JP6854992B1 - Bright pigment judgment method, bright pigment judgment device and bright pigment judgment program - Google Patents

Abstract

The brilliant pigment determination method is a method of determining the brilliant pigment from a subject to be determined containing the brilliant pigment, and is an imaging step of imaging the reflective surface of the subject to be determined by an imaging element and an imaged reflected image. Learning to learn the type and / or content of the bright pigment based on the image feature amount and the extraction step of extracting the image feature amount from the image feature amount, and the relationship between the image feature amount and the bright pigment by machine learning. The actual field size of the reflected image is 0.01 cm2 or more, including a determination step of determining using a completed model.

Description

The present invention relates to a brilliant pigment determination method for determining the brilliant pigment from a subject to be determined containing the brilliant pigment, a brilliant pigment determination device, and a brilliant pigment determination program. The present application claims priority based on Japanese Patent Application No. 2019-121936 filed in Japan on June 28, 2019, the contents of which are incorporated herein by reference.

In order to obtain a high-quality coating color, a brilliant material-containing coating film using a brilliant pigment (also referred to as a brilliant material) such as an aluminum flake pigment or an interference mica pigment is often used for automobile outer panels and the like.
For example, when performing repair coating on an automobile outer panel, it is desired that the type and presence / absence of such a bright material-containing coating film can be easily identified by a non-destructive measurement method.
Conventionally, in order to determine the brilliant pigment contained in the coating film containing the brilliant pigment, a determination operation by visual confirmation or microscopic observation has been performed. However, the judgment accuracy by such a judgment work depends on the experience and knowledge of the observer, the photographing device is expensive or large, and the analysis takes time.

For example, in Patent Document 1, the variation index obtained by numerically processing the variation reflectance data having the light receiving angle as a variable and the spectral reflectance data having a wavelength as a variable are numerically obtained. A method for identifying the type of bright material based on the correlation information of the two indices of the spectral index obtained by processing is described. However, in recent years, in order to further improve the design, multiple bright pigments such as aluminum flake pigments and interfering mica pigments have been used, and the structure has a multi-layer structure such as pearl clear, and there are a wider variety of complicated specifications. , It is becoming difficult to identify by this method.

On the other hand, in Patent Document 2, the image feature amount obtained by extracting the image data of the vicinity portion including one particle of the bright pigment from the image obtained by imaging the coating film containing the bright pigment as the image data to be processed. A method for identifying a brilliant pigment using a database that associates information with the brilliant pigment is described. In the database, the image feature amount and the information of the bright pigment are associated with each other by pre-learning by a neural network. The method for identifying a bright pigment described in Patent Document 1 can easily determine and estimate the brand of a bright pigment by collating a database created in advance. However, it is necessary to take an image with a high-magnification and expensive microscope device, and it is necessary to take a plurality of image data of the vicinity portion including one particle of the bright pigment as the image data to be processed, which takes time. Met. In addition, such a microscope device is usually inconvenient because it cannot be easily carried.

Patent Document 3 describes acquisition of image data from a target coating, image analysis using the processor so as to determine at least one bright spot from the image data, and bright color from the bright spot. To perform hue analysis using the processor, to calculate the bright color distribution using the processor, and to use the processor to have the same appearance as the target coating. Disclosed are computer-implemented methods, including producing coating formulations that are or substantially similar. Although this method is easy to acquire image data, it is indispensable to perform hue analysis and calculate the bright color distribution so that at least one bright spot can be discriminated as much as possible, which is time-consuming. In addition, it is not suitable for identifying a coating film having a complicated structure such as a multi-layer coating film.

Japanese Unexamined Patent Publication No. 2003-294622 JP-A-2007-218895 Special Table 2017-535771

However, in recent years, as bright pigments contained in coating films, those using new materials and those in which a plurality of materials are combined have been developed and used one after another. Therefore, it has become more difficult than before to determine the brilliant pigment contained in the coating film. Therefore, there is a demand for a determination method or the like that can determine the brilliant pigment contained in the coating film containing the brilliant pigment with higher accuracy.

