JP2015099071A - Inspection device, inspection method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To more surely detect more defects of a painted surface.SOLUTION: A wavelet transform unit 103 subjects an object image being an image obtained by photographing a painted surface of an object, to wavelet transform with a plurality of mutually different mother wavelets. An inverse wavelet transform unit 105 subjects, for each mother wavelet, a high band-side component image obtained by wavelet transform with the mother wavelet, to inverse wavelet transform. A superposition unit 107 superposes reconstructed images for respective mother wavelets, which are reconstructed images obtained by inverse wavelet transform. An evaluation determination unit 84 evaluates the presence or absence of defects of the painted surface in accordance with a feature quantity of a superposed image obtained by superposing the reconstructed images.

Description

  The present invention relates to an inspection apparatus, an inspection method, and a program, and more particularly to an inspection apparatus, an inspection method, and a program that can inspect a painted surface.

  When a product made of metal or resin is coated, the inspection of the coating is often performed by human eyes. However, the visual inspection is likely to be overlooked. Conventionally, various methods for taking an image of a painted product and inspecting from the image have been proposed.

  Conventionally, the process of measuring multiple surface images by fixing the object to be evaluated and the camera, changing the position of the point light source, or sequentially turning on the point light sources arranged at multiple positions, using a metal surface reflection model The process of removing strong and strong specular reflections and extracting minute random reflections, removing noise and sharpening defects using complex discrete wavelet transform, and detecting defects by pattern matching. There are some which inspect surface defects (for example, see Patent Document 1).

JP 2008-298703 A

  However, the characteristics of the defective shape that occurs on the painted surface differ for each type of defect such as blisters, dust, and sagging. When inspecting a painted surface from an image of the painted surface, if it is attempted to reliably detect one type of defect, it is often impossible to detect another type of defect. That is, only specific types of defects on the painted surface could be detected. As a result, the number of defects that can be detected could not be increased.

  This invention is made | formed in view of such a condition, and enables it to detect more defects of a coating surface more reliably.

  An inspection apparatus according to one aspect of the present invention is an inspection apparatus that inspects whether or not a painted surface of a coated object is defective, and is an object image that is an image obtained by photographing the painted surface of the object. A wavelet transform means for performing wavelet transform by each of a plurality of mutually different mother wavelets, and a wavelet inverse transform means for performing wavelet inverse transform for each mother wavelet on a high frequency component image obtained by wavelet transform for each mother wavelet. A reconstructed image obtained by inverse wavelet transform, a superimposing unit that superimposes the reconstructed image for each mother wavelet, and a feature value of the superimposed image obtained by superimposing the reconstructed image, Evaluation means for evaluating the presence or absence of defects.

  The wavelet transform means can wavelet transform the object image with each of the mother wavelets, which are four different Dovecy wavelets.

  The wavelet transform means performs wavelet transform on the target image to a predetermined level, and the wavelet inverse transform means performs wavelet inverse transform for each high-frequency side component image for each level for each mother wavelet, and overlaps them. In the matching means, reconstructed images for each level and for each mother wavelet can be superimposed.

  The wavelet transform means can perform wavelet transform on the object image up to level 3.

  Separation means for separating an object image that is a color image into images of each of the three primary colors is further provided, and the wavelet transform means performs wavelet transform on each of the object images of each of the three primary colors by each of the mother wavelets, The wavelet inverse transform means performs wavelet inverse transform for each mother wavelet of the high-frequency component images of the three primary colors obtained by wavelet transform for each mother wavelet, and the superimposing means causes the colors of the three primary colors. Reconstructed images for each and every mother wavelet can be superimposed.

  A light source for irradiating the object with light of each of the three primary colors from different positions for each of the three primary colors can be further provided.

  A wavelet transform unit and a wavelet inverse transform unit can be realized by a GPU (Graphics Processing Unit) that executes a program.

  With a programmed FPGA (Field-Programmable Gate Array), wavelet transform means and wavelet inverse transform means can be realized.

  An inspection method according to one aspect of the present invention is an inspection method for inspecting whether or not a painted surface of a coated object is defective, and is an object image that is an image obtained by photographing the painted surface of the object. Is wavelet transformed by each of a plurality of mutually different mother wavelets, and the high-frequency component image obtained by wavelet transformation for each mother wavelet is subjected to wavelet inverse transformation for each mother wavelet and re-obtained by wavelet inverse transformation. The method includes the step of superimposing the reconstructed images for each mother wavelet and evaluating the presence or absence of a defective coating surface from the feature amount of the superimposed image obtained by superimposing the reconstructed images.

  A program according to one aspect of the present invention is an object that is an image obtained by photographing a painted surface of an object on a computer that controls an inspection device that inspects the presence or absence of a painted surface of the object being painted. The image was wavelet transformed by each of a plurality of different mother wavelets, and the high-frequency component image obtained by wavelet transformation for each mother wavelet was inversely transformed by wavelet for each mother wavelet, and obtained by inverse wavelet transformation. A process including a step of evaluating the presence or absence of a painted surface from a feature amount of a superimposed image obtained by superimposing the reconstructed image for each mother wavelet and overlaying the reconstructed image. Let it be done.

  In one aspect of the present invention, an object image that is an image obtained by photographing a painted surface of an object is wavelet transformed by each of a plurality of mutually different mother wavelets, and obtained by wavelet transformation for each mother wavelet. The obtained high-frequency component image is inversely wavelet transformed for each mother wavelet, and is a reconstructed image obtained by inverse wavelet transformation. The reconstructed image for each mother wavelet is overlaid, and the reconstructed image is overlaid. Whether or not the painted surface is defective is evaluated from the feature amount of the superimposed image obtained in this way.

  As described above, according to one aspect of the present invention, it is possible to detect a defect in the painted surface. Further, according to one aspect of the present invention, it is possible to more reliably detect more defects on the painted surface.

It is a figure which shows the structure of an inspection apparatus. It is a figure which shows the example of the cross section of the defect of the coating surface of articles | goods. It is a figure which shows the high region side component image and low region side component image which are obtained by discrete wavelet transform. It is a figure which shows the example of the waveform of a Dovecy wavelet. It is a figure which shows the example of the waveform of a Dovecy wavelet. It is a block diagram which shows the structural example of the hardware of a process part. It is a block diagram which shows the structure of the function implement | achieved by the process part which performs a program. It is a flowchart which shows the process of a test | inspection. It is a flowchart which shows the detail of pre-processing. It is a flowchart which shows the detail of the process of preparation of the image for evaluation determination. It is a flowchart which shows the detail of the process of preparation of the image for evaluation determination.

  Embodiments of the present invention will be described below. Correspondences between the configuration requirements of the present invention and the embodiments described in the detailed description of the present invention are exemplified as follows. This description is to confirm that the embodiments supporting the present invention are described in the detailed description of the invention. Accordingly, although there are embodiments that are described in the detailed description of the invention but are not described here as embodiments corresponding to the constituent elements of the present invention, It does not mean that the embodiment does not correspond to the configuration requirements. Conversely, even if an embodiment is described here as corresponding to a configuration requirement, that means that the embodiment does not correspond to a configuration requirement other than the configuration requirement. It's not something to do.

  An inspection apparatus according to one aspect of the present invention is an inspection apparatus (for example, the inspection apparatus 11 in FIG. 1) for inspecting whether or not a painted surface of a coated object is defective, and the painted surface of the object is photographed. Wavelet transform means (for example, the wavelet transform unit 103 in FIG. 7) that performs wavelet transform on the object image, which is an image obtained by each of a plurality of mutually different mother wavelets, and wavelet transform for each mother wavelet. A wavelet inverse transform unit (for example, the wavelet inverse transform unit 105 in FIG. 7) that performs wavelet inverse transform for each mother wavelet, and a reconstructed image obtained by the wavelet inverse transform, Superimposing means for superimposing the reconstructed images for each (for example, the overlapping in FIG. 7) Includes an Align 107), from the feature quantity of the resulting superimposed image by superimposing reconstructed images, and evaluation means for evaluating whether a defect of paintwork (e.g., evaluation determining unit 84 in FIG. 7).

  Separation means (for example, the separation unit 91 in FIG. 7) that separates the object image, which is a color image, into images of the three primary colors is further provided, and the wavelet transform means includes the object images of the three primary colors. The wavelet transform is performed by each of the mother wavelets, and the wavelet inverse transform means performs wavelet inverse transform of each of the high-frequency component images of the three primary colors obtained by the wavelet transform for each mother wavelet for each mother wavelet, The superimposing means can superimpose the reconstructed images for each of the three primary colors and for each mother wavelet.

  A light source (for example, the illumination unit 21-1 to illumination unit 21-3 in FIG. 1) that irradiates an object with light of each of the three primary colors from different positions for each of the three primary colors can be further provided.

