JP2018205123A - Image generating apparatus and image generating method for generating inspection image for performance adjustment of image inspection system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an apparatus and a method for generating an image (simulated inspection image) used for adjusting (evaluating, learning, etc.) the performance of an image inspection system for inspecting scratches or defects on the surface of an inspection object. An image generation device that generates a simulated inspection image used for adjusting (evaluating, learning, etc.) the performance of an image inspection system is an original image (32) obtained by photographing the surface of an inspection object and a shape of a scratch or a defect. A storage unit for storing a defect image (39), which is an image showing the above, and an image generation unit for generating a simulated inspection image are provided. The image generation unit generates a base image (36) having texture characteristics similar to the original image from the original image (32) obtained by photographing the surface of the inspection object, synthesizes the base image with the defect image (39), and performs a simulated inspection. Generate an image (40). [Selection diagram] FIG. 7

Description

  The present disclosure relates to an image generation technique for generating an image for adjustment (performance evaluation, machine learning, etc.) for an image inspection system that inspects an object surface for scratches and defects using an image.

  There is an image inspection system that analyzes image data obtained by photographing the surface of an inspection object (work, shape material, etc.) and detects defects such as scratches and defects on the inspection object surface (see Patent Documents 1 and 2, etc.) ).

  In the development of such an image inspection system, the cause, shape, color, etc. of wrinkles are analyzed to formulate appropriate imaging conditions, and at the same time, an image processing algorithm for detection and discrimination is constructed. In other words, the first step in development is to fully understand the phenomenon to be detected and the visible state. In the development process, the system needs to be evaluated, and its reliability is guaranteed by a comprehensive and sufficient amount of test data.

  The problem here is that sample images containing defects necessary for developing an image processing algorithm may not be sufficiently collected during development. For example, it is difficult to obtain a sample image including a defect because the production line is originally designed to suppress the generation of wrinkles, and defects are not essentially generated before the line is operated.

  In addition, with the recent development of machine learning (deep learning) technology, a large amount of sample images are required for the learning. Conventionally, a simulated defect in an inspection image is often produced by machining or the like. However, with actual image data, it is difficult to secure a sufficient number of sample images for machine learning. Therefore, a method of generating sample images by digital image processing is employed. However, generally, an image in which the position, direction, size, sharpness, etc. of the original sample defect image is changed is merely added as a sample image.

Special table 2006-532121 gazette JP 2011-214903 A

Hiroshi Murase, “Generative Learning for Image Recognition”, A theory, Vol.46, No.SIG 15, pp.35-42 (Oct. 2005) Narumi Tsuchiya, Satoshi Yamauchi, Hironobu Fujiyoshi, “Generative Learning for Human Detection and Efficiency of Learning with Negative-Bag MILBoost”, MIRU2012, 2012.8

  In addition to appearance inspection, techniques for generating and adding such sample images by image processing techniques are conventionally known (see Non-Patent Documents 1 and 2).

  However, in appearance inspection, the degree of difficulty in detecting scratches and defect images existing on the surface of the inspection object (roughness, unevenness, shading, etc.) varies greatly. That is, in the generation of the sample image in the above-mentioned document, it is difficult to accurately represent the appearance derived from the cause of the scratch / defect while ensuring the variation of the non-defective substrate.

  The present invention provides an apparatus and method for generating an image used for performance evaluation and machine learning of an image inspection system that performs image inspection of the surface of a shaped material such as casting, forging, pressing, and resin molding.

  In a first aspect of the present invention, there is provided an image generation apparatus that generates a simulated inspection image used for adjusting the performance of an image inspection system that inspects a scratch or a defect on the surface of an inspection object using an image. The image generation apparatus includes a storage unit that stores an original image obtained by photographing the surface of the inspection object, a defect image that is an image indicating a shape of a scratch or a defect, and an image generation unit that generates a simulated inspection image. The image generation unit generates a base image having a texture characteristic similar to that of the original image from the original image obtained by photographing the inspection object surface, and generates a simulated inspection image by synthesizing the defect image with the base image.