Based on the above circumstances, the present invention provides a brilliant pigment determination method, a brilliant pigment determination device, and a brilliant pigment determination program capable of determining a brilliant pigment contained in a coating film more easily and with high accuracy than before. The purpose is to do.

In order to solve the above problems, the present invention proposes the following means.
The method for determining a brilliant pigment according to the first aspect of the present embodiment is a method for determining the brilliant pigment from an object to be determined containing the brilliant pigment, and the reflective surface of the object to be determined is imaged by an image pickup element. Based on the imaging step, the extraction step of extracting the image feature amount from the captured reflection image, and the image feature amount, the type and / or content of the brilliant pigment is determined by the image feature amount and the brilliant pigment. ^{The actual field size of the reflected image is 0.01 cm 2} or more, which includes a determination step of determining the relationship with the image using a trained model learned by machine learning.

The brilliant pigment determination device according to the second aspect of the present embodiment includes an image input unit that captures an image of the reflective surface to be determined by an image pickup device, an extraction unit that extracts an image feature amount from the captured reflected image, and an extraction unit. Based on the image feature amount, a determination unit that determines the type and / or content of the glittering pigment using a learned model in which the relationship between the image feature amount and the glittering pigment is learned by machine learning, and the above. An output unit for displaying the determination result of the determination unit is provided, and the actual field size of the reflected image is 0.01 cm ² or more.

The bright pigment determination program according to the third aspect of the present embodiment is a bright pigment including ^{an image input unit having an image pickup element having an actual field size of 0.01 cm 2} or more, an extraction unit, and a determination unit. A program that controls a determination device, the image input unit is made to image a reflection surface to be judged by the image pickup element, and the extraction unit is made to extract an image feature amount from the captured reflection image, and the determination is made. Based on the image feature amount, the unit is made to determine the type and / or content of the brilliant pigment using a learned model in which the relationship between the image feature amount and the brilliant pigment is learned by machine learning.

According to the brilliant pigment determination method, the brilliant pigment determination device, and the brilliant pigment determination program of the present invention, the brilliant pigment contained in the coating film can be determined more easily and with high accuracy than before.

It is an overall block diagram of the bright pigment determination apparatus which concerns on 1st Embodiment of this invention. It is a figure which shows the outline of the image input part of the same brilliant pigment determination apparatus. This is an example of a reflected image obtained by capturing the reflective surface of the coating film with the image input unit. This is an example of a reflected image obtained by capturing the reflective surface of the coating film with the image input unit. This is an example of a reflected image obtained by capturing the reflective surface of the coating film with the image input unit. It is a figure explaining the Aspecular angle. It is a block diagram of the computer of the same brilliant pigment determination apparatus. It is a functional block diagram of the computer when the computer executes the determination program of a bright pigment.

(First Embodiment)
The first embodiment of the present invention will be described with reference to FIGS. 1 to 8.

[Glittering pigment]
The glitter pigment that can be determined by the glitter pigment determination method according to the present embodiment is, for example, a vapor-deposited metal flake pigment, an aluminum flake pigment, a photocoherent pigment, or the like.

The vapor-deposited metal flake pigment is a pigment obtained by depositing a metal film on a base base material, peeling off the base base material, and then crushing the vapor-deposited metal film. The material of the metal is not particularly limited, and examples thereof include aluminum, gold, silver, copper, brass, titanium, chromium, nickel, nickel chromium, and stainless steel.

Examples of commercially available products that can be used as the vapor-filmed aluminum flake pigments include the "METALURE" series (trade name, manufactured by Ecult), the "Hydrosine WS" series (trade name, manufactured by Ecult), and the "Decomet" series (trade name, manufactured by Schlenk). (Manufactured by BASF), "Metaseen" series (trade name, manufactured by BASF) and the like can be mentioned.

Examples of commercially available products that can be used as the thin-film vapor-deposited chrome flake pigment include the "Metalure Liquid Black" series (trade name, manufactured by Ecult Co., Ltd.) and the like.

