JP6933887B2 - Inspection equipment and inspection method - Google Patents

Description

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

Conventionally, an apparatus has been used in which a three-dimensional object is irradiated with light to take an image, and the appearance of the object is inspected based on the captured image. For example, Patent Document 1 discloses an inspection method for inspecting a circular product. In this method, the inspection image of the circular product is developed and acquired in a strip shape by using an imaging means while rotating the circular product with a rotating jig, and the inspection reference position is pattern-matched to the master image developed in the strip shape. After matching the reference with, the inspection image and the master image are collated.

In addition, Non-Patent Document 1 discloses a method for obtaining an optical flow in two images. The technique uses an energy function that penalizes the assumptions of brightness invariance, brightness gradient invariance, and deviations from smoothness assumptions, and the vector that minimizes this energy is coarse-to-. It is requested by using fine warping).

Japanese Unexamined Patent Publication No. 2014-95579

Thomas Brox et al., "High Accuracy Optical Flow Estimation Based on a Theory for Warping", European Conference on Computer Vision, Springer LNCS 3024, May 2004, vol. 4, p. 25-36

By the way, when an object on the stage is imaged by the imaging unit, if the direction of the object deviates from a predetermined direction, the appearance of the object from the imaging unit changes. In this case, when the target image acquired by the imaging unit is compared with the predetermined reference image, a portion of the target image that does not appear in the reference image appears, so that the portion is not a defect. Regardless, it is detected as a defect (so-called false defect detection).

On the other hand, when the reference image is aligned with respect to the target image by obtaining the optical flow of pixels between the reference image and the target image, a part of the reference image is distorted so as to resemble the above part, which is a false defect. Detection is suppressed. However, in the pixel-by-pixel alignment, the reference image may be excessively resembled to the target image, and defects such as a portion whose dimensions differ greatly from the design value in the target object cannot be detected.

The present invention has been made in view of the above problems, and an object of the present invention is to appropriately align the target image and the reference image to perform a preferable inspection.

The invention according to claim 1 is an inspection device, which comprises a target image storage unit that stores a target image obtained by capturing an image of an object, a reference image storage unit that stores a reference image, and the target image and the reference image. The alignment unit includes a alignment unit for detecting a defect region in the target image by comparing both images after the alignment, and the alignment unit is the target image and the reference. The first alignment processing unit that individually aligns a plurality of regions obtained by dividing the first setting region of one image of the image with respect to the other image is different from the first setting region in the one image. It includes a second alignment processing unit that aligns a region included in the second setting region and larger than any of the plurality of regions with respect to the other image.

The invention according to claim 2 is the inspection apparatus according to claim 1, wherein the comparison unit shifts one of the two images after alignment vertically and horizontally with respect to the other. In, a plurality of difference images are obtained from both of the images, and the defect region is detected based on the plurality of difference images.

The invention according to claim 3 is the inspection apparatus according to claim 1 or 2, wherein the second alignment processing unit aligns the other image in the second setting region by using an affine transformation. I do.

The invention according to claim 4 is the inspection apparatus according to claim 3, wherein the second alignment processing unit acquires an entire processed image of the one image by using an affine transformation, and said that. The first setting area after alignment by the first alignment processing unit is combined with the processed image, and the second setting area, which is an area other than the first setting area in the processed image after composition, is formed. wherein that are aligned with respect to the other image.

The invention according to claim 5 is the inspection apparatus according to any one of claims 1 to 4, wherein the first alignment processing unit is an optical with respect to the other image of each pixel in the first setting area. Alignment is performed by obtaining the flow.

The invention according to claim 6 is an inspection method, in which a) a step of preparing a target image obtained by capturing an image of an object, b) a step of aligning the target image with a reference image, and c) a position. A step of detecting a defect region in the target image by comparing both the combined images is provided, and the b) step is b1) a first setting region of one image of the target image and the reference image. B2) The step of individually aligning a plurality of regions divided into the above with respect to the other image, and b2) the plurality of regions included in the second setting region different from the first setting region in the one image. It includes a step of aligning a region larger than any of the regions with respect to the other image.

The invention according to claim 7 is the inspection method according to claim 6, wherein in the step c), a plurality of relative positions in which one of the two images after the alignment is shifted vertically and horizontally with respect to the other. In the relationship, a plurality of difference images are obtained from the two images, and the defect region is detected based on the plurality of difference images.

