Method for inspecting unevenness in observation image of film

Abstract

To provide a method for inspecting unevenness in an observation image of a film such as interference unevenness in which variation in evaluation is suppressed and which is efficient.SOLUTION: A method for inspecting unevenness in an observation image of a film comprises: a first step of irradiating the film with diffused light, imaging reflected light from the film with an imaging device, and acquiring an image, the image being composed of a two-dimensional matrix of a large number of pixels, each pixel having a gradation value; a second step of dividing the image into a matrix of a plurality of rectangular minimum units; a third step of acquiring a representative value from each of the minimum units with respect to one or more types of numerical items, and thereby acquiring a representative value group for the film; a fourth step of acquiring statistical values of the representative value group for each numerical item as feature quantities; and a fifth step of determining the degree of unevenness by machine learning on the basis of the feature quantities.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a method for inspecting unevenness in an observed image of a film, and more particularly to a method for inspecting the degree of interference unevenness when light is irradiated onto a transparent film and the light reflected from the film is observed. [Background technology]

[0002] In industrial film production, it is necessary to control the quality of films produced in large quantities. Therefore, in order to control the quality of films produced in large quantities, various methods have been proposed for automatically inspecting films from various viewpoints (for example, Patent Documents 1 to 3).

[0003] In particular, films used for optical applications are often required to have minimal unevenness in the optically observed image. In particular, for certain types of films, such as transparent films composed of multiple layers with different refractive indices, interference unevenness may be observed when irradiating the film with light and observing the reflected light from the film, due to minute unevenness in thickness, etc. For films used for optical applications, interference unevenness greater than a certain level is considered an undesirable defect. For example, if a film exhibiting significant interference unevenness is used as a component of a display device such as a liquid crystal display device, it may impair the display quality of the display device. Therefore, when manufacturing films for optical applications, it is required to inspect the manufactured films to ensure that the interference unevenness is below a certain level. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2009-264915 [Patent Document 2] Japanese Patent Application Laid-Open No. 2006-038549 [Patent Document 3] Japanese Patent Application Laid-Open No. 2016-142739 Summary of the Invention [Problem to be solved by the invention]

[0005] Inspection for interference mura is usually performed by an inspector visually observing the film and judging the degree of mura. Such inspections that rely on the inspector's judgment can lead to problems such as variations in evaluations between different inspectors and difficulty in automating the process to improve efficiency. However, no automated method for inspecting interference mura with high accuracy has been found to date.

[0006] Therefore, an object of the present invention is to provide an efficient method for inspecting unevenness in an observed image of a film, such as interference unevenness, which suppresses variations in evaluation. [Means for solving the problem]

[0007] In order to solve the above problems, the inventors conducted research and came up with the idea of irradiating a film with diffused light, capturing the reflected light with an imaging device to obtain an image, and subjecting the image to processing including machine learning. The inventors further conducted research and found that the above problems can be solved when a specific processing is adopted, and thus completed the present invention. That is, the present invention is as follows.

[0008] (1) A method for inspecting unevenness in an observed image of a film, comprising: a first step of irradiating the film with diffused light and capturing the light reflected from the film with an imaging device to obtain an image, the image being composed of a two-dimensional matrix of a large number of pixels, each of the pixels having a grayscale value; a second step of dividing the image into a matrix of a plurality of rectangular minimum units; a third step of acquiring a representative value from each of the minimum units for one or more types of quantification items, and acquiring a group of representative values for the film; A fourth step of acquiring, as a feature, a statistical value of the representative value group for each of the quantified items; and An inspection method including a fifth step of determining the degree of unevenness by machine learning based on the feature amount. (2) The inspection method according to (1), wherein the quantified item obtained in the third step is the standard deviation of the gradation values of the pixels in each of the smallest units, the difference between the maximum and minimum gradation values of the pixels in each of the smallest units, or a combination thereof. (3) The inspection method according to (1) or (2), wherein the statistical value acquired in the fourth step is a mean value, a standard deviation, a median value, a maximum value, a minimum value, a difference between the maximum value and the minimum value, a kurtosis, a skewness, or a combination thereof. [Effects of the Invention]