  In a second aspect of the present invention, there is provided an image generation method for generating a simulated inspection image to be used for adjusting the performance of an image inspection system that inspects an inspection object surface for scratches or defects using an image. . The image generation method includes a first step of storing in a storage unit an original image obtained by photographing the surface of an inspection object and a defect image that indicates a shape of a flaw or a defect, and the control unit from the original image A second step of generating a base image having a texture feature similar to the original image, and a third step of synthesizing the defect image with the base image by the control unit to generate a simulated inspection image are included.

  In a third aspect of the present invention, a program for causing a computer to execute the above-described image generation method is provided.

  According to the present invention, it is possible to artificially generate a large amount of images having texture characteristics similar to the original texture characteristics of the surface of the inspection object as inspection images used for performance evaluation and learning of the image inspection system. it can. Therefore, for example, to stably supply sample images necessary for performance evaluation and machine learning of image systems (image inspection software) for image inspection of the surface of a shaped material such as casting, forging, pressing, resin molding, etc. Can do.

The figure which shows the example of the mock test | inspection image produced | generated by the image production | generation apparatus of one Embodiment of this invention. 1 is a block diagram showing the configuration of an image generation apparatus according to an embodiment of the present invention. The figure which shows the program and data which are stored in the data storage device (A) The figure which shows the example of a base original image, (B) The figure which shows the example of a base seed image Diagram showing examples of shading patterns Diagram showing examples of scratch / defect representative patterns The figure explaining the flow of generation processing of a mock inspection image Flow chart showing the base image composition processing The figure explaining the substrate image generated by texture synthesis from the substrate seed image A diagram illustrating a base image synthesized from various base seed images Flow chart showing flaw / defect image composition processing Diagram explaining deformation process of scratch / defect representative pattern Illustration explaining the shadow added to the scratch / defect image A diagram explaining an example of a composite flaw / defect image generated from a silhouette image Flowchart showing synthesis process of simulated inspection image by image generation device The figure which showed the example of the mock test | inspection image produced | generated by the image generation apparatus

  Hereinafter, an embodiment of an image generating apparatus according to the present invention will be described with reference to the drawings as appropriate.

(Embodiment 1)
The image generation apparatus according to the present embodiment uses an image to evaluate the performance of an image inspection system that inspects scratches and defects on the surface to be inspected or to perform machine learning (deep learning) of the image inspection system ( Hereinafter, it is a device that generates a “simulated inspection image”. FIG. 1 shows an example of a simulated inspection image generated by the image generation apparatus. As shown in FIG. 1, the image generating apparatus includes a simulated inspection image 40, a scratch / defect image (foreground image) 39 indicating a scratch or a defect, and a base image (background image) 36 indicating a base state of the inspection target. And is generated.

1-1. Configuration of Image Generation Device FIG. 2 is a diagram illustrating a configuration of an image generation device according to an embodiment of the present invention. The image generating apparatus 10 can be configured by an information processing apparatus such as a personal computer. Specifically, the image generation apparatus 10 includes a controller 11 that controls the overall operation, a display unit 13 that displays various information, an operation unit 15 for a user to input instructions, a RAM 16, data, And a data storage unit 17 for storing the program. Furthermore, the image generation apparatus 10 includes a first interface 18 for connecting to a network and a second interface 19 for connecting to an external device.

  The display unit 13 is configured by, for example, a liquid crystal display or an organic EL display. The operation unit 15 is a device for a user to perform an input operation on the image generation device 10 and includes a keyboard, a mouse, a touch panel, buttons, and the like.

  The first interface 18 is an interface circuit (module) for connecting to a network by wire or wireless. The first interface 18 communicates with a network according to a predetermined standard (IEEE802.3, IEEE802.11, WiFi, etc.).

  The second interface 19 is an interface for connecting to an external device. For example, the second interface 19 is a circuit (module) for performing communication according to the USB, HDMI (registered trademark), or Bluetooth (registered trademark) standards. For example, a printer and an external storage device (HDD, SSD, memory card, optical disk device, etc.) are connected to the image generation device 10 via the second interface 19.