The invention according to claim 8 is the inspection method according to claim 6 or 7, wherein in the step b2), the alignment with respect to the other image in the second setting region is performed by using the affine transformation. ..

The invention according to claim 9 is the inspection method according to claim 8, wherein in the step b2), the entire processed image of the one image is acquired by using the affine transformation, and the step b1). The first set area after alignment in the above is combined with the processed image, and the second set area, which is a region other than the first set area in the processed image after composition, is relative to the other image. It has been aligned Te that.

The invention according to claim 10 is the inspection method according to any one of claims 6 to 9, wherein in the step b1), an optical flow of each pixel in the first setting region with respect to the other image is obtained. As a result, alignment is performed.

According to the present invention, by individually aligning the target image and the reference image in the first and second setting areas, the alignment according to each setting area is appropriately performed, and a preferable inspection is performed. Can be done.

It is a figure which shows the structure of the inspection apparatus. It is a top view which shows the main body of an inspection apparatus. It is a block diagram which shows the functional structure realized by a computer. It is a figure which shows the flow of the process of inspection of an object. It is a figure which shows the target image. It is a figure which shows the reference image. It is a figure which shows the 1st processed reference image. It is a figure which shows the 2nd processed reference image. It is a figure which shows the composite reference image. It is a figure which shows the part of the composite reference image enlarged. It is a figure for demonstrating the shaking comparison process. It is a figure which shows the defect area image. It is a figure which shows the defect region image by the processing of the comparative example. It is a figure which shows the defect region image by processing of another comparative example. It is a figure which shows the image of the setting area after alignment. It is a figure which shows two setting areas. It is a figure which shows the image of the setting area after alignment.

FIG. 1 is a diagram showing a configuration of an inspection device 1 according to an embodiment of the present invention. FIG. 2 is a plan view showing the main body 11 of the inspection device 1. The inspection device 1 is a device that inspects the appearance of a three-dimensional object 9 having a glossy surface. The object 9 is, for example, a metal part formed by forging or casting, and its surface is satin-finished with minute irregularities. The object 9 is, for example, various parts (cylindrical hub shaft, outer ring, yoke, etc.) used for a universal joint.

As shown in FIG. 1, the inspection device 1 includes a main body 11 and a computer 12. The main body 11 includes a stage 2, a stage rotating unit 21, an imaging unit 3, and a light source unit 4. The object 9 is placed on the stage 2. The stage rotating portion 21 rotates the object 9 together with the stage 2 by a predetermined angle about the central axis J1 facing in the vertical direction. The central axis J1 passes through the center of stage 2. The main body 11 is provided with a light-shielding cover (not shown) for preventing external light from reaching the stage 2, and the stage 2, the image pickup unit 3, and the light source unit 4 are provided in the light-shielding cover.

As shown in FIGS. 1 and 2, the image pickup unit 3 includes one upper image pickup unit 31, four oblique image pickup units 32, and four side image pickup units 33. In FIG. 2, the upper light source unit 31 is not shown (the same applies to the upper light source unit 41 described later). The upper imaging unit 31 is arranged on the central axis J1 above the stage 2. An image obtained by capturing an object 9 on the stage 2 from directly above can be acquired by the upper imaging unit 31.

As shown in FIG. 2, when the main body 11 is viewed from the upper side to the lower side (that is, when the main body 11 is viewed in a plan view), the four oblique imaging units 32 are arranged around the stage 2. .. The four oblique imaging units 32 are arranged at an angular interval (pitch) of 90 ° in the circumferential direction about the central axis J1. On the surface of each oblique imaging unit 32 including the imaging optical axis K2 and the central axis J1 (see FIG. 1), the angle θ2 formed by the imaging optical axis K2 and the central axis J1 is approximately 45 °. An image obtained by capturing an object 9 on the stage 2 from diagonally above can be acquired by each oblique imaging unit 32.

When the main body 11 is viewed in a plan view, the four lateral imaging units 33 are also arranged around the stage 2 in the same manner as the four oblique imaging units 32. The four lateral imaging units 33 are arranged at an angular interval of 90 ° in the circumferential direction. The angle θ3 formed by the imaging optical axis K3 and the central axis J1 on the surface of each side imaging unit 33 including the imaging optical axis K3 and the central axis J1 is approximately 90 °. An image obtained by capturing an object 9 on the stage 2 from the side can be acquired by each side imaging unit 33. The upper image pickup unit 31, the oblique image pickup section 32, and the side image pickup section 33 have, for example, a CCD (Charge Coupled Device), a CMOS (Complementary Metal-Oxide Semiconductor), or the like, and a multi-gradation image is acquired. The upper image pickup unit 31, the oblique image pickup section 32, and the side image pickup section 33 are supported by a support section (not shown).