  A wavelet transform unit and a wavelet inverse transform unit can be realized by a graphics processing unit (GPU) (for example, the GPU 62 in FIG. 6) that executes the program.

  An inspection method according to one aspect of the present invention is an inspection method for inspecting whether or not a painted surface of a coated object is defective, and is an object image that is an image obtained by photographing the painted surface of the object. Are wavelet transformed by a plurality of mutually different mother wavelets (for example, the procedure of step S62 in FIG. 10), and the high-frequency component image obtained by the wavelet transformation for each mother wavelet is wavelet-inverted for each mother wavelet. Transformed (for example, the procedure of step S64 in FIG. 10), and is a reconstructed image obtained by inverse wavelet transform, and the reconstructed image for each mother wavelet is overlaid (for example, the procedure of step S69 in FIG. 10). Evaluate the presence or absence of defects on the painted surface from the features of the superimposed image obtained by overlaying the reconstructed image. (For example, the procedure of step S14 in FIG. 8) includes the step.

  A program according to one aspect of the present invention is an object that is an image obtained by photographing a painted surface of an object on a computer that controls an inspection device that inspects the presence or absence of a painted surface of the object being painted. The image is wavelet transformed by each of a plurality of mutually different mother wavelets (for example, the procedure of step S62 in FIG. 10), and the high-frequency component image obtained by the wavelet transformation for each mother wavelet is converted into a wavelet for each mother wavelet. Inversely transform (for example, the procedure of step S64 in FIG. 10), and reconstructed images obtained by the inverse wavelet transform, and superimpose the reconstructed images for each mother wavelet (for example, the procedure of step S69 in FIG. 10). From the features of the superimposed image obtained by overlaying the reconstructed image, paint Assessing the presence or absence of a defect (e.g., procedures in step S14 of FIG. 8) to perform processing including steps.

  Hereinafter, an inspection apparatus according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a diagram illustrating a configuration of the inspection apparatus 11. The inspection device 11 inspects for the presence or absence of defects in the painted surface of the article 12 such as various products and products. The article 12 is an example of a painted object. For example, the article 12 can be a wiper arm. In addition, the coating method of the article 12 may be any coating method such as solvent coating, powder coating, or electrodeposition coating. The article 12 is placed at a predetermined position by a jig (not shown). Moreover, you may make it test | inspect the articles | goods 12 conveyed by a continuous conveyor or an intermittent feed positioning conveyor.

  The inspection apparatus 11 inspects the presence or absence of a defect on the painted surface of the article 12 by processing an image obtained by photographing the article 12. In the inspection apparatus 11, defects having different shape features are detected by using a plurality of mother wavelets.

  The inspection device 11 includes illumination units 21-1 to 21-3, an imaging unit 22, and a processing unit 23. The illumination unit 21-1 to the illumination unit 21-3 are examples of a light source, and include an incandescent lamp, a fluorescent lamp, a low-pressure discharge lamp, a light emitting diode, a cold cathode fluorescent tube, a laser (device), or the like. 12 is irradiated with light. In the illumination units 21-1 to 21-3, the optical axis of the light emitted from the illumination units 21-1 to 21-3 to the article 12 and the painted surface of the article 12 are at a predetermined angle (for example, , 30 degrees to 60 degrees).

  By doing in this way, the illumination according to the illumination part 21-1 thru | or the illumination part 21-3 becomes easy to produce the shadow according to the shape of the surface of the defect produced on the coating surface of the article | item 12. The shadow referred to here includes not only a portion where no light hits but also a change in brightness due to a change in the angle between the light and the surface. If the image obtained by photographing the article 12 includes a defective shadow, the difference in the image between the normal painted part and the defect can be more reliably extracted in the processing of the image. In addition, the inspection of the presence or absence of a defect on the painted surface of the article 12 becomes more reliable.

  Each of the illumination units 21-1 to 21-3 emits light of each color of three primary colors of additive mixing, that is, the so-called three primary colors by using a color filter or the like or emitting light of a specific wavelength. The article 12 is irradiated. For example, the illumination unit 21-1 irradiates the article 12 with red light, the illumination unit 21-2 irradiates the article 12 with green light, and the illumination unit 21-3 provides the article 12 with blue light. Irradiate.

  The illumination units 21-1 to 21-3 irradiate the article 12 with light of each of the three primary colors from different positions for each of the three primary colors. More specifically, for example, the illumination unit 21-1 to the illumination unit 21-3 are circles around the optical axis of imaging of the imaging unit 22, and are on the circumference of a circle orthogonal to the optical axis. Of the positions that equally divide the circumference, the angle between the optical axis of imaging of the imaging unit 22 and the line connecting each of the illumination units 21-1 to 21-3 and the article 12 is 30 degrees to 60 degrees It is arranged at any position. In this way, a defective image appearing on the shape of the painted surface of the article 12 can be changed for each of the three primary colors. In other words, the illuminating units 21-1 to 21-3 irradiate the article 12 with light having different frequencies (wavelengths) from different positions for light having different frequencies (wavelengths). Hereinafter, each of the three primary colors (light) can be rephrased as light having a frequency (wavelength) different from each other, and each of the three primary colors (each color) can be rephrased as light having a frequency different from each other.

  The imaging unit 22 is composed of, for example, a still camera or a video camera capable of color photography. The imaging unit 22 images the article 12 illuminated by the illumination units 21-1 to 21-3 and supplies the obtained image to the processing unit 23. That is, the imaging unit 22 supplies an image obtained by photographing the painted surface of the article 12 to the processing unit 23. An image taken by the imaging unit 22 and supplied to the processing unit 23 is an example of an object image.

  The image captured by the imaging unit 22 and supplied to the processing unit 23 may be a still image or a moving image. Hereinafter, a case where the object image is a still image will be described as an example. Although detailed description is omitted, when the object image is a moving image, it is possible to inspect whether there is a defect by performing a process similar to the process for a still image on a frame or a field.

  The processing unit 23 includes a personal computer, a workstation, a dedicated processing device, or the like, controls the imaging unit 22, acquires a target image of the article 12 supplied from the imaging unit 22, and acquires the target image. From this, the presence or absence of defects in the painted surface of the article 12 is evaluated and determined.

  That is, in more detail, the processing unit 23 stores in advance data for specifying a plurality of mother wavelets that are different from each other or data that indicates a mother wavelet, and obtains an object image by each of the plurality of mother wavelets that are different from each other. Wavelet transform. Then, the processing unit 23 inversely transforms the high-frequency component image of the images obtained by the wavelet transform for each mother wavelet for each mother wavelet. Further, the processing unit 23 superimposes the reconstructed images obtained by inverse wavelet transform and each mother wavelet. The processing unit 23 evaluates the presence / absence of a coating surface defect from the feature amount of the superimposed image obtained by superimposing the reconstructed images.

  FIG. 2 is a diagram illustrating an example of a defective cross section of the painted surface of the article 12. The defect of the coating surface detected by the inspection device 11 is a defect that appears in the shape of the surface of the coating film 32 formed by painting on the surface of the article 31 that is the article 12 before painting. In other words, the inspection apparatus 11 detects a defect in a painted surface having a shape different from a normal painted surface. As shown in FIG. 2, the feature of the defective shape that occurs on the painted surface differs for each type of defect. FIG. 2 shows an example of a defective cross section of a type called “butsu”, “trash”, “hajiki”, “scratch”, and “pinhole”.

  FIG. 2 (A) is a diagram showing an example of a cross section of a defective painted surface called “butsu”. The irregularity is generated when foreign matter such as iron powder, dust, or paint residue adheres to the coating film 32. Since the flowed coating material is pulled by the foreign matter due to the surface tension, the coating film 32 around the foreign material has a shape that rapidly rises in the coating film 32 where the flaws occur. There are various shapes and sizes of foreign matters such as iron powder, dust, and paint residue.

  FIG. 2B is a diagram showing an example of a cross section of a defective painted surface called dust. The dust is generated when fibers such as clothes adhere to the coating film 32. Since the fibers are placed on the coating film 32, the surface of the coating film 32 on which the fibers are placed has a fine uneven shape in the coating film 32 where dust is generated.

  FIG. 2C is a diagram showing an example of a cross section of a defective painted surface called a repellent. Repelling occurs when oil or the like adheres to the surface of the object 31 before painting. In the repelling, a hole reaching the surface (base surface) of the article 31 is formed. Holes caused by repellency spread over a wide range compared to trash and dust, and the inclination of the holes becomes gentle.

  FIG. 2D is a diagram showing an example of a cross section of a defective painted surface called a scratch. The scratch is a scratch or a scratch on the coating film 32 caused by a person or an object coming into contact with the coating film 32 after being applied and before being cured. The scratched coating film 32 has a relatively sharp and larger uneven shape than the coating film 32 with dust.