  The RAM 16 is composed of a semiconductor device such as a DRAM or SRAM, for example, and temporarily stores data and functions as a work area for the controller 11.

  The data storage unit 17 is a recording medium that stores parameters, data, a control program, and the like necessary for realizing a predetermined function. The data storage unit 17 can be composed of, for example, a hard disk (HDD), a semiconductor storage device (SSD), and / or an optical disk device.

  The controller 11 (an example of an image generation unit and a control unit) implements an image generation function to be described later by executing a control program 30 (software). In the present embodiment, an image generation program is installed in advance in the data storage unit 17 as a control program. The controller 11 implements the functions described later by executing this image generation program.

  The control program executed by the controller 11 may be provided by a recording medium such as a DVD-ROM or CD-ROM, or may be downloaded from a server on the network via a communication line. In the present embodiment, the function of the controller 11 is realized by the cooperation of hardware and software (application program). However, the function is realized only by a hardware circuit designed exclusively for realizing a predetermined function. May be. Therefore, the controller 11 can be composed of not only a CPU and MPU, but also a DSP, FPGA, ASIC, and the like.

1-1-1. Various Image Data FIG. 3 is a diagram illustrating programs and data stored in the data storage unit 17. The data storage unit 17 stores an image generation program 30 that is a control program for realizing the functions of the image generation apparatus 10 and various image data 33 to 39 used for image generation. Note that these image data 33 to 39 may be provided from a server on the network connected via the first interface 18 or an external storage device connected via the second interface 19. The data storage unit 17 also stores simulated inspection image data 40 that is finally generated.

  FIG. 4B shows an example of a base seed image generated from the original image. The base seed image 33 is an image used for generating the base image 36. The base seed image 33 is generated by cutting out a part of the rectangular area from the original image obtained by photographing the actual surface of the inspection object. FIG. 4A shows an example of an original image obtained by photographing the surface of a frying pan as an inspection object. A part of the rectangular regions R1 to R3 is cut out from the base image 32 shown in FIG. 4A, and base seed images 33a to 33c as shown in FIG. 4B are generated. In selecting this rectangular area, the state of shading may be taken into consideration. Depending on the shading state of the rectangular area, it is possible to generate shading in the simulated inspection image 40 after synthesis. The position of the rectangular area can be selected as appropriate.

  FIG. 5 is a diagram illustrating an example of the shading pattern 34. The shading pattern 34 is an image including only shading information. It is an image for giving a shadow (shading) in the final simulated inspection image 40. The shading pattern 34 is used to artificially represent the shade due to the illumination direction, the inclination of the component, and the shape of the component, which are included in the actual inspection image.

  FIG. 6 shows an example of the scratch / defect representative pattern 35. The image generating apparatus 10 holds an image showing representative silhouette shapes (two-dimensional shapes) of various scratches and defects in the data storage unit 17 as a scratch / defect representative pattern 35. A plurality of scratch / defect images 39 can be generated from each of the representative patterns 35 shown in FIG. As the shape of the scratch / defect representative pattern 35, for example, the shape described in the contaminant measurement chart can be used.

1-2. Operation of Image Generation Device The operation of the image generation device 10 configured as described above will be described below.

1-2-1. Overview FIG. 7 is a diagram for explaining an overview of the flow of processing for generating a simulated inspection image 40 by the image generation apparatus 10 according to the present embodiment. As described above, conventionally, as a method of generating a simulated inspection image used for evaluation and learning of an inspection system for appearance inspection, a self-determined image is directly synthesized with a background image, or a foreground and a background are converted simultaneously. There were many things. On the other hand, in the present embodiment, in the generation of the simulated inspection image, the foreground (scratch / defect image 39) and the background (background image 36) are respectively synthesized and combined to generate the simulated inspection image 40. To do.