The light source unit 4 includes one upper light source unit 41, eight oblique light source units 42, and eight side light source units 43. The upper light source unit 41 is a light source unit in which a plurality of LEDs (light emitting diodes) are arranged in a ring shape centered on the central axis J1. The ring-shaped upper light source unit 41 is fixed to the upper image pickup unit 31 so as to surround the periphery of the upper image pickup unit 31. The upper light source unit 41 can irradiate the object 9 on the stage 2 with light from directly above along the direction parallel to the central axis J1.

When the main body 11 is viewed in a plan view, the eight oblique light source portions 42 are arranged around the stage 2. The eight oblique light source units 42 are arranged at an angular interval of 45 ° in the circumferential direction. Each oblique light source unit 42 is a light source unit in which a plurality of LEDs are arranged in a bar shape extending in the tangential direction of the circumference centered on the central axis J1. When the line connecting the center of the emission surface of each oblique light source unit 42 and the object 9 (center) is called an "illumination axis", the surface including the illumination axis and the central axis J1 of the oblique light source unit 42 The angle formed by the illumination axis and the central axis J1 is approximately 45 °. Each oblique light source unit 42 can irradiate the object 9 on the stage 2 with light from diagonally above along the illumination axis. In the inspection device 1, four of the eight oblique light source units 42 are fixed to the four oblique light source units 32, and the remaining four oblique light source units 42 are shown in the figure. Supported by an abbreviated support.

When the main body 11 is viewed in a plan view, the eight side light source units 43 are arranged around the stage 2. The eight side light source units 43 are arranged at an angular interval of 45 ° in the circumferential direction. Each side light source unit 43 is a light source unit in which a plurality of LEDs are arranged in a bar shape extending in the tangential direction of the circumference centered on the central axis J1. Similar to the oblique light source unit 42, when the line connecting the center of the emission surface of each side light source unit 43 and the object 9 is called an "illumination axis", the illumination axis and the central axis J1 of the side light source unit 43 are called "illumination axes". The angle formed by the illumination axis and the central axis J1 on the surface including and is approximately 90 °. Each side light source unit 43 can irradiate the object 9 on the stage 2 with light from the side along the illumination axis. In the inspection device 1, four side light source units 43 out of eight side light source units 43 are fixed to each of the four side light source units 33, and the remaining four side light source units 43 are not shown. Supported by an abbreviated support.

For example, the distance between the upper image pickup unit 31 and the upper light source unit 41 and the object 9 is about 55 cm (centimeters). The distance between the oblique imaging unit 32 and the oblique light source unit 42 and the object 9 is about 50 cm, and the distance between the side imaging unit 33 and the side light source unit 43 and the object 9 is about 50 cm. It is 40 cm. In the upper light source unit 41, the oblique light source unit 42, and the side light source unit 43, a type of light source other than the LED may be used.

FIG. 3 is a block diagram showing a functional configuration realized by the computer 12. The computer 12 includes an inspection unit 5. The inspection unit 5 includes a target image storage unit 51, a reference image storage unit 52, an alignment unit 53, and a comparison unit 54. The target image storage unit 51 stores data of an image of the object 9 (hereinafter, referred to as “target image”) captured by the image pickup unit 3 (the image pickup unit). The reference image storage unit 52 stores the data of the image of the object 9 without defects (hereinafter, referred to as “reference image”) acquired by the same imaging unit as the target image. The alignment unit 53 aligns the target image and the reference image. The alignment unit 53 includes a first alignment processing unit 531, a second alignment processing unit 532, and a synthesis unit 533. The comparison unit 54 detects a defective region in the target image by comparing both the aligned images. The details of the processing by the inspection unit 5 will be described later. The computer 12 also plays a role as a control unit that controls the entire inspection device 1.