  FIG. 2E is a diagram showing an example of a cross section of a defective painted surface called a pinhole. The pinhole is a small hole such as a needle hole generated in the coating film 32, and refers to a hole reaching from the surface of the coating film 32 to the surface (substrate) of the article 31. Many holes are densely packed.

  As described above, the feature of the defective shape generated on the painted surface is different for each type of defect. The inspection apparatus 11 is capable of detecting more defects on the painted surface by more reliably detecting defects having different shape characteristics by using wavelet transforms of a plurality of different mother wavelets. It is.

・・・式（１） Here, an outline of the wavelet transform will be described. The wavelet transform is defined by convolution with the mother wavelet ψ (x) subjected to translation and expansion / contraction operations. The wavelet function ψ _{a, b} (x) uses the real number a (> 0) and real number b for the mother wavelet ψ (x) as parameters, and is multiplied by 1 / a ^1/2 in the vertical axis direction and a in the horizontal axis direction. , B are translated and represented by the formula (1).

... Formula (1)

  In equation (1), the real number a is referred to as a scale parameter, and the real number b is referred to as a shift parameter. The mother wavelet is also called a basic wavelet or a prototype wavelet.

  Wavelet transformation is roughly divided into two types, continuous wavelet transformation and discrete wavelet transformation. The continuous wavelet transform is one in which the scale parameters are continuous but the basis functions constituting the wavelet function are non-orthogonal, and the discrete wavelet transform is one in which the basis functions are orthogonal but the parameters are discrete. The continuous wavelet transform is applied to time-series data analysis using the feature of time / frequency analysis, and the discrete wavelet transform is applied to the field of image compression and encoding using the feature of multi-resolution analysis. Has been. The inspection apparatus 11 employs discrete wavelet transform.

The wavelet transform is a signal transform based on the wavelet function ψ _{a, b} (x) given by Equation (1). First, continuous wavelet transform will be described.

・・・式（２） When a signal to be analyzed is expressed by a function g (x), a wavelet transform of the signal g (x) is defined by Expression (2).

... Formula (2)

・・・式（３） From Equation (2), the inverse wavelet transform for restoring the signal g (x) is defined as Equation (3).

... Formula (3)

・・・式（４） In formula (3), C _ψ is represented by formula (4).

... Formula (4)

  Ψ (ω) with a circumflex symbol (hat symbol) indicates a Fourier transform of ψ (ω). The set R indicates a real number set.

Next, discrete wavelet transform will be described. In the discrete wavelet transform, using the integer i and the integer k, the real number a and the real number b in the formula (1) are set to values shown in the formula (5).
a = 2- ⁱ
b = 2- ⁱ k
... Formula (5)

・・・式（６） The discrete wavelet transform is expressed by Equation (6).

... Formula (6)

・・・式（７） Also, the discrete wavelet inverse transform is expressed by Equation (7).

... Formula (7)

・・・式（８） In Expression (7), focusing on the integer i and expressing the right side as h _i (x) as in Expression (8), Expression (9) can be expressed.

... Formula (8)

・・・式（９）

... Formula (9)

Expression (9) represents that the signal g _i (x) is subdivided into wavelet components h _i-1 (x), h _i-2 (x). Here, the integer i indicates a level. When the level i is lowered by 1 (the value of the level i is increased by 1), the resolution is halved. This level is the level of spatial resolution.

Assuming that the signal g _i (x) is an image S, as shown in FIG. 3, when a discrete wavelet transform is applied to the image S, a low-frequency component image L ₁ and a high-frequency component image H _{1 of} level 1 are obtained. can get. Further, when the discrete wavelet transform is applied to the level 1 low frequency component image L ₁ , the level 2 low frequency component image L ₂ and the high frequency component image H ₂ are obtained. Similarly, applying a discrete wavelet transform of level 2 to the low frequency band component images L _2, a low frequency band component images L ₃ and the high-frequency component image H ₃ was obtained Level 3, the low-frequency side of the Level 3 When the discrete wavelet transform is applied to the component image L ₃ , a level 4 low-frequency component image L ₄ and a high-frequency component image H ₄ are obtained.

  In this way, when the discrete wavelet transform of the same algorithm is applied to the image S or the low-frequency component image L at a predetermined level, the low-frequency component image L and the high-frequency component image H at the next lower level are applied. And is obtained.

High-frequency component image H ₁ level 1, compared to the high frequency band components of the component images of H ₂ level 2, including higher frequency components. Similarly, the high-frequency component image of H ₂ level 2, as compared to the components in the high frequency band component images H ₃ Level 3 includes higher frequency components, the high-frequency component image H ₃ Level 3 , compared to components in the high frequency band component images H ₄ level 4, including higher frequency components.

  Hereinafter, the high-frequency component image is also referred to as a detailed image.

As described above, in the discrete wavelet transform, for example, when analyzing the image S as the signal g (x), the reference mother wavelet is extended to various scales, and the wavelet function becomes a measure of the change in the pixel value of the pixel. A large number of ψ _{a, b} (x) is prepared, and information on changes in pixel values of pixels applied to the image S while moving in the x-axis direction (horizontal direction), y-axis direction (vertical direction) or diagonal direction of the image S Will be obtained. The wavelet transform is equivalent to obtaining a correlation value between the signal g (x) and the wavelet function ψ _{a, b} (x). In the wavelet transform, in essence, the wavelet that is a ruler in the signal g (x). It is determined how many components similar to the function ψ _{a, b} (x) are included.

  As described above, in the inspection apparatus 11, an image obtained by photographing the painted surface of the article 12 is wavelet transformed by each of a plurality of different mother wavelets. Here, an example of a mother wavelet employed for wavelet transform by the inspection apparatus 11 will be described. For example, in the inspection apparatus 11, wavelets included in the wavelet family proposed by Ingrid Daubechies, so-called Dovecy wavelets, can be adopted as mother wavelets. Note that the dovecy wavelet is specified by an integer N, and the dovecy wavelet with N = 1 is equivalent to the wavelet proposed by Haar.

  FIG. 4 and FIG. 5 are diagrams showing examples of the waveform of the Dovecy wavelet. FIG. 4A is a diagram illustrating an example of a waveform of a dovecy wavelet with N = 1. A dovecy wavelet with N = 1 has a stepped waveform.

  FIG. 4B is a diagram illustrating an example of a waveform of a Dovecy wavelet with N = 2. The dovecy wavelet with N = 2 is curved toward a sharp tip, and has a pair of saw-like waveforms on the top and bottom where fine waves are superimposed.

  FIG. 4C is a diagram illustrating an example of the waveform of the Dovecy wavelet with N = 3. The N = 3 Dovecy wavelet has a waveform in which a triangular wave having a sharp tip is formed which extends greatly upward and extends downward on both sides of the triangular wave having a sharp tip.

  FIG. 5A, FIG. 5B, and FIG. 5C are diagrams showing examples of the waveform of each Dovecy wavelet with N = 4 to 6, respectively. The N = 4 Dovecy wavelet has a waveform that extends upward and has a triangular wave with a relatively rounded tip that swings up and down on both sides of the triangular wave with a relatively rounded tip. Become. N = 5 Dovecy wavelet has a triangular wave with a relatively rounded tip, which swings up and down on both sides of a triangular wave with a relatively rounded tip. Compared to the case, the difference between the amplitude of the central wave and the amplitude of the waves on both sides is smaller. The N = 6 dovecy wavelet has a wider wave width (lateral length in FIG. 5C) than the N = 5 dovecy wavelet.

When the image obtained by photographing the painted surface of the article 12 is wavelet transformed by each of a plurality of different mother wavelets, a cross section of a defective painted surface is converted into an image obtained by wavelet transformation using one of the mother wavelets. The feature of appears. As described above, in the wavelet transform, how much a component similar to the wavelet function ψ _{a, b} (x) is included in the image is measured, and the result appears in the high frequency component image.

  For example, since the pinhole is a small hole such as a needle hole generated in the coating film 32, when an image obtained by photographing the painted surface where the pinhole is generated is wavelet transformed by each of a plurality of mother wavelets different from each other, A feature appears in the high-frequency component image obtained by wavelet transform using the dovecy wavelet with N = 1. For example, dust generated due to the attachment of fibers such as clothes causes a fine uneven shape on the surface of the coating film 32. Therefore, a plurality of images obtained by photographing a painted surface to which fibers such as clothes are attached are different from each other. When the wavelet transform is performed by each of the mother wavelets, a feature appears in the high frequency component image obtained by the wavelet transform by the N = 2 Dovecy wavelet.