[0009] According to the present invention, an efficient method for inspecting unevenness in an observed image of a film, such as interference unevenness, is provided, which suppresses variations in evaluation. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a side view that schematically shows an example of an imaging system for performing the first step of the inspection method of the present invention, and an example of how the first step is performed using the imaging system. [Figure 2] FIG. 2 is a plan view that schematically shows an example of the image of the film in the image acquired in the first step and divisions of the image. [Figure 3] FIG. 3 is a block diagram showing an example of the functions and configuration of an information processing device for performing the second to fifth steps of the inspection method of the present invention. [Figure 4] FIG. 4 is a flowchart showing an example of the second to fifth steps of the inspection method of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0011] The present invention will be described in detail below with reference to embodiments and examples. However, the present invention is not limited to the embodiments and examples shown below, and can be implemented with any modifications within the scope of the claims of the present invention and their equivalents.

[0012] In this application, a "long" film refers to a web having a length that is 5 times or more its width, preferably 10 times or more its width, and specifically refers to a web having a length that can be wound into a roll for storage or transportation. There is no particular upper limit to the length of a long web, and it can be, for example, 100,000 times or less its width.

[0013] (Inspection method: Overview and inspection target) The inspection method of the present invention is a method for inspecting unevenness in an observation image of a film, and specifically, can be useful as a method for inspecting interference unevenness in an observation image of a transparent film. More specifically, it is particularly useful as a method for inspecting interference unevenness in an optical film that is required to have small unevenness in an optical observation image. Even more specifically, it is particularly useful as a method for inspecting interference unevenness in a transparent optical film that is composed of multiple layers with different refractive indices, in which the occurrence of interference unevenness is often a problem.

[0014] A preferred example of a film to be inspected in the inspection method of the present invention is a transparent optical film consisting of multiple layers, including a layer of an alicyclic structure-containing polymer resin. Examples of the alicyclic structure-containing polymer in such an optical film include a polymer obtainable by a polymerization reaction using a cyclic olefin as a monomer, or a hydrogenated product thereof. Furthermore, the alicyclic structure-containing polymer can be either a polymer containing an alicyclic structure in the main chain or a polymer containing an alicyclic structure in the side chain. Of these, it is preferable that the alicyclic structure-containing polymer contains an alicyclic structure in the main chain. Examples of the alicyclic structure include a cycloalkane structure and a cycloalkene structure, with a cycloalkane structure being preferred from the standpoint of thermal stability, etc.

[0015] The number of carbon atoms contained in one alicyclic structure is preferably 4 or more, more preferably 5 or more, more preferably 6 or more, and is preferably 30 or less, more preferably 20 or less, particularly preferably 15 or less. When the number of carbon atoms contained in one alicyclic structure is within the above range, a high level of balance between mechanical strength, heat resistance, and moldability is achieved.

[0016] The proportion of repeating units having an alicyclic structure in the alicyclic structure-containing polymer is preferably 30% by weight or more, more preferably 50% by weight or more, even more preferably 70% by weight or more, and particularly preferably 90% by weight or more. By increasing the proportion of repeating units having an alicyclic structure as described above, heat resistance can be improved. In the alicyclic structure-containing polymer, the remainder other than the repeating unit having the alicyclic structure is not particularly limited and can be appropriately selected depending on the intended use.

[0017] Examples of the polymer having an alicyclic structure include (1) norbornene polymers, (2) monocyclic olefin polymers, (3) cyclic conjugated diene polymers, (4) vinyl alicyclic hydrocarbon polymers, and hydrogenated versions thereof. Among these, norbornene polymers and hydrogenated versions thereof are preferred from the viewpoints of transparency and moldability.

[0018] Examples of norbornene-based polymers include ring-opening polymers of monomers having a norbornene structure and their hydrogenated products; and addition polymers of monomers having a norbornene structure and their hydrogenated products. Examples of ring-opening polymers of monomers having a norbornene structure include ring-opening homopolymers of one type of monomer having a norbornene structure, ring-opening copolymers of two or more types of monomers having a norbornene structure, and ring-opening copolymers of a monomer having a norbornene structure and any monomer copolymerizable therewith. Examples of addition polymers of monomers having a norbornene structure include addition homopolymers of one type of monomer having a norbornene structure, addition copolymers of two or more types of monomers having a norbornene structure, and addition copolymers of a monomer having a norbornene structure and any monomer copolymerizable therewith. Among these, hydrogenated ring-opening polymers of monomers having a norbornene structure are particularly suitable from the viewpoints of moldability, heat resistance, low moisture absorption, low moisture permeability, dimensional stability, and light weight.