FIG. 4 is a diagram showing a flow of processing for inspection of the object 9 by the inspection device 1. First, the object 9 to be inspected is placed on the stage 2 (step S11). For example, a plurality of pins for alignment are provided on the stage 2, and by bringing a predetermined portion of the object 9 into contact with the plurality of pins, the object is placed at a predetermined position on the stage 2. The 9s are (ideally) arranged in a predetermined orientation. Subsequently, the imaging setting information for the object 9 on the stage 2 is acquired based on the input by the operator or the like (step S12). Here, the imaging setting information includes an imaging unit used in the imaging unit 3 (hereinafter referred to as “selective imaging unit”) and a light source unit (hereinafter referred to as “selective imaging unit”) that lights up in the light source unit 4 when the selective imaging unit acquires an image. , "Selected light source unit"). Here, it is assumed that one oblique imaging unit 32 is selected as the selective imaging unit, and the light source unit around the selective imaging unit is selected as the selective light source unit. Of course, other imaging units and light source units may be selected as the selective imaging unit and the selective light source unit. When the imaging setting information is acquired, the target image 71 (data) shown in FIG. 5 is acquired by using the selective light source unit and the selective imaging unit (step S13).

The target image 71 is a multi-gradation image showing the target object 9, and is stored and prepared in the target image storage unit 51. In FIG. 5, the same reference numeral 9 is attached to the object 9 in the target image 71 (the same applies to the reference image 8 in FIG. 6 described later). The object 9 has a shape in which a cylindrical portion 902 is provided in the center of a plate-shaped base portion 901. The object 9 at the time of imaging is in a direction rotated by a slight angle about the cylindrical portion 902 from a predetermined direction. As a result, in the target image 71, the corner side surface 903, which is originally a hidden portion (so-called occlusion), appears in the base portion 901. Further, a defect 904 exists at the edge of the base portion 901, and the length of the cylindrical portion 902 is also shorter than the design value (see FIG. 6 described later).

On the other hand, in the inspection device 1, a plurality of reference images (data) are acquired in advance by imaging another object 9 having no defects by using various combinations of a light source unit and an imaging unit. The reference image may be created from a plurality of images obtained by capturing a plurality of objects 9 having no defects, or may be created from the design data of the objects 9. The inspection unit 5 identifies the reference image acquired by using the same combination of the light source unit and the imaging unit as the target image 71, and outputs the reference image together with the target image 71 to the alignment unit 53.

FIG. 6 is a diagram showing a reference image 8 corresponding to the target image 71. In the first alignment processing unit 531 and the second alignment processing unit 532, the reference image 8 and the target image 71 are aligned. Here, the alignment (registration) is a process of moving a region showing each part of the object 9 to a position showing the part in the other image in one image of the target image 71 and the reference image. Yes, the alignment reduces the difference between the two images.

The first alignment processing unit 531 is required to obtain an optical flow for the target image 71 of each pixel of the reference image 8. Then, the reference image 8 is aligned with the target image 71 based on the optical flow, and the aligned reference image 81 (hereinafter, referred to as “first processed reference image 81”) shown in FIG. 7 is acquired. (Step S14).

Here, in the calculation of the optical flow between the reference image 8 and the target image 71, for example, "High Accuracy Optical Flow Estimation Based on a Theory for Warping" by Thomas Brox et al. (European Conference on Computer Vision, Springer LNCS 3024, The method described in May 2004, vol. 4, p. 25-36) (Non-Patent Document 1) can be used. In this method, an energy function that penalizes the assumption of brightness invariance, the assumption of brightness gradient invariance, and the deviation from the assumption of smoothness is used, and the vector that minimizes this energy is coarse-to- It is requested by using fine warping).

As described above, in the alignment processing by the first alignment processing unit 531, one pixel is used for microscopic alignment in pixels (that is, when searching for a similar portion in the other image). Alignment) is performed. As a result, the corner side surface 903 in the target image 71 that does not appear in the reference image 8 is expressed to some extent in the first processed reference image 81 by the alignment of the surrounding pixels in the reference image 8 having a relatively close brightness. Will be done. Further, in the first processed reference image 81, the length of the cylindrical portion 902 is also adjusted to some extent to the length in the target image 71. Other methods may be used in the calculation of the optical flow.

Further, in the second alignment processing unit 532, the reference image 8 is aligned with respect to the target image 71 by using the affine transformation, and the reference image 82 after the alignment shown in FIG. 8 (hereinafter, “second processed reference”). Image 82 ”) is acquired (step S15). In an example of alignment using the affine transformation, a plurality of feature points on the object 9 (for example, points on a line indicating the outer shape of the object 9) are predetermined in the reference image 8, and the target image 71 is the same. A plurality of positions indicating a plurality of feature points are obtained by pattern matching. Then, the values of the elements of the affine transformation matrix (here, the elements related to translation and rotation) are obtained by using the coordinates of the plurality of feature points in the reference image 8 and the coordinates of the plurality of positions in the target image 71. , The affine transformation is performed on the reference image 8 using the transformation matrix.