  In addition, since the coating film 32 with the blisters suddenly rises, the wavelet transform is performed by each of a plurality of mother wavelets obtained by photographing the painted surface to which foreign matter such as iron powder, dust, or paint residue is adhered. Then, a feature appears in the high-frequency component image obtained by wavelet transform using the N = 4 Dovecy wavelet. Furthermore, since the inclination of the hole caused by the repellency becomes gentle, when an image obtained by photographing the painted surface where the repellency occurs is subjected to wavelet transformation by each of a plurality of different mother wavelets, N = 6 dovecy Features appear in the high-frequency component image obtained by wavelet transform using the wavelet. Similarly, the shape of the scratched coating film 32 is relatively sharp and has larger irregularities, so that an image of the scratched painted surface is wavelet transformed by each of a plurality of different mother wavelets. Then, a feature appears in the high frequency component image obtained by wavelet transform using the N = 3 Dovecy wavelet.

  In this way, in the inspection apparatus 11, the image obtained by photographing the painted surface is wavelet transformed by each of a plurality of mother wavelets whose waveforms are different from each other according to the shape of the defective painting, so that the irregularity, dust, repelling, scratches, etc. Even if the feature of the shape of a defect that occurs on the painted surface, such as a pinhole, differs for each type of defect, the feature appears in the high-frequency component image obtained by wavelet transform using any of the wavelets. In the inspection apparatus 11, the presence or absence of a defective painted surface of the article 12 is inspected from the features that appear in the high-frequency component image obtained by wavelet transform using any one of the wavelets.

  If a wavelet transform is performed with one mother wavelet and the resulting image is inspected for defects on the painted surface, detect any one defect such as blisters, dust, repellent, scratches, or pinholes. Although it becomes possible, for example, it becomes difficult to detect a plurality of types of defects having different shape features such as repelling and pinholes.

  Although it is possible to combine wavelet transform and other transforms such as Fourier transform, it is necessary to use different algorithms, and it becomes necessary to combine results obtained by different transforms. Adjustment of settings becomes difficult.

  If the wavelet transform is adopted, for example, many different mother wavelets such as Dovecy wavelet have been proposed, and can be selected according to the characteristics of the defective shape generated on the painted surface.

  FIG. 6 is a block diagram illustrating a hardware configuration example of the processing unit 23 that executes a series of inspection processes to be described later using a program.

  In the processing unit 23 which is a computer, a CPU (Central Processing Unit) 51, a ROM (Read Only Memory) 52, and a RAM (Random Access Memory) 53 are connected to each other by a bus 54.

  An input / output interface 55 is further connected to the bus 54. The input / output interface 55 includes an input unit 56 composed of a keyboard, mouse, microphone, etc., an output unit 57 composed of a display, a speaker, etc., a storage unit 58 composed of a hard disk and a non-volatile memory, and a communication unit 59 composed of a network interface. A drive 60 for driving a removable medium 61 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory is connected. Further, the input / output interface 55 is connected to the imaging unit 22, the imaging unit 22 is controlled via the input / output interface 55, and an object image is acquired from the imaging unit 22.

  In the computer configured as described above, the CPU 51 loads the program stored in the storage unit 58 to the RAM 53 through the input / output interface 55 and the bus 54 and executes the program, for example. Is performed.

  The program executed by the computer (CPU 51) is, for example, a magnetic disk (including a flexible disk), an optical disk (CD-ROM (Compact Disc-Read Only Memory), DVD (Digital Versatile Disc), etc.), a magneto-optical disk, or a semiconductor. It is recorded on a removable medium 61, which is a package medium composed of a memory or the like, or is provided via a wired or wireless transmission medium such as a local area network, the Internet, or digital satellite broadcasting.

  The program can be installed in the computer by installing the removable medium 61 in the drive 60 and storing it in the storage unit 58 via the input / output interface 55. Further, the program can be installed in a computer by being received by the communication unit 59 via a wired or wireless transmission medium and stored in the storage unit 58. In addition, the program can be installed in the computer in advance by storing it in the ROM 52 or the storage unit 58 in advance.

  Further, by connecting a graphics processing unit (GPU) 62 for performing image processing operations to the input / output interface 55 and operating the GPU 62 as an arithmetic unit for performing general-purpose calculations, a series of inspection processes described later is performed. It can also be executed by a program. Since the GPU 62 has a configuration suitable for numerical calculation, the GPU 62 can execute processing at higher speed.

  FIG. 7 is a block diagram illustrating a configuration of functions realized by the processing unit 23 that executes a program. When the processing unit 23 executes the program, an original image acquisition unit 81, a preprocessing unit 82, an evaluation determination image creation unit 83, an evaluation determination unit 84, and a result output unit 85 are realized.

  The original image acquisition unit 81 controls the input / output interface 55 and acquires an original image that is a target image from the imaging unit 22 connected to the input / output interface 55. The preprocessing unit 82 performs preprocessing necessary to create an image for evaluation determination. The evaluation determination image creation unit 83 creates an image for evaluation determination from the image and data obtained as a result of the preprocessing. The evaluation determination unit 84 evaluates and determines the presence or absence of a defective coating surface of the article 12 from the image for evaluation determination. The result output unit 85 outputs the result of evaluation and determination of the presence / absence of a defect in the painted surface of the article 12.

  The preprocessing unit 82 includes a separation unit 91, a luminance adjustment unit 92, a gray scale conversion unit 93, a mask generation unit 94, a coordinate acquisition unit 95, a trimming unit 96, and a masking unit 97. The separation unit 91 separates the original image that is the object image into single color images of the three primary colors of light. The brightness adjustment unit 92 adjusts the brightness of the image. The gray scale unit 93 converts the original image, which is a color image, to gray scale, and generates a gray scale image, which is a gray scale image. The mask generation unit 94 generates a mask for deleting unnecessary portions of the image from the grayscale image or the like.

  The coordinate acquisition unit 95 acquires coordinates indicating the position of the image of the article 12 among the coordinates on the image from the object mask. The trimming unit 96 refers to the acquired coordinates and cuts out the image of the article 12 from the single color image or the gray scale image. The masking unit 97 extracts an image of the article 12 from which unnecessary regions are removed by applying a mask to the original image that is the object image.

  The evaluation determination image creating unit 83 includes a mother wavelet storage unit 101, a weight coefficient storage unit 102, a wavelet transform unit 103, a selection unit 104, a wavelet inverse transform unit 105, a masking unit 106, a superposition unit 107, and a weighting unit 108. Composed. The mother wavelet storage unit 101 is formed in the storage area of the ROM 52 or RAM 53 and stores data indicating a plurality of different mother wavelets. For example, the mother wavelet storage unit 101 stores in advance data for specifying a plurality of mother wavelets that are predetermined N-valued dovecy wavelets or data indicating the mother wavelet. Specifically, the mother wavelet storage unit 101 may store only the N value of the Dovecy wavelet or may store data indicating a waveform. The mother wavelet storage unit 101 may store filter coefficients corresponding to a plurality of mother wavelets, which are predetermined N-valued Dovecy wavelets.

  The weighting coefficient storage unit 102 is formed in a storage area of the ROM 52 or the RAM 53, and stores a weighting coefficient that is a coefficient for weighting an image. For example, the weight coefficient storage unit 102 stores a weight coefficient for each mother wavelet and each level. The wavelet transform unit 103 performs wavelet transform on an original image, which is an object image that has been subjected to predetermined preprocessing, by each of a plurality of mutually different mother wavelets indicated by data stored in the mother wavelet storage unit 101. To do. The selection unit 104 selects a detailed image at each level from images obtained by wavelet transform. The detailed image is an example of a high frequency component image.

  The wavelet inverse transform unit 105 performs wavelet inverse transform on each of the selected detail images at each level to generate a reconstructed image. The masking unit 106 removes unnecessary areas from the reconstructed image by applying a mask to the reconstructed image, and extracts an image portion consisting only of an image corresponding to the image of the article 12 from each portion of the reconstructed image. To do. The superimposing unit 107 superimposes the reconstructed images. The weighting unit 108 weights the reconstructed image with the weighting coefficient stored in the weighting coefficient storage unit 102.

  The evaluation determination image creation unit 83 supplies the evaluation determination unit 84 with the total image generated by superimposing the weighted reconstructed images.

  The evaluation determination unit 84 includes a binarization unit 111, an image adjustment unit 112, a feature amount extraction unit 113, a determination unit 114, and a setting unit 115. The binarization unit 111 binarizes the sum image supplied from the evaluation determination image creation unit 83 and generates a binarized sum image. The image adjustment unit 112 adjusts the binarized sum image so as to facilitate particle analysis. The feature amount extraction unit 113 calculates the feature amount of each particle included in the binarized total image. The determination unit 114 compares a predetermined threshold value with the feature amount of each particle included in the binarized total image, and determines whether the feature amount is larger than the threshold value. The setting unit 115 sets, to the result of the evaluation determination, that a failure has been detected or that a failure has not been detected.