1-2-2. Composition of Base Image An explanation will be given of the composition processing of the base image 36 as a background in the simulated inspection image. The surface of the inspection object has a texture derived from the manufacturing method of the inspection object. In other words, an image obtained by imaging the surface of the inspection target has the same characteristics with respect to the texture characteristics, although the image data differs for each region. In the present embodiment, the texture synthesis method of computer graphic technology is applied to the generation of a background image as a background. As the texture synthesis method, there are a pixel-based method, a patch-based method, an optimization-based method, a coordinate-based method, and the like. In this embodiment, the texture synthesis method is described in “Fast Texture Synthesis using Tree-structured Vector Quantization”, Li-Yi Wei, and M. Levoy, SIGGRAPH (2000) (hereinafter referred to as “Non-Patent Document 1”). The pixel-based method is used. Which texture synthesis method is used may be determined depending on the target appearance inspection.

  With reference to the flowchart of FIG. 8, the composition process of the base image 36 by the image generation apparatus 10 will be described. It is assumed that the image generation apparatus 10 stores the image data of the generated raw original image 32 obtained by photographing the inspection target surface in the data storage unit 17 or the RAM 16 before starting this processing.

  The user sets a rectangular area in the area of the base original image 32 via the operation unit 15. Only one rectangular area or a plurality of rectangular areas may be set. It is possible to arbitrarily set which rectangular region is selected or combined in the base image 32 at the time of combining. The controller 11 reads the rectangular area set by the user (S21), cuts out the rectangular area from the raw original image 32, and generates the basic seed image 33 (S22). The cut-out base seed image 33 is stored in the data storage unit 17 (or RAM 16). The controller 11 generates a base image 36 from the base seed image 33 using a texture synthesis method (S23). In the method of the present embodiment, since the base seed image 33 is not enlarged or reduced, the base texture information does not deteriorate, and a base image 36 having a texture characteristic close to that of the original base can be obtained.

  FIG. 9 is a diagram illustrating an example of a base image 36 generated from the base seed image 33 by texture synthesis. As shown in the figure, a texture similar to the texture of the base seed image 33 but different from that of the base seed image 33 can be obtained by appropriately setting the texture synthesis parameters from one base seed image 33 (see FIG. 9A). A plurality of base images 36 (see FIG. 9B) are generated.

  FIG. 10 is a diagram illustrating an example of the base image 36 synthesized by texture synthesis from the base seed image 33 having various textures. FIG. 10A shows an example of a base image 36 generated from a base seed image 33 having a cast skin texture. FIG. 10B shows an example of a base image 36 generated from a base seed image 33 having a cast skin texture. FIG. 10C shows an example of a substrate image 36 generated from a substrate seed image 33 having a fiber texture. FIG. 10D shows an example of a base image 36 generated from a base seed image 33 having a plywood texture.

1-2-3. Scratch / Defect Image Combining The flaw / defect image 39 combining process will be described. The scratch / defect image 39 is generated by deforming the scratch / defect representative pattern 35. Furthermore, there are various phenomena such as scratches, dents and the like that change the surface three-dimensionally, stains and dirt that adhere to the surface, and wrinkles that are buried on the surface. Therefore, in the present embodiment, first, attention is focused on the silhouette shape of a flaw / defect. That is, an image (scratch / defect silhouette image) 38 showing a two-dimensional shape of a scratch / defect is generated. If an image 38 showing a two-dimensional shape can be generated, shading and unevenness are added to the inside of the figure according to the assumed state of wrinkles. The use of the texture synthesis method in computer graphic technology is also conceivable in the synthesis of scratch / defect images. In particular, a technique for synthesizing image variations slightly different from any sample image from several sample images having a roughly common structure is effective.

FIG. 11 is a flowchart showing a process for combining the flaw / defect image 39.
First, the controller 11 deforms each of the scratch / defect representative patterns 35 to generate a plurality of scratch / defect silhouette images 38 (S31). Specifically, each scratch / defect representative pattern 35 is subjected to an affine transformation process to generate a plurality of scratch / defect silhouette images 38 having different shapes, sizes, and orientations. For example, as shown in FIG. 12, a plurality of different scratch / defect silhouette images 38a to 38d,... Are generated by transforming one scratch / defect representative pattern 35b by affine transformation. In other words, more scratch / defect silhouette images 38 can be generated from the scratch / defect representative pattern 35.