As described above, in the alignment processing by the second alignment processing unit 532, the macroscopic alignment (that is, the position where the unit region is a set of a large number of pixels) is set in units of a relatively large area surrounded by a plurality of feature points. Matching) is performed. As a result, between the reference image 8 before processing and the reference image 82 after the second processing, the dimensions of each part of the object 9 (for example, the length and diameter of the cylindrical portion 902, the width of the base portion 901, etc.) are approximately. Be maintained. The area surrounded by the plurality of feature points also includes an area outside the setting area A1 described later. On the other hand, in the second processed reference image 82, the region corresponding to the corner side surface 903 that does not appear in the reference image 8 (the region indicated by the same reference numeral 903 in FIG. 8) is, for example, a background.

In the reference image 8 of FIG. 6, the setting area A1 indicated by the broken line rectangle is predetermined by the input of the operator or the like, and in the synthesis unit 533, the setting area A1 of the first processed reference image 81 of FIG. 7 is set. The portion is extracted and replaced with the portion of the setting area A1 in the second processed reference image 82 of FIG. That is, the set area A1 after the alignment by the first alignment processing unit 531 is combined with the second processed reference image 82 after the alignment by the second alignment processing unit 532. As a result, the composite reference image 83 shown in FIG. 9 is acquired (step S16).

The comparison unit 54 detects a defective region in the target image 71 by comparing the target image 71 with the composite reference image 83 (step S17). Here, as shown in FIG. 10 in which the region A0 surrounded by the alternate long and short dash line in FIG. 9 is enlarged, the edge of the object 9 may be slightly displaced at the boundary of the setting region A1. This is because when the setting area A1 of the first processed reference image 81 is combined with the second processed reference image 82, the positions of the objects 9 shown by both images do not always completely match. Therefore, the comparison unit 54 performs a shaking comparison process.

FIG. 11 is a diagram for explaining the shaking comparison process, and shows a plurality of (nine) positional relationships between the target image 71 and the composite reference image 83 side by side in 3 rows and 3 columns. In the shaking comparison process, the target image 71 shown by the broken line rectangle in FIG. 11 is completely overlapped with the composite reference image 83 shown by the solid line rectangle (see the center in FIG. 11) in each of the eight directions. The difference between the value of the pixel and the value of the pixel of the composite reference image 83 that overlaps with the pixel for each pixel of the target image 71 after moving by a predetermined number of pixels (for example, one pixel). Absolute value) is required. In other words, a difference image between the moved target image 71 and the composite reference image 83 is obtained. Similarly, a difference image is obtained at a position where the target image 71 completely overlaps with the composite reference image 83.

Then, in a plurality of (here, nine) difference images, the minimum value of the pixel value at the same position is specified, and the image to which the minimum value is given to the position is acquired as the result image of the shaking comparison process. Will be done. The result image is binarized at a predetermined threshold value, and a binary defect candidate image indicating a defect candidate region is acquired. In the comparison unit 54, among the defect candidate regions shown by the defect candidate image, those satisfying a predetermined condition (for example, an area or the like) are specified as the defect region. As a result, as shown in FIG. 12, the defect region image 61 showing the defect region 611 is acquired, and the defect region in the target image 71 is detected. In FIG. 12, the object 9 shown by the object image 71 is shown by a chain double-dashed line (the same applies to FIGS. 13 and 14 described later). The steps S11 to S17 are repeated for the other object 9. The result image obtained by the shaking comparison process may be derived from the above-mentioned plurality of difference images. For example, the difference image having the smallest sum of pixel values among the plurality of difference images is the result image as it is. May be used as.