  Next, the inspection process will be described with reference to the flowchart of FIG. In step S <b> 11, the original image acquisition unit 81 instructs the imaging unit 22 to image the object to be inspected, that is, the article 12 by instructing the imaging unit 22 to perform imaging via the input / output interface 55. Then, the original image acquisition unit 81 acquires an original image that is a color image obtained by photographing the article 12 via the input / output interface 55. The acquired original image is an example of the object image.

  In step S <b> 12, the preprocessing unit 82 performs preprocessing necessary for creating a sum image that is an image for evaluation determination. Details of the preprocessing will be described with reference to the flowchart of FIG.

  In step S <b> 13, the evaluation determination image creation unit 83 executes a process of creating an evaluation determination image that creates a sum image from the image and data obtained as a result of the preprocessing. Details of the process of creating the evaluation determination image will be described with reference to the flowchart of FIG.

  In step S <b> 14, the evaluation determination unit 84 evaluates the presence / absence of a coating surface defect of the article 12 from the summed image, and details of the evaluation determination process for executing the evaluation determination process is illustrated in the flowchart of FIG. 11. To explain.

  In step S <b> 15, the result output unit 85 outputs the result of the evaluation determination indicating whether the painted surface of the article 12 is defective or not, and the inspection process ends. For example, in step S <b> 15, the result output unit 85 pays attention to an image (such as an image indicating the occurrence of a defect) or a sound (such as an image indicating the occurrence of a defect) corresponding to the result of the evaluation determination indicating whether the painted surface of the article 12 is defective. A file in which data indicating the result of evaluation determination indicating whether there is a defect in the painted surface of the article 12 is stored in the storage unit 58. To write to.

  FIG. 9 is a flowchart showing details of the preprocessing. In step S31, the separation unit 91 separates the original image, which is the object image, into single color images of the three primary colors of light. That is, for example, when each pixel of the original image includes a red subpixel, a green subpixel, and a blue subpixel, the separation unit 91 extracts the red subpixel of each pixel of the original image. The extracted sub-pixel is used as a pixel to generate a red single-color image, the green sub-pixel of each pixel of the original image is extracted, and the extracted sub-pixel is used as a pixel to produce a green single-color image. , The blue subpixel of each pixel of the original image is extracted, and the extracted subpixel is used as a pixel to generate a blue monochromatic image. In this manner, the separation unit 91 separates the original image into red, green, and blue single color images. Since the illumination unit 21-1 to the illumination unit 21-3 irradiate the article 12 with light of each of the three primary colors from different positions for each of the three primary colors, the single color image of each of the three primary colors of light is one Different shapes will be shown for each defect.

  In step S32, the brightness adjusting unit 92 adjusts the brightness of each monochrome image obtained in the procedure of step S31. In step S <b> 33, the grayscale unit 93 converts the original image that is a color image to grayscale, and generates a grayscale image that is a grayscale image. For example, the gray scale conversion unit 93 calculates a value indicating the luminance of each pixel from the pixel value of each pixel of the original image, which is a color image, and sets the luminance value to each pixel, so that white to black A gray scale image expressed only by the brightness and darkness is generated.

  In step S <b> 34, the mask generation unit 94 extracts only the part of the image showing the article 12 that is the object from the generated grayscale image, and deletes the part of the image other than the image showing the article 12. A mask for an object which is a mask for generating the object is generated. In step S <b> 35, the coordinate acquisition unit 95 acquires circumscribed rectangular coordinates of the image of the article 12 that is the object from the object mask. When the horizontal direction of the image is represented by the X axis, the vertical direction of the image is represented by the Y axis, and the upper left corner of the image is the origin, the circumscribed rectangular coordinates are the position of the right end of the image of the article 12 and the position of the article 12 on the X axis. The position of the left end of the image and the position of the upper end of the image of the article 12 and the position of the lower end of the image of the article 12 on the Y axis are shown.

  In step S36, the trimming unit 96 refers to the circumscribed rectangle coordinates acquired in step S35, and trims each single-color image of each of the three primary colors of light with the circumscribed rectangle defined by the circumscribed rectangle coordinates. That is, the trimming unit 96 removes an image in a range outside the circumscribed rectangle from each of the single color images of the three primary colors of light. In step S37, the trimming unit 96 refers to the circumscribed rectangle coordinates acquired in step S35, and trims the original image with the circumscribed rectangle defined by the circumscribed rectangle coordinates. That is, the trimming unit 96 removes an image in a range outside the circumscribed rectangle from the original image.

  In step S38, the mask generator 94 generates a halation mask from the trimmed original image. More specifically, the mask generation unit 94 detects an area having a luminance equal to or higher than the threshold by comparing the pixel value of each pixel of the trimmed original image with a predetermined threshold, and an area in which halation occurs A halation mask is generated as a mask for removing (regions having a predetermined luminance or higher).

  In step S39, the trimming unit 96 refers to the circumscribed rectangle coordinates acquired in step S35, and trims the object mask with the circumscribed rectangle defined by the circumscribed rectangle coordinates. That is, the trimming unit 96 removes an image in a range outside the circumscribed rectangle from the object mask.

  In step S40, the mask generation unit 94 generates a monochrome image mask image that is a mask (image) to be applied to each monochrome image from the trimmed object mask and halation portion mask. For example, the mask generation unit 94 superimposes the trimmed object mask and the halation mask, and can remove a region removed by either the trimmed object mask or the halation mask. An image mask image is generated.

  In step S41, the masking unit 97 applies the single-color image mask image to each single-color image and masks each single-color image. As a result, an image of the article 12 from which unnecessary areas are removed is extracted from each of the monochrome images. That is, the masking unit 97 generates an image by applying a single-color image mask image to each single-color image and removing a region where halation has occurred from the image portion showing the article 12 as the object.

  In step S42, the mask generation unit 94 generates a wavelet transform image mask image, which is a mask image for each wavelet transform level and each color, from the single color image mask image. As described above, the wavelet transform level indicates the number of times wavelet transform for one image is repeated. When wavelet transform is applied to an image, the transformed image is reduced. The wavelet transform image mask image is generated for each color, and the size of the wavelet transform image mask image is adjusted according to the level of the wavelet transform.

  In step S <b> 43, the preprocessing unit 82 outputs an inspection region extraction image that is a masked single color image to the evaluation determination image creation unit 83. In step S44, the preprocessing unit 82 outputs the wavelet transform image mask image to the evaluation determination image creating unit 83, and the preprocessing ends.

  In this way, in the pre-processing, an inspection area extraction image obtained by removing an area where halation has occurred from the image portion showing the article 12 that is the object, and a wavelet transform. A wavelet transform image mask image that matches the level of the image is generated and supplied to the evaluation determination image creating unit 83.

Next, the details of the process of creating the evaluation determination image will be described with reference to the flowchart of FIG. In step S <b> 61, the wavelet transform unit 103 reads data indicating a plurality of different mother wavelets from the mother wavelet storage unit 101. In step S62, the wavelet transform unit 103 performs wavelet transform of the inspection region extraction image to a predetermined level for each of a plurality of different mother wavelets and colors. That is, the wavelet transform unit 103 performs wavelet transform of the inspection region extraction image to a predetermined level for each of a plurality of mother wavelets and colors having different waveforms according to the shape of the defective coating. More specifically, the wavelet transform unit 103 applies each of the wavelet functions ψ _{a, b} (x) for each of _a plurality of different mother wavelets ψ (x) to the inspection area extraction image for each color up to a predetermined level. The discrete wavelet transform represented by the equation (6) using is applied.

  For example, in step S62, the wavelet transform unit 103 is a mother wavelet having a waveform corresponding to the shape of each of “butsu”, “trash”, “hajiki”, and “scratches”, N = 2, N = 3, N = 4, and N = 6. The inspection region extraction image is wavelet transformed by each of the four different Dovecy wavelets. By doing in this way, among the defects of the painted surface of the article 12, it is possible to more reliably detect defects, dust, cissing and scratches that are relatively frequently occurring.

  Also, for example, in step S62, the wavelet transform unit 103 performs wavelet transform on the inspection region extraction image up to level 3. In this way, even if a small defect can be detected, a large defect cannot be detected, or even if a large defect can be detected, the detection of the small defect cannot be detected. Therefore, the defect can be detected more reliably regardless of the size of the defect (spatial size).