[0019] Examples of monomers having a norbornene structure include bicyclo[2.2.1]hept-2-ene (common name: norbornene), tricyclo[4.3.0.1 2,5 ]Deca-3,7-diene (common name: dicyclopentadiene), 7,8-benzotricyclo[4.3.0.1 2,5 ]dec-3-ene (common name: methanotetrahydrofluorene), tetracyclo[4.4.0.1 2,5 .1 7,10 ]dodec-3-ene (trivial name: tetracyclododecene) and derivatives of these compounds (for example, those having a substituent on the ring). Examples of the substituent include an alkyl group, an alkylene group, and a polar group. These substituents may be the same or different, and a plurality of them may be bonded to the ring. The monomer having a norbornene structure may be used alone or in combination of two or more kinds in any ratio.

[0020] The thickness of the film to be inspected and the thickness of each layer constituting the film are not particularly limited, but the inspection method of the present invention can be advantageously applied to films having a thickness that is likely to cause interference unevenness at a level that is likely to cause variations in evaluation by visual judgment.Specifically, the total thickness of the film is preferably 5 μm or more, more preferably 10 μm or more, while preferably 200 μm or less, more preferably 100 μm or less.Furthermore, when the film to be inspected is a multilayer film, the thickness of each layer is preferably 0.1 μm or more, more preferably 0.5 μm or more, while preferably 7 μm or less, more preferably 10 μm or less.

[0021] The testing method of the present invention includes the following first to fifth steps. First step: A step of irradiating a film with diffused light and capturing the light reflected from the film with an imaging device to obtain an image, the image consisting of a two-dimensional matrix of a large number of pixels, each of which has a gray scale value. Second step: Dividing the image into a matrix of minimum rectangular units. Third step: A step of obtaining a representative value from each of the smallest units for one or more types of quantification items, and obtaining a group of representative values for the film. Step 4: A step of obtaining the statistical values of the representative values for each quantification item as features. Fifth process: The degree of unevenness is determined using machine learning based on the feature values.

[0022] (Preliminary process: Laminating the light-shielding layer) In the inspection method of the present invention, in order to more clearly image unevenness, a step of attaching a light-shielding layer to the film to be inspected may be carried out prior to the first step. Such a light-shielding layer may be an adhesive tape for visual inspection with a uniform black surface (for example, a product name "Kukkiri Mieru" manufactured by Tomoegawa Paper Co., Ltd.). The step of attaching the light-shielding layer may be carried out by attaching such a light-shielding layer to one surface of the film to be inspected. By attaching the light-shielding layer, the influence of the background of the film can be reduced, allowing for clearer images of unevenness to be captured under consistent photographing conditions, thereby better suppressing variations in evaluation. When the light-shielding layer is attached, the first step can be carried out by subjecting the resulting inspection target-light-shielding layer composite to the first step.

[0023] When the film to be inspected is a long film and the film is transported in the longitudinal direction and the inspection is to be performed continuously, the same effect can be achieved by supporting the film using a black support device instead of attaching such a light-shielding layer. Examples of such a support device include a support roll and a floating transport device having a black support surface that supports the surface of the film opposite the imaging device.

[0024] (1st step) In the first step, diffused light is irradiated onto the film, and the light reflected from the film is captured by an imaging device to obtain an image.

[0025] Fig. 1 is a side view showing an imaging system for carrying out the first step of the inspection method of the present invention, and a schematic diagram of an example of implementing the first step using the imaging system. In Fig. 1, imaging system 10 includes an inspection stand 110, a diffused light source 120, and an imaging device 130. Inspection stand 110 includes a surface 111 on which a film to be inspected, or a composite of the film to be inspected and a light-shielding layer, can be placed and imaged.