Here, the processing of a comparative example in which only the second alignment processing unit 532 is used (only the affine transformation is used) when the target image 71 and the reference image 8 are aligned will be described. In the processing of the comparative example, when the defective region is detected, the defective region image 91 shown in FIG. 13 is acquired by comparing the second processed reference image 82 of FIG. 8 with the target image 71 of FIG. Will be done. In the defect region image 91 of FIG. 13, similarly to the defect region image 61 of FIG. 12, the defect region 911 existing at the edge of the base portion 901 and the defect region 911 indicating that the length of the cylindrical portion 902 is short are detected. Will be done. On the other hand, as described above, in the second processed reference image 82, the region corresponding to the corner side surface 903 that does not appear in the reference image 8 becomes, for example, the background. Therefore, in the defect region image 91, the corner side surface 903 that is not a defect is detected as the defect region 911, and the defect is over-detected (a false defect is detected). Further, due to the presence of the corner side surface 903 in the target image 71, the reference image 8 and the target image 71 may be misaligned by the affine transformation, and many false defects may be detected. The problem is that when the reference image 8 is acquired by imaging another object 9 having no defects, the object 9 is slightly rotated from a predetermined direction, that is, the reference image 8. The same applies when occlusion occurs in.

Further, when aligning the target image 71 and the reference image 8, processing of another comparative example in which only the first alignment processing unit 531 is used (only the optical flow is used) will be described. In the processing of the other comparative example, when the defect region is detected, the first processed reference image 81 of FIG. 7 and the target image 71 of FIG. 5 are compared, so that the defect region image 92 shown in FIG. 14 is compared. Is obtained. In the defect region image 92 of FIG. 14, the defect region 921 existing at the edge of the base portion 901 is detected as in the defect region image 61 of FIG. Further, as described above, the corner side surface 903 that does not appear in the reference image 8 is represented to some extent in the first processed reference image 81 by the surrounding pixels having relatively close brightness in the reference image 8. Therefore, in the defect candidate image obtained by the processing of the other comparative example, the defect candidate region corresponding to the corner side surface 903 becomes relatively small, and detection as a defect region is suppressed. On the other hand, in the first processed reference image 81, the length of the cylindrical portion 902 is adjusted to some extent to the length of the target image 71, so that there is a defect region indicating that the length of the cylindrical portion 902 is short in the target image 71. It will not be detected. That is, a defect related to the length of the cylindrical portion 902 is overlooked. In FIG. 14, the region 922 showing the defect is shown by a broken line.

On the other hand, in the inspection device 1, both the first alignment processing unit 531 and the second alignment processing unit 532 are used when the target image 71 and the reference image 8 are aligned. In the first alignment processing unit 531, by obtaining the optical flow for each pixel of the reference image 8 with respect to the target image 71, a plurality of (all) pixels of the reference image 8 are individually aligned with respect to the target image 71. NS. In the second alignment processing unit 532, the region of the object 9 in the reference image 8 is aligned with respect to the target image 71 as a whole by using the affine transformation. Then, the setting area A1 after the alignment by the first alignment processing unit 531 is combined with the reference image (second processed reference image 82) after the alignment by the second alignment processing unit 532.

In the above processing, for the setting area A1 in which false defects are likely to occur due to the rotation of the object 9 or the allowable range of dimensional defects of the object 9 is large, the alignment by the first alignment processing unit 531 is substantially performed. It is done in. Further, for other regions where the occurrence of false defects due to the rotation of the object 9 is unlikely to occur (in the above example, the cylindrical portion 902 which is a rotating body) or the allowable range of dimensional defects of the object 9 is small, the first The alignment by the two alignment processing unit 532 is substantially performed. In this way, by aligning the target image 71 and the reference image 8 individually (that is, in unit areas of different sizes) in the setting area A1 and the areas other than the setting area A1, each area is set. Appropriate alignment can be performed appropriately, and a preferable inspection can be performed.

Further, in the comparison unit 54, a plurality of difference images are obtained from both images in a plurality of relative positional relationships in which the target image 71 is shifted vertically and horizontally with respect to the reference image (composite reference image 83) after alignment. The defect region is detected based on the plurality of difference images. As a result, it is possible to suppress the occurrence of false defects caused by the misalignment between the first processed reference image 81 and the second processed reference image 82.

The inspection device 1 can be modified in various ways.

The first alignment processing unit 531 may perform processing only on the setting area A1 in the reference image 8 to acquire the image 81a of the setting area A1 after the alignment as shown in FIG. In this case, the image 81a of the setting area A1 is combined with the entire processed image (second processed reference image 82) of the reference image 8 by the second alignment processing unit 532, and the composite reference image 83 similar to FIG. 9 is combined. Is obtained. Further, the second alignment processing unit 532 performs processing only on the area other than the setting area A1 in the reference image 8, and an image showing the area after the alignment is acquired, and the first processed reference image 81 is obtained. May be synthesized in.