This will be described in more detail by taking as an example a case where wavelet transform is performed up to level 3 by using respective dovecy wavelets of N = 2, N = 3, N = 4, and N = 6. The wavelet transform unit 103 performs wavelet transform on the red inspection region extracted image by N = 2 Dovecy wavelet, and wavelet transforms the obtained level 1 low-frequency component image L ₁ by N = 2 Dovecy wavelet. , wavelet transform the low frequency band component images L ₂ obtained level 2 by Dobe Sea wavelets N = 2. The wavelet transform unit 103 performs wavelet transform on the red inspection region extracted image by N = 3 Dovecy wavelet, and wavelet transforms the obtained level 1 low-frequency component image L ₁ by N = 3 Dovecy wavelet. , wavelet transform the low frequency band component images L ₂ obtained level 2 by Dobe Sea wavelets N = 3. The wavelet transform unit 103 performs wavelet transform on the red inspection region extracted image by N = 4 Dovecy wavelet, and wavelet transforms the obtained level 1 low-frequency component image L ₁ by N = 4 Dovecy wavelet. , wavelet transform the low frequency band component images L ₂ obtained level 2 by Dobe Sea wavelets N = 4. The wavelet transform unit 103 performs wavelet transform on the red inspection region extracted image by N = 6 Dovecy wavelet, and wavelet transforms the obtained level 1 low-frequency component image L ₁ by N = 6 Dovecy wavelet. , wavelet transform the low frequency band component images L ₂ obtained level 2 by Dobe Sea wavelets N = 6.

Further, the wavelet transform unit 103 performs wavelet transform on the green inspection region extracted image by N = 2 Dovecy wavelet, and wavelet transforms the obtained level 1 low-frequency component image L ₁ by N = 2 Dovecy wavelet. conversion, resulting in the level 2 low-frequency component image L ₂ to the wavelet transform by Dobe Sea wavelets N = 2. The wavelet transform unit 103 performs wavelet transform on the green inspection region extracted image by N = 3 Dovecy wavelet, and wavelet transforms the obtained level 1 low-frequency component image L ₁ by N = 3 Dovecy wavelet. , wavelet transform the low frequency band component images L ₂ obtained level 2 by Dobe Sea wavelets N = 3. The wavelet transform unit 103 performs wavelet transform on the green inspection region extracted image by N = 4 Dovecy wavelet, and wavelet transforms the obtained level 1 low-frequency component image L ₁ by N = 4 Dovecy wavelet. , wavelet transform the low frequency band component images L ₂ obtained level 2 by Dobe Sea wavelets N = 4. The wavelet transform unit 103, the green inspection region extraction image wavelet transform by Dobe Sea wavelets N = 6, the low frequency band component images L ₁ of the obtained level 1 wavelet transform by Dobe Sea wavelets N = 6 , wavelet transform the low frequency band component images L ₂ obtained level 2 by Dobe Sea wavelets N = 6.

Further, the wavelet transform unit 103 performs wavelet transform on the blue inspection region extracted image by N = 2 Dovecy wavelet, and wavelet transforms the obtained level 1 low-frequency component image L ₁ by N = 2 Dovecy wavelet. conversion, resulting in the level 2 low-frequency component image L ₂ to the wavelet transform by Dobe Sea wavelets N = 2. The wavelet transform unit 103 performs wavelet transform on the blue inspection region extracted image by N = 3 Dovecy wavelet, and wavelet transforms the obtained level 1 low-frequency component image L ₁ by N = 3 Dovecy wavelet. , wavelet transform the low frequency band component images L ₂ obtained level 2 by Dobe Sea wavelets N = 3. The wavelet transform unit 103 performs wavelet transform on the blue inspection region extracted image by N = 4 Dovecy wavelet, and wavelet transforms the obtained level 1 low-frequency component image L ₁ by N = 4 Dovecy wavelet. , wavelet transform the low frequency band component images L ₂ obtained level 2 by Dobe Sea wavelets N = 4. The wavelet transform unit 103 performs wavelet transform on the blue inspection region extracted image by N = 6 Dovecy wavelet, and wavelet transforms the obtained level 1 low-frequency component image L ₁ by N = 6 Dovecy wavelet. , wavelet transform the low frequency band component images L ₂ obtained level 2 by Dobe Sea wavelets N = 6.

In this way, the level 1 low-frequency component image L ₁ and the high-frequency component image H ₁ obtained by wavelet transforming the red inspection region extraction image with the N = 2 Dovecy wavelet, the level 2 low frequency A side component image L ₂ and a high frequency component image H ₂ , and a level 3 low frequency component image L ₃ and a high frequency component image H ₃ are obtained. Level 1 low-frequency component image L ₁ and high-frequency component image H ₁ , level 2 low-frequency component image L ₂ and high-frequency component image H ₂ , and level 3 obtained by wavelet transform using wavelets A low-frequency component image L ₃ and a high-frequency component image H ₃ are obtained, and the red inspection region extraction image is wavelet transformed by N = 4 Dovecy wavelet to obtain a level 1 low-frequency component image L ₁ Hoyo High-frequency component image H _1, the low-frequency component image L ₂ and high-side component image of H ₂ level _2, and the low-frequency component image L ₃ and high-side component image H ₃ Level 3 is obtained, Level 1 low-frequency component image L _1, high-frequency component image H ₁ , and level 2 low-frequency component image L obtained by wavelet transforming the red inspection region extracted image with N = 6 Dovecy wavelet ₂ and a high-frequency component image H ₂ , and a low-frequency component image L ₃ and a high-frequency component image H _{3 of level 3} are obtained.

Further, the level 1 low-frequency component image L ₁ and the high-frequency component image H ₁ , and the level 2 low-frequency component obtained by wavelet transforming the green inspection region extracted image with the N = 2 Dovecy wavelet An image L ₂ and a high-frequency component image H ₂ , and a low-frequency component image L ₃ and a high-frequency component image H _{3 of level 3} are obtained, and a green inspection region extraction image is obtained by N = 3 Dovecy wavelet. Level 1 low-frequency component image L ₁ and high-frequency component image H ₁ obtained by wavelet transform, level 2 low-frequency component image L ₂ and high-frequency component image H ₂ , and level 3 low A band-side component image L ₃ and a high-band side component image H ₃ are obtained, and a level 1 low-band component image L ₁ obtained by wavelet transforming a green inspection region extraction image with an N = 4 Dovecy wavelet. And high range Component images H _1, the low-frequency component image L ₂ and high-side component image of H ₂ level _2, and the low-frequency component image L ₃ and high-side component image H ₃ Level 3 is obtained, a green test Level 1 low-frequency component image L ₁ and high-frequency component image H ₁ , level 2 low-frequency component image L ₂ and high obtained by wavelet transforming the region-extracted image with N = 6 Dovecy wavelet A band-side component image H ₂ , a level 3 low-frequency component image L _3, and a high-frequency component image H ₃ are obtained.

Further, the level 1 low-frequency component image L ₁ and the high-frequency component image H ₁ , and the level 2 low-frequency component obtained by wavelet transforming the blue inspection region extracted image with the N = 2 Dovecy wavelet An image L ₂ and a high-frequency component image H ₂ , and a low-frequency component image L ₃ and a high-frequency component image H _{3 of level 3} are obtained, and a blue inspection region extraction image is obtained by N = 3 Dovecy wavelet. Level 1 low-frequency component image L ₁ and high-frequency component image H ₁ obtained by wavelet transform, level 2 low-frequency component image L ₂ and high-frequency component image H ₂ , and level 3 low A band-side component image L ₃ and a high-band side component image H ₃ are obtained, and a level 1 low-band side component image L ₁ obtained by wavelet transforming a blue inspection region extraction image with an N = 4 Dovecy wavelet. And high Side component images H _1, the low-frequency component image L ₂ and high-side component image of H ₂ level _2, and the low-frequency component image L ₃ and high-side component image H ₃ Level 3 is obtained, blue Level 1 low-frequency component image L ₁ and high-frequency component image H ₁ , level 2 low-frequency component image L _2, obtained by wavelet transforming the inspection region extraction image with N = 6 Dovecy wavelets, and A high-frequency component image H ₂ and a low-frequency component image L ₃ and a high-frequency component image H _{3 at level 3} are obtained.

In step S63, the selection unit 104 selects a detailed image at each level, which is an example of a high-frequency component image at each level, from the wavelet transform images for each mother wavelet and color obtained by the wavelet transform. . For example, as shown in FIG. 3, if the inspection region extraction image is wavelet transform up to the level 4, the selection unit 104, the mother wavelet and per each color, the high-frequency component image H ₁ Level 1 and Level 2 High-frequency side component image H ₂ , level 3 high-frequency side component image H _3, and level 4 high-frequency side component image H ₄ are selected.

When wavelet transform is performed up to level 3 using each of the dovecy wavelets of N = 2, N = 3, N = 4, and N = 6, the selection unit 104 converts the red inspection region extraction image to N = 2, Level 1 high-frequency component image H ₁ , level 2 high-frequency component image H ₂ , and level obtained by wavelet transform using the respective Dovecy wavelets with N = 3, N = 4, and N = 6 3 high frequency side component images H ₃ were selected, and the green inspection region extracted image was obtained by wavelet transform using N = 2, N = 3, N = 4, and N = 6 Dovecy wavelets. A level 1 high frequency component image H ₁ , a level 2 high frequency component image H ₂ , and a level 3 high frequency component image H ₃ are selected, and a blue inspection region extraction image is N = 2, N = 3, N = 4, and N = 6 The level 1 high-frequency component image H ₁ , the level 2 high-frequency component image H ₂ , and the level 3 high-frequency component image H ₃ obtained by wavelet transform using the respective Dovecy wavelets are selected. .