The vapor-deposited aluminum flake pigment may have a surface treated with silica.

Aluminum flake pigments are generally produced by pulverizing and grinding aluminum in a ball mill or an attritor mill in the presence of a pulverizing medium solution using a pulverizing aid, and usually has an average particle size (D50) of 1 for paints. A material having a thickness of about 50 μm, particularly about 5 to 20 μm, is used from the viewpoint of stability in the paint and the finish of the formed coating film.

Examples of commercially available products that can be used as the aluminum flake pigment include the "Alpaste" series (trade name, manufactured by Toyo Aluminum Co., Ltd.).

The aluminum flake pigment may have a surface treated with silica.

The photocoherent pigment is a pigment in which a transparent or translucent scaly substrate is coated with a metal oxide. Examples of the scaly substrate include natural mica, artificial mica, alumina flakes, silica flakes, glass flakes and the like. The natural mica is a scaly base material obtained by crushing ore mica (mica). On the other hand, the above-mentioned artificial mica, fluorphlogopite _{(KMg 3 AlSi 3 O10F 2)} , potassium tetrasilisic mica _{(KMg 25 AlSi 4 O 10 F} 2), sodium tetrasilicic mica _{(NaMg 25 AlSi 4 O 10 F} 2), Industrial raw materials such as Na teniolite (NaMg ₂ LiSi ₄ O ₁₀ F ₂ ) and LiNa teniolite (LiMg ₂ LiSi ₄ O ₁₀ F ₂ ) are heated, melted at a high temperature of about 1500 ° C., cooled and crystallized for synthesis. Compared with natural mica, it has less impurities and is uniform in size and thickness.

The alumina flakes mean scaly (flaky) aluminum oxide and are colorless and transparent. The aluminum oxide does not have to be a single component and may contain oxides of other metals.

The silica flakes are scaly silica. The glass flakes are scaly glass.

Examples of the metal oxide include titanium oxide and iron oxide. It is possible to develop different interference colors depending on the thickness of the metal oxide that coats the scaly substrate.

It is preferable that the photocoherent pigment is surface-treated from the viewpoint of obtaining a coating film having excellent storage stability and weather resistance.

Examples of commercially available products that can be used as the photointerfering pigment include the "Iriodin" series (trade name, manufactured by Merck Japan) and the "Shiraric" series (trade name, manufactured by Merck Japan).

[Bright pigment determination device 100]
FIG. 1 is an overall configuration diagram of the glitter pigment determination device 100.
The glitter pigment determination device 100 includes an image input unit 1, a computer 2 capable of executing a program, and an output unit 3 such as an LCD monitor.

FIG. 2 is a diagram showing an outline of the image input unit 1.
The image input unit 1 includes a light source 11 that irradiates the coating film F to be determined, and an image sensor 12 that captures the reflected light from the coating film F. This embodiment is an example. For example, the image sensor 12 may be arranged at the position of the light source 11, and the light source 11 may be arranged at the position of the image sensor 12.

The light source 11 may be any one that irradiates white light, and is, for example, a white LED or a halogen light. For example, the light source 11 is REVOX SLG-150U.

The image sensor (image sensor) 12 receives the reflected light obtained from the reflecting surface of the coating film F that reflects the incident light from the light source 11. As the image sensor 12, one mounted on a general digital camera or the like can be used. Depending on the allowable range of optical conditions, the reflected light may be condensed by using a lens, or the reflected light may be received by the image sensor 12 without using a lens. For example, the image acquisition size by the image sensor 12 is 6x10 mm, and the field of view size is 15x40 mm. The image acquisition size and the field of view size of the image sensor 12 are not limited to this. The image sensor 12 can capture a sufficient image in the subsequent image processing as long as the actual field size of the reflected image obtained by capturing the reflective surface of the coating film F by the image input unit 1 is 0.01 cm ^{2 or more.} From the viewpoint that it is desirable that the actual field size of the reflected image to be captured is 0.01 cm ² or more, a general digital camera can be used as the image sensor 12. Acquiring a reflected image using an electron microscope is unsuitable in that it is not possible to obtain the actual visual field size required for subsequent image processing and that it is not easy. A more preferable actual visual field size is 0.5 cm ² or more, further 1 cm ² or more, and the upper limit is not particularly limited, but may be ^{50 cm 2} or less and 10 cm ^{2 or less.} The image sensor 12 is an example of an image sensor. The image pickup device is, for example, a CCD module, a CMOS module, a photodiode array, or the like.