  Finally, the controller 11 adds a shadow or unevenness to each of the flaw / defect silhouette images 38 to generate a flaw / defect image 39 (S32). Here, with reference to FIG. 13, the shadow added to an image is demonstrated.

  FIGS. 13A to 13D are diagrams illustrating the difference in the appearance of scratches (or defects) 64 formed on the surface of the inspection object 61 due to the positional relationship among the inspection object 61, the viewpoint 62, and the illumination 63. FIG. It is.

  Depending on the direction and angle of the illumination 63 with respect to the inspection object 61 and the position of the viewpoint 62, there are many patterns of how the scratches 64 appear. FIG. 13 illustrates four patterns. FIG. 13A shows how the scratch 64 looks when the scratch 64 is viewed from an oblique direction with respect to the inspection object 61. In this case, as shown in the image 39a, the scratch 64 looks white on the left side of the scratch and black on the right side. FIG. 13B shows an example of how the scratch 64 is seen when viewed from the opposite direction to FIG. In this case, as shown in the image 39b, the scratches appear black on the left side of the scratches and whitish on the right side. FIG. 13C illustrates an example of how the inspection target 61 is viewed when the inspection target 61 is tilted and the scratch 64 is viewed from substantially the front of the inspection target 61. In this case, as shown in the image 39c, the scratch 64 appears whiter toward the center of the scratch. FIG. 13D shows an example of how the inspection object 61 is made of a material different from that shown in FIG. In this case, as shown in the image 39d, the scratches appear black at the center of the scratches.

  As described above, even if the scratch 64 has the same shape, depending on the relative positional relationship between the illumination 63 and the viewpoint 62 with respect to the inspection object 61, the way in which the shadow 64 is shaded differs and the appearance of the scratch 64 is different. . Therefore, in consideration of such shadows, the scratch / defect image 39 is generated by adding various shadows to the scratches 64 having the same shape. FIG. 14 shows an example of a flaw / defect image 39 generated by the above method. In the figure, scratch / defect images 39x and 39y with different shading and unevenness are generated from the scratch / defect silhouette image 38.

1-2-4. Synthesis of Simulated Inspection Image When the background image 36 and the scratch / defect image 39 as the background are generated as described above, the simulated inspection image 40 is combined by combining the images 36 and 39 and the shading pattern 34. Generate.

  With reference to the flowchart of FIG. 15, the generation operation of the simulated inspection image 40 in the image generation apparatus 10 will be described. It is assumed that the base image 36 and the flaw / defect image 39 are stored in the data storage unit 17 before this processing is started.

  The controller 11 of the image generation apparatus 10 acquires one of the plurality of base images 36 stored in the data storage unit 17 (S11). Furthermore, the controller 11 acquires one of the plurality of shading patterns 34 stored in the data storage unit 17 (S12). Further, the controller 11 acquires one of the plurality of scratch / defect images 39 stored in the data storage unit 17 (S13). Then, the controller 11 combines the acquired base image 36, the scratch / defect image 39, and the shading pattern 34 to generate a simulated inspection image 40 (S14).

  When the scratch / defect image 39 is synthesized (superimposed) on the base image 36, the brightness near the defect / defect image 39 and its outline is adjusted by mask processing so that the foreground (scratch / defect image) does not stand out. For example, a technique that blends the foreground and background with Poisson Blending is known. The superposition method may be appropriately changed depending on the type of scratch or defect.

  In the above description, the method of combining the base image 36 and the flaw / defect image 39 after combining the images has been described. However, both images may be combined in an intermediate stage of the combining processing of the respective images. Specifically, a scratch / defect image with shadows / unevenness added to the initial random noise image in texture synthesis is superimposed first, and the background and the foreground are synthesized simultaneously. This makes it possible to generate a simulated inspection image 40 with a natural impression in which the boundary between the base image 36 and the scratch / defect image 39 is not conspicuous.

  As described above, two methods for synthesizing the base image 36 and the scratch / defect image 39 have been presented. However, the selection of each method and the setting of parameters may be appropriately changed depending on the type of defect (scratch, defect).