In step S64, the wavelet inverse transformation unit 105 performs wavelet inverse transformation on each of the selected detailed images at each level for each mother wavelet and each color, thereby generating a reconstructed image. In other words, the wavelet inverse transform unit 105 uses the wavelet functions ψ _{a, b} (x) for each of _a plurality of different mother wavelets ψ (x) for the selected detailed images at each level for each color. The discrete wavelet inverse transform expressed by (7) is applied. In this case, the detailed image obtained by the discrete wavelet transformation using the wavelet function ψ _{a, b} (x) of the predetermined mother wavelet ψ (x) is used in the discrete wavelet transformation for obtaining the detailed image. The discrete wavelet inverse transform using the wavelet function ψ _{a, b} (x) of the mother wavelet ψ (x) is applied. For example, as shown in FIG. 3, when the inspection region extraction image is wavelet transformed up to level 4, the wavelet inverse transformation unit 105 performs the level 1 high-frequency component image H ₁ , for each mother wavelet and color. Each of the high-frequency component image H ₂ at level ₂ , the high-frequency component image H ₃ at level ₃ , and the high-frequency component image H ₄ at level 4 is inversely wavelet transformed to generate a reconstructed image.

For example, when wavelet transform is performed up to level 3 using each of the dovecy wavelets of N = 2, N = 3, N = 4, and N = 6, the wavelet inverse transform unit 105 converts the red inspection region extraction image. A level 1 high frequency component image H ₁ obtained by wavelet transform using a N = 2 Dovecy wavelet, a level 2 high frequency component image H ₂ , and a level 3 high frequency component image H ₃ , respectively. Are subjected to wavelet inverse transform, and a red inspection region extraction image is wavelet transformed by N = 3 Dovecy wavelets, and a level 1 high frequency component image H ₁ , a level 2 high frequency component image H ₂ , and each of the high-frequency component image H ₃ of level 3 is the inverse wavelet transform, the red examination region extraction image Dobe sea wavelet N = 4 Accordingly, the level 1 high-frequency component image H ₁ , the level 2 high-frequency component image H ₂ , and the level 3 high-frequency component image H _{3 obtained by} the wavelet transform are each subjected to wavelet inverse transform and red. Level 1 high-frequency component image H ₁ , level 2 high-frequency component image H ₂ , and level 3 high-frequency component obtained by wavelet transforming the inspection region extracted image of N = 6 by Dovecy wavelet each component image H ₃ is the inverse wavelet transform.

In addition, the wavelet inverse transform unit 105 performs wavelet transform on the green inspection region extracted image with the N = 2 Dovecy wavelet, and the level 1 high frequency component image H ₁ and the level 2 high frequency component image Each of H ₂ and level 3 high-frequency side component image H ₃ is inversely wavelet transformed, and the level 1 high-frequency side obtained by wavelet transforming the green inspection region extracted image with N = 3 Dovecy wavelet The component image H ₁ , the high-frequency component image H ₂ at level ₂ , and the high-frequency component image H ₃ at level ₃ are each subjected to wavelet inverse transform, and the green inspection region extracted image is converted into N = 4 Dovecy wavelets. Level 1 high frequency component image H ₁ obtained by wavelet transform, Level 2 high frequency component image H ₂ , and Level 3 high frequency component Each image H ₃ is the inverse wavelet transform, the green inspection region extracting high frequency band components images H ₁ of the image level 1, obtained by wavelet transform by Dobe Sea wavelets N = _6, level 2 higher frequency Each of the component image H ₂ and the level 3 high-frequency component image H ₃ is subjected to inverse wavelet transform.

Furthermore, the level 1 high-frequency component image H ₁ and the level 2 high-frequency component image obtained by wavelet transforming the blue inspection region extracted image with the N = 2 Dovecy wavelet by the wavelet inverse transform unit 105. Each of H ₂ and level 3 high frequency side component image H ₃ is inversely wavelet transformed, and the level 1 high frequency side obtained by wavelet transforming the blue inspection region extracted image with N = 3 Dovecy wavelet The component image H ₁ , the high-frequency component image H ₂ of level ₂ , and the high-frequency component image H ₃ of level ₃ are each inversely wavelet transformed, and the blue inspection region extraction image is converted into N = 4 Dovecy wavelets. Level 1 high frequency component image H ₁ obtained by wavelet transform, Level 2 high frequency component image H ₂ , and Level 3 high frequency component image Each partial image H ₃ is the inverse wavelet transform, the high-frequency component image H ₁ level 1 blue inspection region extraction image obtained by wavelet transform by Dobe Sea wavelets N = _6, level 2 high pass The side component image H ₂ and the level 3 high frequency side component image H ₃ are each subjected to wavelet inverse transform.

  Thus, the low-frequency component image is not used for generating the reconstructed image. By doing in this way, the feature point according to the defect of a coating surface is expressed more strongly in a reconstructed image.

  In step S65, the masking unit 106 applies each color and level wavelet transform image mask image to each reconstructed image for each mother wavelet, each color, and each level to mask each reconstructed image. . That is, the masking unit 106 removes unnecessary areas from the reconstructed image for each mother wavelet, for each color, and for each level, and among the parts of the reconstructed image, the part of the image consisting only of the image corresponding to the image of the article 12 To extract.

  In step S66, the superimposing unit 107 superimposes the masked reconstructed images corresponding to the three primary colors of light, and generates an added reconstructed image for each mother wavelet and each level. In other words, the superimposing unit 107 superimposes the reconstructed images for three colors for each mother wavelet and each level. That is, the superimposing unit 107 adds the pixel values of the respective pixels at the corresponding positions of the reconstructed image for the three colors, and sets the sum of the pixel values of the pixels at the positions, thereby setting the three colors. Superimpose the reconstructed images.

  In step S <b> 67, the weighting unit 108 reads the weighting coefficient for each mother wavelet and each level from the weighting coefficient storage unit 102. In step S68, the weighting unit 108 weights the added reconstructed image for each mother wavelet and each level by using a weighting coefficient for each mother wavelet and each level. That is, the weighting unit 108 multiplies the pixel value of each pixel of the addition reconstructed image for each mother wavelet and each level by the weighting coefficient for each mother wavelet and each level, and uses the obtained product as the pixel value of that pixel. And

  In step S69, the superimposing unit 107 superimposes the weighted reconstructed images for each mother wavelet and each level to generate a total image. That is, the superimposing unit 107 adds the pixel values of the respective pixels at the corresponding positions of the weighted mother wavelet and the added reconstructed image for each level, and sets the sum of the pixel values of the pixels at the positions. Thus, the summed image is generated by superimposing the addition reconstructed images.

  By superimposing the additive reconstructed images for each level, the image of particles appearing in the additively reconstructed image of one level, the image of particles appearing in the additively reconstructed image of other levels, and the addition of one level Since the image surrounding the particle in the reconstructed image and the image appearing in the addition reconstructed image at another level are added, the feature point is expressed more strongly. In addition, the illumination units 21-1 to 21-3 irradiate the article 12 with light of each of the three primary colors from different positions for each of the three primary colors, whereby a single-color image of each of the three primary colors of light. However, since each defect has a different shape, the feature point corresponding to the defect is expressed more strongly and larger in the total image. Therefore, a defect is detected more reliably.

  Further, the additive reconstructed images for each mother wavelet and each level are overlapped to generate one total image, and from this one total image, the presence / absence of a defective coating surface is evaluated. Therefore, for each color and each mother wavelet As compared with the case where evaluation is performed for each level, the evaluation determination process can be simplified.

  In step S70, the evaluation determination image creation unit 83 outputs the sum image to the evaluation determination unit 84, and the process of creating the evaluation determination image ends.

  As described above, in the process of creating the evaluation determination image, the wavelet transform unit 103 differs from the inspection region extraction image that is the object image obtained by photographing the painted surface of the article 12 that is the object. Wavelet transform by each of multiple mother wavelets. The wavelet inverse transformation unit 105 inversely transforms the detailed image, which is a high-frequency component image obtained by wavelet transformation for each mother wavelet, for each mother wavelet. Then, the superimposing unit 107 superimposes the reconstructed images obtained by inverse wavelet transform and each mother wavelet.

  The process for creating the evaluation determination image may be executed by the GPU 62 that executes the program. In particular, the wavelet transform in step S62 and the wavelet inverse transform in step S64 in the process of creating the evaluation determination image may be executed by the GPU 62 that executes the program. In this case, the wavelet transform unit 103 and the wavelet inverse transform unit 105 are realized by a GPU 62 that executes a program.