Here, the amount of calculation related to the alignment of regions of the same size is larger in the first alignment processing unit 531 than in the second alignment processing unit 532. Therefore, in order to reduce the amount of calculation in the alignment unit 53, the second alignment processing unit 532 processes all of the reference image 8, and the first alignment processing unit 531 performs a part of the reference image 8. It can be said that it is preferable to perform processing only on the area (setting area A1).

In the above processing example, in the reference image 8, one setting area A1 mainly including the base portion 901 is provided, but as shown in FIG. 16, another setting area A2 mainly including the cylindrical portion 902 is a setting area. It may be provided in addition to A1. In this case, for example, the second alignment processing unit 532 performs processing only on the setting area A2 in the reference image 8 (specifically, the alignment is performed by affine transformation based on a plurality of feature points in the setting area A2). Then, as shown in FIG. 17, the image 82a of the set area A2 after the alignment is acquired. Then, the image 81a of the setting area A1 after the alignment of FIG. 15 and the image 82a of the setting area A2 after the alignment of FIG. 17 are combined to obtain the reference image after the alignment.

Further, in the setting area A1 and the setting area A2 (hereinafter, referred to as "first setting area A1" and "second setting area A2", respectively), comparison with the target image 71 may be performed individually. For example, by comparing the image 81a of the first setting area A1 after the alignment in FIG. 15 with the target image 71, the comparison result for the first setting area A1 (the portion of the first setting area A1 of the defect area image). Is obtained. Further, by comparing the image 82a of the second setting area A2 after the alignment in FIG. 17 with the target image 71, the comparison result for the second setting area A2 (the portion of the second setting area A2 of the defect area image). Is obtained. In this case, the amount of calculation can be significantly reduced.

In the above processing example of acquiring the first processed reference image 81 and the second processed reference image 82, the setting area A1 in the reference image 8 is regarded as the first setting area A1, and all the areas other than the setting area A1 are the first. 2 It is also possible to regard it as a setting area. In this case, it can be said that the first processed reference image 81 is an image in which a plurality of pixels included in the first setting area A1 of the reference image 8 are individually aligned with respect to the target image 71. Further, it can be said that the second processed reference image 82 is an image in which the region of the object 9 included in the second setting region of the reference image 8 is aligned with respect to the target image 71.

In the alignment unit 53, the target image 71 may be aligned with respect to the reference image 8. In this case, the first setting area and the second setting area are set in the target image 71, and a plurality of pixels included in the first setting area of the target image 71 are set with respect to the reference image 8 by the first alignment processing unit 531. Are individually aligned. Further, the second alignment processing unit 532 aligns the area of the object 9 included in the second setting area of the target image 71 with respect to the reference image 8.

The first alignment processing unit 531 divides, for example, the first setting area of one of the target image 71 and the reference image 8 to acquire a plurality of areas, and each area (for example, a set of several pixels). The image of the first setting region after the alignment may be acquired by aligning the image with respect to the other image using the affine transformation. In this way, by individually aligning a plurality of regions (which may be pixels) included in the first setting region of one image with respect to the other image, the first setting region can be used. Microscopic alignment (alignment in units of minute regions) can be realized. If a gap is generated between the plurality of regions in the first setting region after alignment, interpolation processing for filling the gap is appropriately performed (the same applies to other processes in which the gap is generated).

In the second alignment processing unit 532, various methods other than the affine transformation may be adopted as long as the macroscopic alignment with respect to the second setting region is realized. Here, the alignment in the second setting area by the second alignment processing unit 532 may be performed by using a region larger than the unit of alignment by the first alignment processing unit 531 as the unit of alignment. That is, in the second alignment processing unit 532, a plurality of regions included in the second setting area of one of the target image 71 and the reference image 8 and being the unit of alignment in the first alignment processing unit 531. Areas larger than any of are aligned with respect to the other image. In the inspection device 1, the target image 71 and the reference image 8 are aligned in unit areas having different sizes in the first and second setting areas, so that the alignment according to each setting area is appropriately performed. , A favorable inspection can be realized.

A plurality of first setting areas may be set in the alignment unit 53, and a plurality of second setting areas may be set in the same manner. Further, in the region of the object 9 in the target image 71, a non-inspection region in which the defect is not inspected may be set.