  Thereafter, the controller 11 determines whether or not the simulated inspection image 40 has been generated for all combinations of the base image 36, the scratch / defect image 39, and the shading pattern 34 stored in the data storage unit 17 (S15). . If the simulated examination image 40 has not been generated for some combinations (NO in S15), the controller 11 returns to step S11 and performs the above-described processing of steps S11 to S14 for other combinations. The controller 11 performs the processes of steps S11 to S14 described above until the simulated inspection images 40 are generated for all combinations (YES in S15).

  As a result of the above processing, when l base images 36, m scratch / defect images 39, and n shading patterns 34 are stored in the data storage unit 17, finally, l × m × n simulation inspection images 40 are generated. FIG. 16 is a diagram showing an example of the simulated inspection image 40 finally generated as described above. In the figure, white arrows point to images representing scratches or defects. In the figure, only the image 40a in the upper left corner is an image obtained by photographing an actual defect, and the others are synthesized images. From this figure, it can be seen that the roughness of the cast skin, the shape / density of defects, and the overall shading can be expressed.

1-3. Evaluation Using CNN Specifically, the identification performance of a CNN (Convolutional Neural Network) in which a simulated inspection image 40 generated by the method disclosed in the present embodiment was learned as learning data was confirmed. Five original images were prepared for each non-defective product and defective product, 1000 learning data were generated from these images, and two CNNs were trained. One CNN was trained using the learning data generated by the method of the present embodiment. The other CNN is for comparison, and learning was performed using learning data generated by affine transformation such as rotation, parallel movement, and enlargement / reduction of the original image. The configuration and parameters of the two CNNs are common. After learning, a good / bad test was performed using 2000 test images. The correct answer rate of CNN learned using data generated by the method of the present embodiment was 97.4%. On the other hand, the correct answer rate of CNN learned using data generated by affine transformation was 94.7%. The latter is a simple machine learning method, but it is considered that the performance deteriorated because the entire original image, that is, the base material is deformed. That is, it is considered that the method of the present embodiment can provide better quality learning data suitable for CNN learning.

1-4. Effects As described above, the image generation apparatus 10 according to the present embodiment adjusts the performance (evaluation, machine learning (deep learning), etc.) of an image inspection system that inspects scratches or defects on the surface of an inspection object using images. This is a device for generating a simulated inspection image 40 used in the above. The image generation apparatus 10 includes a raw base image 32 (an example of the base image of the present invention) obtained by photographing the surface of the inspection object, and a flaw / defect image 39 (an image of the defect image of the present invention) indicating an image of a flaw or a defect. Data storage unit 17 (an example of the storage unit of the present invention) and a controller 11 (an example of the image generation unit of the present invention) that generates a simulated inspection image. The controller 11 generates a base image 36 having a texture characteristic similar to that of the original image from the base original image 32 obtained by photographing the surface of the inspection object, and synthesizes a scratch / defect image 39 with the base image 36 to simulate the inspection image 40 ( An example of a simulated inspection image of the present invention is generated.

  According to the image generating apparatus 10 having the above-described configuration, a base image 36 having texture characteristics similar to that of the base original image 32 obtained by photographing the surface of the inspection object is generated and used as the background image of the simulated inspection image 40. For use, it is possible to generate an inspection image similar to the original texture characteristics of the surface of the inspection object. Therefore, an image inspection system with improved inspection accuracy can be provided by performing performance evaluation and learning of the image inspection system using such a simulated inspection image 40. Further, the number of simulation inspection images 40 can be generated by the number of combinations of the base image 36 and the scratch / defect image 39, and a large amount of image data for inspection can be generated. Therefore, it is possible to stably supply a large amount of accurate sample images necessary for performance evaluation of the image inspection system and machine learning (deep learning).

(Other embodiments)
The above embodiment shows one specific embodiment of the present invention, and the present invention is not limited to the contents of the above embodiment. Specific examples to which the present invention is applied are not limited to the contents of the above-described embodiment, and the above-described components can be appropriately changed, replaced, added, omitted, and the like.