[0026] In the example of FIG. 1, inspection object-light-shielding layer composite 190 is a composite of film 180, which is the inspection object, and light-shielding layer 191, and film 180 is a film made up of two transparent layers 181 and 182. Composite 190 is installed so that the surface on the light-shielding layer 191 side is in contact with rack surface 111. By installing composite 190 in this manner, it is possible to achieve good imaging of the inspection object.

[0027] The light irradiated onto the film in the first step is diffused light, which is light that includes light rays traveling in multiple non-parallel directions, and can be generated by combining a light source with a diffuser plate that diffusely transmits or reflects light.

[0028] In the example of FIG. 1 , the diffuse light source 120 irradiates the composite 190 with diffuse light from a direction that forms a polar angle θ2 in a downward azimuth with respect to the normal direction L1 of the frame surface 111. The diffuse light source 120 includes a light emitter 121, a diffusion plate 122, and a housing 123. The light emitter 121 is provided inside the housing 123. The housing 123 itself is light-blocking, and its inner surface is a surface that has high light reflecting properties, such as a surface painted white, which allows the light from the light emitter 121 to exit from the opening.

[0029] Light from the light emitter 121 reaches the diffuser plate 122 provided at the opening of the housing 123, either directly or after being reflected by other elements within the housing 123, and passes through the diffuser plate 122 before being emitted to the outside of the housing 123. Due to reflection within the housing 123 and diffusion by the diffuser plate 122, the light emitted from the diffused light source 120 becomes diffused light traveling in various directions.

[0030] Light that reaches the composite 190 on the frame 111 is reflected by the surface or an internal layer of the composite 190. The imaging device 130 is a device that can capture this reflected light as an image. Such an image is made up of a two-dimensional matrix of many pixels, each of which has a grayscale value, and the imaging device is a device that can acquire electronic information as such an image.

[0031] 1, the imaging device 130 captures an image of the reflected light from the composite 190 from a direction that forms a polar angle θ3 in an upward direction with respect to the normal direction L1 of the frame surface 111. In this way, by irradiating diffused light from one direction (downward in the example of FIG. 1) and capturing an image from the opposite direction (upward in the example of FIG. 1), it is possible to easily capture a clear image of unevenness.

[0032] The number of images taken for one piece of film to be inspected may be one or more. When multiple images are taken for one piece of film to be inspected, they may be averaged to cancel out environmental errors, and the image obtained may be used for the next process.

[0033] In order to clearly grasp the unevenness of the observed image, it is preferable to perform the first step in an environment with little light other than that from a predetermined diffuse light source. Specifically, it is preferable to place the imaging system 10 described above in a darkroom environment and perform the first step there.

[0034] (2nd process) In the second step, the image acquired in the first step is divided into a matrix of a plurality of rectangular minimum units.

[0035] Fig. 3 is a block diagram showing an example of the functions and configuration of an information processing device for performing steps 2 to 5 of the inspection method of the present invention. In Fig. 3, information processing device 30 includes a division processing unit 301 for performing step 2, a representative value group acquisition unit 302 for performing step 3, a feature value acquisition unit 303 for performing step 4, and a learning processing unit 304 for performing step 5.

[0036] The information processing device 30 may be, for example, an information processing device such as a PC (Personal Computer). The functions of the classification processing unit 301, the representative value group acquisition unit 302, the feature value acquisition unit 303, and the learning processing unit 304 of the information processing device 30 may be realized by a control unit (not shown) reading and executing a program stored in a storage unit (not shown). The storage unit may include a volatile storage area such as a random access memory (RAM); a non-volatile storage area such as a read-only memory (ROM) or a hard disk drive (HDD). The control unit may be a central processing unit (CPU), a graphical processing unit (GPU), or a general-purpose computing on graphics processing unit (GPGPU).

[0037] After the first step is performed, the image acquired in the first step is captured in an information processing device 30 shown in FIG.

[0038] FIG. 4 is a flowchart showing an example of the second to fifth steps of the inspection method of the present invention. The following describes how the second to fifth steps are performed using the information processing device shown in FIG. 3. First, in the second step S10, the image acquired in the first step is imported into the information processing device 30, and the division processing unit 301 divides the acquired image into a matrix of a plurality of rectangular minimum units. The dimensions of the minimum units can be adjusted as appropriate to ensure good inspection. Specifically, the length of one side of the minimum unit is preferably about 10 mm to 100 mm, corresponding to the actual size of the corresponding film. The shape of the minimum unit can be a square or a rectangle.