  The process for creating the evaluation determination image may be executed by a programmed FPGA (Field-Programmable Gate Array) (not shown). For example, the FPGA can be provided in place of the GPU 62 shown in FIG. For example, the FPGA can be provided in the same manner as the GPU 62 shown in FIG. In particular, in the process of creating the evaluation determination image, the wavelet transform in step S62 and the wavelet inverse transform in step S64 may be executed by a programmed FPGA. In this case, the wavelet transform unit 103 and the wavelet inverse transform unit 105 are realized by a programmed FPGA.

  Next, the details of the evaluation determination process will be described with reference to the flowchart of FIG. In step S91, the binarization unit 111 binarizes the total image obtained by the process of creating the evaluation determination image, and generates a binarized total image. That is, the binarization unit 111 sets the pixel value of each pixel to 0 or 1 by comparing with a predetermined threshold value or a threshold value obtained from a feature amount such as brightness of the total image. Generate a valuated sum image.

  In step S92, the image adjustment unit 112 adjusts the binarized sum image so that particle analysis is easy, such as deletion of minute particles. In step S93, the feature amount extraction unit 113 includes the feature amount of each particle included in the binarized total image, for example, the size of the particle (the width in the x-axis direction, the height in the y-axis direction, or the area) The average luminance of the particles is calculated from the total image.

  In step S94, the determination unit 114 compares the size of each particle with a predetermined size threshold value to determine whether there is a particle larger than the size threshold value. If it is determined in step S94 that there are particles larger than the size threshold, the procedure proceeds to step S95, and the determination unit 114 refers to the summed image and determines the average luminance of the particles determined to be larger than the size threshold. Is compared with a predetermined brightness threshold to determine whether the average brightness of the particles is higher than the brightness threshold. If there are a plurality of particles larger than the size threshold, in step S95, a determination process is performed for each particle.

  If it is determined in step S95 that the average luminance of any one of the particles is higher than the luminance threshold value, the procedure proceeds to step S96, and the setting unit 115 sets that the defect is detected as the result of the evaluation determination, and evaluates the evaluation. The determination image creation process ends.

  On the other hand, if it is determined in step S94 that there are no particles larger than the size threshold, the procedure proceeds to step S97, and the setting unit 115 sets that the result of the evaluation determination is that no defect is detected, and the evaluation The determination image creation process ends. If it is determined in step S95 that the average luminance of any particle is not higher than the luminance threshold, the procedure proceeds to step S97, and the setting unit 115 indicates that no failure has been detected as a result of the evaluation determination. Then, the process for creating the evaluation determination image is completed.

  As described above, in the evaluation determination process, the evaluation determination unit 84 evaluates the presence / absence of a defective coating surface from the feature amount of the summed image that is a superimposed image obtained by superimposing the reconstructed images.

  In the evaluation determination process, the evaluation determination unit 84 uses a statistical method such as a supervised learning or unsupervised learning learning machine such as a support vector machine to check whether there is a defect in the painted surface from the feature amount of the summed image. May be evaluated.

  In this way, the object image, which is an image obtained by photographing the painted surface of the object, is wavelet transformed by each of a plurality of mutually different mother wavelets, and a high frequency obtained by wavelet transformation for each mother wavelet. Wavelet inverse transform of each side component image for each mother wavelet, reconstructed image obtained by inverse wavelet transform, overlayed by overlaying the reconstructed image for each mother wavelet When the presence / absence of a defect on the painted surface is evaluated from the feature amount of the combined image, more defects on the painted surface can be detected more reliably.

  In addition, although it demonstrated that the presence or absence of the defect of the coating surface of the articles | goods 12 was performed from the object image which is a color image, not only this but the coating of the articles | goods 12 from the object image which is a gray scale image You may make it test | inspect for the presence or absence of the defect of a surface.

  The program executed by the computer may be a program that is processed in time series in the order described in this specification, or in parallel or at a necessary timing such as when a call is made. It may be a program for processing.

  The series of processes described above can be executed by hardware or can be executed by software. When a series of processing is executed by software, a program constituting the software executes various functions by installing a computer incorporated in dedicated hardware or various programs. For example, it is installed from a program recording medium in a general-purpose personal computer or the like.

  The embodiments of the present invention are not limited to the above-described embodiments, and various modifications can be made without departing from the scope of the present invention.

DESCRIPTION OF SYMBOLS 11 Inspection apparatus, 12 articles | goods, 21-1 thru | or 21-3 illumination part, 22 imaging part, 23 processing part, 31 to-be-coated object, 32 coating film, 51 CPU, 52 ROM, 53 RAM, 58 memory | storage part, 61 Removable media , 62 GPU, 81 original image acquisition unit, 82 preprocessing unit, 83 evaluation determination image creation unit, 84 evaluation determination unit, 85 result output unit, 91 separation unit, 92 luminance adjustment unit, 93 gray scale unit, 94 mask generation Unit, 95 coordinate acquisition unit, 96 trimming unit, 97 masking unit, 101 mother wavelet storage unit, 102 weight coefficient storage unit, 103 wavelet transform unit, 104 selection unit, 105 wavelet inverse transform unit, 106 masking unit, 107 superposition unit , 108 weighting unit, 111 binarizing unit, 112 image adjusting unit, 113 feature quantity extracting unit, 14 judging unit, 115 setting unit

Claims (10)

In inspection equipment that inspects the presence or absence of defects on the painted surface of the object being painted,
Wavelet transform means for wavelet transforming the object image, which is an image obtained by photographing the painted surface of the object, with each of a plurality of different mother wavelets;
Wavelet inverse transform means for inversely transforming wavelet inverse for each mother wavelet, the high-frequency component image obtained by wavelet transform for each mother wavelet;
A reconstructed image obtained by inverse wavelet transform, and superimposing means for superimposing the reconstructed image for each mother wavelet;
An inspection apparatus comprising: an evaluation unit that evaluates the presence / absence of a defective painted surface from a feature amount of a superimposed image obtained by superimposing the reconstructed images. The inspection apparatus according to claim 1,
The wavelet transform unit is an inspection device that wavelet transforms the object image by each of the mother wavelets, which are four different Dovecy wavelets. The inspection apparatus according to claim 1,
The wavelet transform unit performs wavelet transform on the object image to a predetermined level,
The wavelet inverse transform means performs wavelet inverse transform for each of the mother wavelet for each of the high-frequency component images for each level,
The superimposing means superimposes the reconstructed image for each level and for each mother wavelet. The inspection apparatus according to claim 3, wherein
The wavelet transform unit is an inspection apparatus that wavelet transforms the object image up to level 3. The inspection apparatus according to claim 1,
Separating means for separating the object image, which is a color image, into images of the three primary colors;
The wavelet transform unit performs wavelet transform on each of the object images of the three primary colors by each of the mother wavelets,
The wavelet inverse transforming means performs wavelet inverse transform for each of the mother wavelets for each of the high-frequency side component images of the three primary colors obtained by wavelet transform for each of the mother wavelets,
The superimposing means superimposes the reconstructed image for each of three primary colors and for each mother wavelet. The inspection apparatus according to claim 5, wherein
An inspection apparatus further comprising a light source that irradiates the object with light of each of the three primary colors from different positions for each of the three primary colors. The inspection apparatus according to claim 1,
An inspection apparatus in which the wavelet transform unit and the wavelet inverse transform unit are realized by a GPU (Graphics Processing Unit) that executes a program. The inspection apparatus according to claim 1,
An inspection apparatus in which the wavelet transform unit and the wavelet inverse transform unit are realized by a programmed FPGA (Field-Programmable Gate Array). In an inspection method for inspecting the presence or absence of defects on the painted surface of the object being painted,
The object image, which is an image obtained by photographing the painted surface of the object, is wavelet transformed by each of a plurality of different mother wavelets,
The high-frequency side component image obtained by wavelet transformation for each mother wavelet is subjected to wavelet inverse transformation for each mother wavelet,
A reconstructed image obtained by inverse wavelet transform, wherein the reconstructed images for each mother wavelet are superimposed,
An inspection method including a step of evaluating whether or not a painted surface is defective from a feature amount of a superimposed image obtained by superimposing the reconstructed images. In the computer that controls the inspection device that inspects the coated surface of the object being painted for defects
The object image, which is an image obtained by photographing the painted surface of the object, is wavelet transformed by each of a plurality of different mother wavelets,
The high-frequency side component image obtained by wavelet transformation for each mother wavelet is subjected to wavelet inverse transformation for each mother wavelet,
A reconstructed image obtained by inverse wavelet transform, wherein the reconstructed images for each mother wavelet are superimposed,
A program for performing a process including a step of evaluating the presence / absence of a coating surface defect from a feature amount of a superimposed image obtained by superimposing the reconstructed images.

Images