  For example, in the above-described embodiment, the example in which various image data such as the base image 32, the base image 36, and the flaw defect image 39 used for the synthesis of the simulated inspection image 40 is stored in the data storage unit 17 has been described. The various image data 32 may be stored in the RAM 16. Further, the various image data used for the synthesis of the simulated inspection image 40 need not always be held in the image generation apparatus 10. The image generation device 10 may acquire various image data from an external storage device or the cloud as necessary.

  The examples of the shading pattern 34 and the scratch / defect representative pattern 35 shown in FIGS. 5 and 6 are also examples, and various other variations are possible.

  In the above embodiment, the base image synthesis process (FIG. 8), the flaw / defect image synthesis process (FIG. 11), and the simulated inspection image synthesis process (FIG. 15) are executed independently. The example to do was explained. These processes may be executed as a series of processes linked to each other.

  In the above embodiment, the base image 36, the flaw / defect image 39, and the shading pattern 34 are combined to generate the simulated inspection image 40. However, the shading pattern 34 is not necessarily required for combining. The simulated inspection image 40 may be generated by combining only the base image 36 and the scratch / defect image 39.

(This disclosure)
The above disclosure discloses the following technical idea.

(1) An apparatus (10) for generating a simulated inspection image (40) for use in adjusting the performance of an image inspection system for inspecting a scratch or a defect on the surface of an inspection object using an image,
A storage unit (16, 17) that stores an original image (32) obtained by imaging the surface of the inspection object and a defect image (39) that is an image showing the shape of a flaw or a defect;
An image generation unit (11) for generating a simulated inspection image,
The image generation unit generates a base image (36) having a texture characteristic similar to that of the original image from the original image (32) obtained by photographing the surface of the inspection object, and synthesizes the defect image (39) with the base image to perform a simulated inspection. Generating an image (40),
Image generation device.

(2) In the image generation device of (1), the image generation unit performs texture synthesis using an image (33) obtained by cutting out a partial region of the original image (32), thereby generating a base image (36). It may be generated.

(3) In the image generating apparatus of (1), the defect image (39) may be an image generated by adding a shadow set in consideration of the appearance of scratches or defects.

(4) In the image generation device according to any one of (1) to (3), the storage unit may store a plurality of defect images. The image generation unit may generate a plurality of simulated inspection images by combining each of the plurality of defect images with the base image.

(5) In the image generation device of (4), the storage unit may store a defect pattern image (35) that is an image showing a predetermined shape related to a scratch or a defect. The image generation unit may generate a plurality of defect images (39) by deforming the defect pattern image (35) by various methods (for example, affine transformation processing).

(6) In the image generation apparatus according to any one of (1) to (5), the image generation unit includes a base image (36), a defect image (39), and a shading pattern image (34) including only shadow information. ) May be combined to generate a simulated inspection image (40).

(7) In the image generation device according to any one of (1) to (6), the simulated inspection image is an image used for performance evaluation or learning of the image inspection system, for example.

The above embodiment discloses the following image generation method.
(8) A method for generating a simulated inspection image (40) used for adjusting the performance of an image inspection system for inspecting a scratch or a defect on the surface of an inspection object using an image by a computer,
A first step of storing, in the storage unit (16, 17), an original image (32) obtained by photographing the surface of the inspection object and a defect image (39) that is an image showing the shape of a flaw or a defect;
A second step of generating a base image (36) having a texture characteristic similar to that of the original image from the original image (32) by the control unit (11);
And a third step of generating a simulated inspection image (40) by synthesizing the defect image with the base image by the control unit (11).

(9) In the image generation method of (8), the control unit may generate a base image by cutting out a part of the original image and performing texture synthesis using the cut-out image (33). .

(10) In the image generation method of (8), the defect image (39) may be generated by adding a shadow set in consideration of the appearance of a flaw or a defect by the control unit.

(11) In the image generation method according to any one of (8) to (10), a plurality of defect images (39) may be stored in the storage unit. A plurality of simulated inspection images (40) may be generated by synthesizing each of the plurality of defect images (39) with the base image by the control unit.

(12) In the image generation method of (11), the control unit may generate a plurality of defect images by deforming the pattern image (35) indicating a predetermined shape related to scratches or defects by various methods.