[0039] The number of minimum units corresponding to one sheet of film can be adjusted as appropriate to enable good inspection. Specifically, it can be adjusted as appropriate depending on the actual size of the film, the preferred dimensions of the minimum units described above, etc. The number of rows and columns in the matrix of minimum units can be adjusted as appropriate to obtain the appropriate number of minimum units and to obtain a matrix with a shape corresponding to the shape of the film. When the film to be inspected is a sheet of film, the matrix of minimum units can be set at a position within the image that occupies a portion that is the image of the film. In particular, it is preferable to set the matrix of minimum units near the center of the image of the film.

[0040] FIG. 2 is a plan view schematically illustrating an example of the image of the film in the image acquired in the first step and a division of the image. In FIG. 2, 190im is an image of the inspection object-light-shielding layer complex 190 in the image captured by the imaging device 130. Because the image was captured from diagonally above the rectangular complex 190, the image is slightly distorted and has a trapezoidal shape. Even if there is some distortion, as long as the image reflects the unevenness actually observed on the film, the inspection method can be successfully implemented.

[0041] Within the image 190im, an inspection area is set, indicated by a rectangular frame defined by an upper edge 211U, a lower edge 211D, a right edge 211R, and a left edge 211L. The inspection area is set in the portion of the film to be inspected, and is usually set in the center of the image 190im. That is, its vertical position is determined so that the line dividing the inspection area into two equal parts, vertically and horizontally, coincides with the line LC, which also divides the image 190im into two equal parts vertically and horizontally, and its horizontal position is determined so that the line dividing the inspection area into two equal parts horizontally and horizontally coincides with the line also dividing the image 190im into two equal parts horizontally. A matrix consisting of a large number of minimum units 212 is set within this area. As a result, in this example, the image 190im is divided into a matrix of minimum units, 4 vertically and 10 horizontally, at its center.

[0042] (3rd step) In the third step S20, the representative value group acquisition unit 302 acquires a representative value for one or more types of quantification items from each of the minimum units determined by dividing the image in the second step S10. As a result of acquiring a representative value for each of the many minimum units set in the image corresponding to one piece of film, many representative values for one piece of film are obtained, and these are acquired as a representative value group for that piece of film.

[0043] In this application, a quantified item is some statistical item that can define a representative value for a set of gradation values of pixels in the smallest unit. Examples of quantified items include the standard deviation of gradation values of pixels in the smallest unit, the difference between the maximum and minimum gradation values of pixels in the smallest unit, or a combination of these. There may be only one type of quantified item, but it is preferable to use two or more types of quantified items in order to perform an inspection with higher judgment accuracy.

[0044] (4th step) In the fourth step S30, the feature value acquisition unit 303 acquires the statistical values of the representative value groups for each quantification item as feature quantities. As a result of acquiring feature quantities for a group of representative values obtained for a certain quantification item of a single film, one feature value is obtained for that quantification item of a single film. The statistical quantity obtained for a group of representative values may be one type or multiple types. Specifically, if n representative value groups GR(1), GR(2), ..., and GR(n) are obtained for a single inspection target film, and m feature quantities are obtained for each representative value group, the number of feature quantities obtained for a single inspection target film will be n x m.

[0045] In this application, a statistical value is any statistical item that can further define a representative value for a group of representative values, which is a set of representative values. Examples of statistical values include the mean, standard deviation, median, maximum value, minimum value, difference between the maximum and minimum values, kurtosis, skewness, or a combination thereof.

[0046] (5th step) In the fifth step S40, the learning processing unit 304 determines the degree of unevenness through machine learning based on the feature values obtained in the fourth step S40. Specific machine learning techniques may include various known techniques. For example, techniques from various categories, such as supervised learning, unsupervised learning, and reinforcement learning, may be employed. Since it is relatively easy to collect a large amount of training data when inspecting unevenness in an observed image of a film, supervised learning techniques can be easily used. Therefore, supervised learning is preferable because it allows for easy execution of an inspection that is highly accurate and allows for the establishment of an inspection process.