In the shaking comparison process in the comparison unit 54, the composite reference image 83 may be moved with respect to the target image 71. That is, in the shaking comparison process, after the target image 71 and the reference image 8 are aligned, one of the two images is shifted vertically and horizontally with respect to the other in a plurality of relative positional relationships. The difference image of is obtained. Depending on the design of the comparison unit 54, the two images may be compared without performing the shaking process. Further, the comparison between the aligned target image 71 and the reference image 8 may be performed by, for example, obtaining the ratio of the values of the corresponding pixels of both images, in addition to the method of obtaining the difference image of both images. ..

Depending on the memory area of the comparison unit 54, the first defect area image showing the comparison result between the target image 71 and the first processed reference image 81 and the comparison result between the target image 71 and the second processed reference image 82 may be displayed. The second defect region image shown may be acquired. In this case, for example, the defect region image is acquired by synthesizing the portion of the setting region A1 in the first defect region image with the second defect region image.

The inspection target in the inspection device 1 may be a plate-shaped object, a film-shaped object, or the like, in addition to the three-dimensional object 9.

The above-described embodiment and the configurations in each modification may be appropriately combined as long as they do not conflict with each other.

1 Inspection device 8 Reference image 9 Target image 51 Target image storage unit 52 Reference image storage unit 53 Alignment unit 54 Comparison unit 71 Target image 82 Second processed reference image 83 Synthetic reference image 531 First alignment processing unit 532 Second Alignment processing unit 611 Defect area A1 First setting area A2 Second setting area S11 to S17 Steps

Claims (10)

It ’s an inspection device,
A target image storage unit that stores the target image obtained by capturing the target object,
A reference image storage unit that stores a reference image,
An alignment unit that aligns the target image and the reference image, and
A comparison unit that detects a defective region in the target image by comparing both images after alignment, and a comparison unit.
With
The alignment part
A first alignment processing unit that individually aligns a plurality of regions obtained by dividing the first setting area of one image of the target image and the reference image with respect to the other image.
A second alignment processing unit that aligns a region included in a second setting region different from the first setting region in the one image and larger than any of the plurality of regions with respect to the other image. When,
An inspection device comprising. The inspection device according to claim 1.
The comparison unit obtains a plurality of difference images from the two images in a plurality of relative positional relationships in which one of the two images after the alignment is shifted vertically and horizontally with respect to the other, and is based on the plurality of difference images. An inspection device characterized by detecting the defect region. The inspection device according to claim 1 or 2.
An inspection apparatus characterized in that the second alignment processing unit aligns the other image in the second setting region by using an affine transformation. The inspection device according to claim 3.
The second alignment processing unit acquires the entire processed image of the one image by using the affine transformation.
The first set area after alignment by the first alignment processing unit is combined with the processed image, and the image is combined with the processed image .
Inspection device the second setting area which is an area other than the first prescribed area in the processed image after the synthesis is characterized that you have been aligned relative to the other image. The inspection device according to any one of claims 1 to 4.
An inspection apparatus characterized in that the first alignment processing unit performs alignment by obtaining an optical flow of each pixel in the first setting region with respect to the other image. It ’s an inspection method,
a) The process of preparing an object image that is an image of an object, and
b) A step of aligning the target image and the reference image, and
c) A step of detecting a defective region in the target image by comparing both images after alignment, and
With
The step b)
b1) A step of individually aligning a plurality of regions obtained by dividing the first setting region of one image of the target image and the reference image with respect to the other image.
b2) A step of aligning a region included in a second setting region different from the first setting region in the one image and larger than any of the plurality of regions with respect to the other image.
An inspection method characterized by comprising. The inspection method according to claim 6.
In the step c), a plurality of difference images are obtained from the two images in a plurality of relative positional relationships in which one of the two images after the alignment is shifted vertically and horizontally with respect to the other, and the plurality of difference images are obtained. An inspection method characterized in that the defect region is detected based on the above. The inspection method according to claim 6 or 7.
The inspection method, characterized in that, in the step b2), the alignment with respect to the other image in the second setting region is performed by using the affine transformation. The inspection method according to claim 8.
In the step b2), the entire processed image of the one image is acquired by using the affine transformation.
The first set area after the alignment in the step b1) is combined with the processed image .
Inspection method the second set area which is an area other than the first setting region is characterized that you have been aligned relative to the other images in the processed image after the synthesis. The inspection method according to any one of claims 6 to 9.
The inspection method, characterized in that, in the step b1), alignment is performed by obtaining an optical flow of each pixel in the first setting region with respect to the other image.

Images