(13) In the image generation method according to any one of (8) to (12), the control unit combines the base image, the defect image, and the shading pattern image (34) including only the shadow information, A simulated inspection image may be generated.

(14) In the image generation method according to any one of (8) to (13), the simulated inspection image is an image used for performance evaluation or learning of the image inspection system, for example.

(15) The embodiment described above discloses a program for causing a computer to execute any one of the image generation methods (8) to (14).

  The above-described embodiments are for illustrating the technology in the present disclosure, and various modifications, replacements, additions, omissions, and the like can be made within the scope of the claims or an equivalent scope thereof.

  The present invention can be applied to an apparatus that generates image data for use in evaluation and learning of an inspection apparatus that inspects scratches and defects on the surface of an object based on an image.

DESCRIPTION OF SYMBOLS 10 Image generation apparatus 11 Controller 13 Display part 15 Operation part 16 RAM
Reference Signs List 17 Data storage unit 18 First interface 19 Second interface 32 Base image 33 Base seed image 34 Shading pattern 35 Scratch / defect representative pattern 36 Base image 38 Scratch / defect silhouette image 39 Scratch / defect image 40 Simulation inspection image

Claims (15)

An apparatus for generating a simulated inspection image used for adjusting the performance of an image inspection system for inspecting a scratch or a defect on the surface of an inspection object using an image,
A storage unit that stores an original image obtained by photographing the surface of the inspection object, and a defect image that is an image showing a shape of a scratch or a defect;
An image generation unit for generating the simulated inspection image,
The image generation unit generates a base image having a texture characteristic similar to the original image from an original image obtained by photographing the inspection object surface, and generates the simulated inspection image by synthesizing the defect image with the base image. To
Image generation device.   The image generation apparatus according to claim 1, wherein the image generation unit generates the base image by performing texture synthesis using an image obtained by cutting out a partial region of the original image.   The image generation apparatus according to claim 1, wherein the defect image is an image generated by adding a shadow set in consideration of a flaw or a defect appearance. The storage unit stores a plurality of defect images,
The image generation device according to claim 1, wherein the image generation unit generates a plurality of simulated inspection images by combining each of a plurality of defect images with the base image. The storage unit stores a defect pattern image that is an image showing a predetermined shape related to a scratch or a defect,
The image generation unit generates a plurality of defect images by deforming the defect pattern image by various methods.
The image generation apparatus according to claim 4.   The image generation unit generates the simulated inspection image by combining the base image, the defect image, and a shading pattern image including only shadow information. The image generating apparatus described.   The image generating apparatus according to claim 1, wherein the simulated inspection image is an image used for performance evaluation or learning of the image inspection system. A method of generating a simulated inspection image for use in adjusting the performance of an image inspection system for inspecting a scratch or a defect on the surface of an inspection object using an image by a computer,
A first step of storing, in the storage unit, an original image obtained by photographing the surface of the inspection object and a defect image that is an image showing a shape of a scratch or a defect;
A second step of generating, from the original image, a base image having a texture characteristic similar to the original image by the control unit;
A third step of synthesizing the defect image with the base image and generating the simulated inspection image by the control unit;
Image generation method.   The image generation method according to claim 8, wherein the control unit generates the base image by cutting out a partial region of the original image and performing texture synthesis using the cut-out image.   The image generation method according to claim 8, wherein the control unit generates the defect image by adding a shadow set in consideration of a flaw or a defect appearance. Storing a plurality of defect images in the storage unit;
The image generation method according to claim 8, wherein the control unit generates a plurality of simulated inspection images by combining each of the plurality of defect images with the base image. The control unit generates a plurality of defect images by deforming a pattern image indicating a predetermined shape related to scratches or defects by various methods.
The image generation method according to claim 11. The image according to any one of claims 8 to 12, wherein the control unit generates the simulated inspection image by combining the base image, the defect image, and a shading pattern image including only shadow information. Generation method.   The image generation method according to claim 8, wherein the simulated inspection image is an image used for performance evaluation or learning of the image inspection system.   The program for making a computer perform the image generation method in any one of Claims 8-14.