[0047] Various known classification models may be used as the machine learning classification model. Examples of the classification model include the MLP Classifier, Extra Trees Classifier, Gaussian Process Classifier, Bagging Classifier, KNeighbors Classifier, Linear SVC, Ridge Classifier CV, Calibrated Classifier CV, Label Spreading, Linear Discriminant Analysis, NuSVC, Random Forest Classifier, Ridge Classifier, SVC, and Gradient Boosting Classifier. The present inventors have found that, of these, the Extra Trees Classifier, Bagging Classifier, Ridge Classifier CV, and Random Forest Classifier are preferable because they can achieve particularly high-accuracy testing.

[0048] Specifically, the fifth step S40 using supervised learning can be performed as follows. That is, first, for a large number of films to obtain training data, the above-described first to fourth steps are performed to obtain feature quantities, and a visual evaluation of unevenness is performed to obtain data on the relationship between the feature quantities and the visual evaluation results. This is used as training data, and based on this, an appropriate machine learning separation model is used to learn the relationship between the feature quantities and the evaluation results. For an inspection target film whose unevenness level is unknown, the first to fourth steps are performed to obtain feature quantities, and based on these feature quantities and the results of the above learning, a task is performed to determine what evaluation result will be obtained. This makes it possible to determine the unevenness level of the inspection target film based on machine learning.

[0049] Optional Components and Modifications The inspection method of the present invention is not limited to the specific example described above, and any other components may be added. Also, the specific example described above may be modified.

[0050] For example, in the example described above, the film to be inspected is a rectangular sheet of film, but the present invention is not limited to this. The inspection object may be a long film, which is divided along lines along the width direction into multiple sections connected in the longitudinal direction, and each of these sections may be inspected individually, with unevenness being inspected for each section. [Example]

[0051] The present invention will be described in detail below with reference to examples. However, the present invention is not limited to the examples shown below, and can be implemented with any modifications within the scope of the claims of the present invention and their equivalents.

[0052] In the following description, the "%" and "parts" that represent amounts are by weight unless otherwise specified. Furthermore, the operations described below were carried out under normal temperature and pressure conditions unless otherwise specified.

[0053] Example 1 (1-1. Manufacturing of the object to be inspected) An alicyclic structure-containing polymer resin (product name "ZEONOR 1430", manufactured by Zeon Corporation) was dissolved in a solvent to prepare a resin solution with a solid content of 10.9%. The solvent used was a mixed solvent of cyclohexane and ethylcyclohexane (the ratio of cyclohexane to ethylcyclohexane was 2:1 (weight ratio)).

[0054] A PET film (product name "Uni-Peel P786-38", thickness 30 μm, manufactured by Unitika Ltd.) was prepared. A resin solution was applied to the PET film and dried to produce a test film having a layer structure of (PET film layer) / (alicyclic structure-containing polymer resin layer). The film thickness of the alicyclic structure-containing polymer resin layer in the test film was set to 3 μm by adjusting the amount of application.

[0055] (1-2. Lamination of light-shielding layer) A black adhesive tape for light-shielding appearance inspection (product name "Kukkiri Mieru", manufactured by Tomoegawa Paper Co., Ltd.) was prepared as the light-shielding layer. A light-shielding layer was attached to the PET film layer side of the inspection target film obtained in (1-1), and cut to appropriate dimensions to obtain 60 inspection target-light-shielding layer composites with a layer structure of (light-shielding layer) / (PET film layer) / (alicyclic structure-containing polymer resin layer). Each inspection target-light-shielding layer composite was rectangular in shape, with short sides of 200 mm and long sides of 300 mm.

[0056] (1-3. First step and visual evaluation) An imaging system 10, shown schematically in FIG. 1, was set up in a darkroom environment.

[0057] 1, the inspection object-light-shielding layer composite 190 was placed on the rack surface 111 of the stand 110, and diffused light was irradiated onto the composite 190 from the diffused light source 120. The composite 190 was placed in such a direction that the surface on the light-shielding layer 191 side was in contact with the rack surface 111, and the long side was in the horizontal direction (corresponding to the front-to-back direction on the paper surface of FIG. 1), and the short side was in the approximately vertical direction (corresponding to the up-and-down direction on the paper surface of FIG. 1).

