram showing a developed image of the pattern on the inspection object in FIG. 9;
12 is a diagram showing a dark window image corresponding to FIG.
13 is a diagram showing an edge mask image corresponding to FIG.
14 shows an inspection mask image corresponding to FIG.
FIG. 15 is a principle diagram of a second smoothing difference inspection in the first to third inventions.
FIG. 16 is a flowchart for explaining the operation as one embodiment of the second invention;
FIG. 17 is a flowchart for explaining the operation during color image processing as another embodiment of the second invention;
FIG. 18 is an external view of an inspection object as an embodiment of the third invention.
19 is a diagram showing a developed image of a pattern on the inspection object in FIG. 18;
FIG. 20 is a flowchart for explaining the operation as one embodiment of the third invention;
[Explanation of symbols]
01 Image processing device
01a Image processing apparatus main body
1 A / D converter
2 Image memory
3 Mask memory
4 Defect detection circuit
5 Microprocessor
6 Camera (Line type camera)
7 Defect count section
8 Bus
11 Light source
12 line type fiber
13 Illumination light
13a Illuminated line
14 Motor
15 Workpiece rotation axis
20 work pieces
20a Work torso
20b Curved surface of workpiece
30 Unfolded image (print screen)
30M mask image, edge mask image, individual edge mask image
30MK common edge mask image
30D dark window image
30T inspection mask image
31 designs
32 (32-1 to 32-5) Individual inspection area
33 Reference position
41, 42 patterns
41E Edge of pattern 41
42E Edge of pattern 42
43 Horizontal Stripe Pattern
100 Black printing part
100D dark area
101 Print peeling in the black print section
102 background
200 Striped pattern printing department
201 Print stripping of the striped pattern printing section
Po pixel of interest
Pb Back background pixel
Pbb Average value of back background pixels
PGbb Rear background pixel group average value
Pf Front background pixel
Pfb Front background pixel average value
PGfb Front background pixel group average value
PGb Rear comparison pixel group, rear background pixel group
PGf Front comparison pixel group, front background pixel group
d Span amount
m Number of background pixels in the averaging area (horizontal axis direction)
n Number of background pixels in the averaging area (vertical direction)
AR Defect detection scan direction

Claims (5)

It is attached by the printing metal on the inspection Butsujo to underlying images a pattern formed by having a predetermined gray distribution,
A digital image corresponding to the picture (hereinafter referred to as an inspected image) is generated by A / D converting the picture of the picture obtained by this imaging,
A visual inspection apparatus that processes density information of the image to be inspected to detect defects such as stains and print defects on the image to be inspected, and determines the quality of the object to be inspected.
An edge as a portion having a density gradient equal to or higher than a predetermined value on the first inspected image obtained from the inspected object determined in advance as a non-defective product by two-dimensional differentiation processing;
All edge pixels obtained by the detection, per one or a plurality of discrete inspection area set before Symbol in the first inspection image in accordance with the shape of the object to be inspected, N pixels (where N is an integer including 0 determined according to the individual inspection region), and an expanded edge mask image is generated.
After obtaining a dark part binary image indicating pixels in the dark part region having a density level equal to or lower than a predetermined value on the first inspection image for the inspected object determined as the non-defective product , separately from the other pixels. Then, a dark part window image formed by reducing the pixels of the dark part region on the dark part binary image by N pixels (where N is an integer including 0 determined according to the dark part region) is generated,
Inspection that masks pixels other than the pixel to be inspected on the dark part window image, using pixels obtained by removing pixels corresponding to edge pixels on the edge mask image from pixels in the dark part region on the dark part window image as inspection target pixels Means for generating a mask image;
Masking the second inspection image obtained from the inspection object to be judged as good or bad by the inspection mask image so that the positions of the patterns corresponding to both images match,
Previous SL second inspection image after the mask targets, and a defect detection device for performing the defect detecting each region in which the inspection target pixel is formed by connecting a respective one of the individual test area An appearance inspection apparatus characterized by that. The appearance inspection apparatus according to claim 1 ,
When the image to be inspected is a color image, the edge mask image generated with the one color component at the time of the defect detection performed for each RGB color component is selected from any of the edge mask images for each color component. An appearance inspection apparatus characterized by replacing all pixels masked by a new edge mask image. The appearance inspection apparatus according to claim 1 or 2 ,
While the defect detection means moves the pixel of interest along the scanning line pixel by pixel, the density value of the pixel of interest and the interval determined according to the individual inspection area before and after along the scanning line of the pixel of interest To obtain a difference from the density average value of the pixel area formed by connecting the number of pixels determined according to the individual inspection area in the scanning direction, and any of the difference before and after this is the individual inspection. An appearance inspection apparatus comprising first fine defect detection means for determining a target pixel as a defective pixel when a threshold value determined in accordance with a region is exceeded. The appearance inspection apparatus according to any one of claims 1 to 3 ,
While the defect detection means moves the pixel of interest along the scanning line pixel by pixel, the density value of the pixel of interest and the interval determined according to the individual inspection area before and after along the scanning line of the pixel of interest The difference between each of the two pixel areas of m pixels × n pixels (where m and n are integers determined according to the individual inspection area) is obtained, Appearance characterized by comprising second fine defect detection means for determining that the pixel of interest is a defective pixel when any of the subsequent differences exceeds a threshold determined in accordance with the individual inspection area Inspection device. In the appearance inspection apparatus according to any one of claims 1 to 4,
The defect detection means includes m pixels × n pixels (where m and n are determined in accordance with the individual inspection area, respectively) with a predetermined interval in accordance with the individual inspection area before and after the scanning line. The representative pixel of one of the predetermined comparison pixel groups in each pixel group (hereinafter referred to as a comparison pixel group) in the two pixel regions of the two pixel regions is a pixel of interest, and the pixel of interest is one pixel along the scanning line. While moving one by one, a difference between the respective density average values of the comparison pixel groups before and after this is obtained, and when the difference exceeds a threshold determined according to the individual inspection area, the target pixel is determined as a defective pixel. An appearance inspection apparatus comprising a light defect detection means for determining.

Images