[0058] The diffused light source 120 had a three-wavelength fluorescent tube (product name "FL20SS.EX-N / 18F3" manufactured by Panasonic) as the light emitter 121 and three stacked diffusion layers as the diffuser 122. The three diffusion layers each had a thickness of 2 mm, and their haze values were 45%, 65%, and 90%, respectively, starting from the layer closest to the light emitter 121. The direction of the diffuser was adjusted so that the normal direction from the center of the light exit surface on the diffuser was directed toward the center of the composite 190 to be inspected. The distance (the distance indicated by the length of the dashed line L2 in FIG. 1) was 475 mm, and the direction of the diffused light source 120 with respect to the normal direction L1 of the frame surface 111 (the angle indicated by θ2 in FIG. 1) was 49°.

[0059] The reflected light from the composite 190 was captured by the imaging device 130 to obtain an image. If interference unevenness occurs due to interference of reflected light from layer 181 (alicyclic structure-containing polymer resin layer) and layer 182 (PET film layer) constituting the film 180 to be inspected, such interference unevenness is observed in the observation image of the composite 190 in the image. As the imaging device 130, a device combining a USB monochrome camera (product name "BU406M", manufactured by Toshiba Teli Corporation) and a C-mount lens (product name "3Z4S-LE VS-LLD18", manufactured by Omron Corporation) was used. The distance from the composite 190 to the imaging device 130 (the distance indicated by the length of line L3 in FIG. 1) was 340 mm, and the direction of the imaging device 130 relative to the normal direction L1 of the frame 111 (the angle indicated by θ3 in FIG. 1) was 15°.

[0060] Using the imaging system 10 installed as described above, images were taken of each of the 60 inspection object-light-shielding layer composites 190 obtained in (1-2), and images were acquired.

[0061] In parallel with the imaging operation, one observer visually evaluated the interference unevenness for all 60 inspection target-light-shielding layer composites 190. That is, the same observer visually inspected each of the inspection target-light-shielding layer composites 190 in a state where they were placed on the rack surface 111 and irradiated with diffused light from the diffused light source 120, and evaluated the degree of interference unevenness on a three-point scale: Class 1 was for those with very little interference unevenness, Class 2 was for those with more interference unevenness than Class 1 but no problem as a product, and Class 3 was for those with interference unevenness to the extent that it was problematic as a product.

[0062] (1-4. 2nd process) A second step was carried out for each of the 60 images of the inspection object-light-shielding layer complex 190 obtained in (1-3), in which the images were divided into a matrix of a plurality of rectangular minimum units, as shown schematically in FIG.

[0063] As shown in FIG. 2, the central region of the image 190im of the composite 190 was divided into a matrix of 24 minimum units 212, with a 4×6 matrix. The inspection region in which the minimum units 212 were set was an area indicated by a rectangular frame defined by an upper edge 211U, a lower edge 211D, a right edge 211R, and a left edge 211L. The vertical dimension of the inspection region was set to a dimension equivalent to 110 mm, the actual size of the film itself. Furthermore, along line LC, which divides the image 190im into two equal halves at the top and bottom, the horizontal dimension of the inspection region (the dimension corresponding to the length of arrow A2) was set to a dimension equivalent to the central 220 mm of the actual size of the film itself, 300 mm (the dimension corresponding to the length of arrow A1). As a result, the lengths of the upper edge 211U and the lower edge 211D were approximately equivalent to 211 mm and 230 mm, respectively, the actual size of the film itself. Each minimum unit had 118 pixels vertically by 118 pixels horizontally, that is, 13924 pixels. The gradation of each pixel was 8 bits monochrome (0 for black to 255 for white).

[0064] (1-5. 3rd step) Representative values were obtained for the quantified items of 13,924 pixels in each minimum unit. The quantified items were (a) the standard deviation of the gradation values, and (b) the range of the gradation values (difference between the maximum and minimum values). In other words, the representative value R(a), which is the standard deviation of the gradation values of the pixels contained in each minimum unit, and the representative value R(b), which is the range (difference between the maximum and minimum values), were obtained. As a result, for each test film, a representative value group GR(b), which is a group of 24 representative values R(a), and a representative value group GR(b), which is a group of 24 representative values R(b), were obtained.

[0065] (1-6. 4th step) The feature quantities were calculated for the representative value groups GR(a) and GR(b) for each of the films to be inspected. Specifically, the calculated feature quantities are as follows:

[0066] Regarding the representative value group GR(a): Average of representative values R(a) Standard deviation of representative value R(a) Median of representative value R(a) Maximum representative value R(a) Minimum representative value R(a) Typical value R(a) range Kurtosis of the representative value R(a) Skewness of the typical value R(a)

[0067] Regarding the representative value group GR(b): Average of representative values R(b) Standard deviation of representative value R(b) Median of representative value R(b) Maximum representative value R(b) Minimum representative value R(b) Typical value R(b) range Kurtosis of the representative value R(b) Skewness of the typical value R(b)

[0068] Therefore, a group of 16 features was obtained for each film under test.

[0069] (1-7. Machine learning judgment) The data from 48 films (80%), or 80% of the 60 films to be inspected, was used as training data, and the data from 12 films (20%), or 20%, was used as test data, and machine learning was used to determine the degree of film interference unevenness. That is, the relationship between the 16 feature values of the training data and the visual evaluation results of interference unevenness was learned, and the task was to predict which classification the evaluation results of the test data would fall into based on each of the 16 feature values of the test data. The classification models used for machine learning were each of the several types listed in Table 1 below.

[0070] For each of the 12 pieces of data in the training data, the classification predicted from the results of performing the task using each classification model was compared with the classification determined by visual evaluation, and if they matched, it was considered correct. The percentage of data that were correct out of the 12 pieces of data was calculated to determine the accuracy of the machine learning judgment. The accuracy of each classification model is shown in Table 1 below.

[0071] [Table 1]

[0072] As can be seen from the results in Table 1, the inspection method of the present invention makes it possible to inspect for interference unevenness with an accuracy approaching that of visual inspection to a certain extent. Therefore, this inspection method makes it possible to automate the inspection of interference unevenness and other unevenness, suppress variation, and perform efficient inspection.

[0073] Furthermore, the results in Table 1 show that among the various classification models, particularly when using classification models such as the Extra Trees Classifier, Bagging Classifier, Ridge Classifier CV, and Random Forest Classifier, highly accurate testing is possible, with an accuracy of 0.9 or higher. [Explanation of symbols]

[0074] 10: Imaging system 30: Information processing device 110: Inspection stand 111: Bridge 120: Diffused light source 121: Luminous object 122: Diffuser 123: Cabinet 130: Imaging device 180: Film to be inspected 181: layer 182: layer 190: Inspection object-light-shielding layer complex 190im: Image of the inspection object-light-shielding layer complex 190 191: Light blocking layer 211D: Bottom 211L: Left side 211R: Right side 211U:Top 212: Smallest unit 301: Segmentation processing unit 302:Representative value group acquisition part 303: Feature value acquisition unit 304: Learning processing unit L1: Normal direction of frame surface 111 LC: A line that divides the image 190 mm into two equal parts, top and bottom, at equal heights.

Claims

1. A method for inspecting unevenness in an observation image of a film, comprising: a first step of irradiating the film with diffused light and capturing the light reflected from the film with an imaging device to obtain an image, the image being composed of a two-dimensional matrix of a large number of pixels, each of the pixels having a grayscale value; a second step of dividing the image into a matrix of a plurality of rectangular minimum units; a third step of acquiring a representative value from each of the minimum units for one or more types of quantification items, and acquiring a group of representative values for the film; A fourth step of acquiring, as a feature, a statistical value of the representative value group for each of the quantified items; and An inspection method including a fifth step of determining the degree of unevenness by machine learning based on the feature amount.

2. 2. The inspection method according to claim 1, wherein the quantified item acquired in the third step is a standard deviation of the gradation values of the pixels in each of the minimum units, a difference between the maximum and minimum gradation values of the pixels in each of the minimum units, or a combination thereof.

3. 3. The inspection method according to claim 1, wherein the statistical value acquired in the fourth step is a mean value, a standard deviation, a median value, a maximum value, a minimum value, a difference between the maximum value and the minimum value, a kurtosis, a skewness, or a combination thereof.

Images