The center position of the coating film F irradiated by the light source 11 is defined as the “irradiation center point O”. As shown in FIG. 2, the image sensor 12 is arranged at a position where the reflected light reflected in the normal direction N at the irradiation center point O of the coating film F is vertically incident on the image sensor 12.

3 to 5 are examples of reflection images obtained by capturing the reflection surface of the coating film F with the image input unit 1. In FIGS. 3 to 5, the light source 11 is arranged at positions where the incident angles of the incident light on the irradiation center point O of the coating film F with respect to the normal direction N are about 25 degrees, about 45 degrees, and about 75 degrees, respectively. This is a reflection image obtained by imaging the reflection surface of the coating film F. By changing the position of the light source 11, it is possible to image the reflective surface of the coating film F having different aspecular angles. The reflected image may be a reflected image obtained by capturing the reflecting surface of the coating film F that receives light from one incident angle with a plurality of image sensors 12.

FIG. 6 is a diagram illustrating an aspecular angle.
The specular angle is an angle indicating the inclination of the optical path of the reflected light from the normal reflection direction. For example, the illumination angle is expressed by 45 ° from the sample normal direction, and the light receiving angle is expressed by the angle from the normal reflected light angle. It is the angle. For example, the five angles standardized by the international standard ASTM E2194 (sometimes expressed as as15 °, 25 °, 45 °, 75 °, 110 °) can be mentioned. The "as" is an abbreviation for specular and means an angle from specular reflection.

In the present embodiment, the image input unit 1 saves the captured reflection image as an RGB image and transfers it to the computer 2. The storage format of the captured reflection image is not limited to the RGB image format, and may be another image format.

FIG. 7 is a configuration diagram of the computer 2.
The computer 2 is a program-executable processing device having a processor 20, a memory 21 capable of reading a program, a storage unit 22 capable of storing programs and data, and an input / output control unit 23. The function of the computer 2 is realized by the processor 20 executing the program provided to the computer 2.

FIG. 8 is a functional block diagram of the computer 2 when the computer 2 executes a determination program for a bright pigment. The brilliant pigment determination program includes a sorting unit 51, an extraction unit 52, and a determination unit 53 as functional blocks.

(Image conversion process)
The sorting unit 51 may convert the reflected image of the RGB image format captured by the image input unit 1 into an HSV image or a Lab image before the processing by the sorting unit 51 to be performed later (image conversion step). By converting the image format, it is possible to change the amount of image features that are easy to extract in the subsequent processing. Further, the sorting unit 51 may convert an image in the RGB image format into a monochrome image or a grayscale image for use. Further, the sorting unit 51 may convert the image in the RGB image format or the converted image into an image subjected to processing such as FFT (Fast Fourier Transform), Laplace transform, wavelet transform, and convolution filter.

(Threshold processing)
In the sorting unit 51, the reflected image captured by the image input unit 1 is input, and the area of the reflecting surface in the reflected image is a bright range (existence range E) in which there is a high possibility that a bright pigment is present by threshold processing based on the brightness difference. Select. First, the sorting unit 51 selects the presence range E of the bright pigment as the extraction target of the image feature amount from the region of the reflecting surface in the reflected image. The sorting unit 51 specifies the existence range E of the brilliant pigment by the threshold processing without obtaining the hue distribution of the entire reflected image.

The extraction unit 52 extracts the image feature amount from the existence range E separated by the sorting unit 51. The image feature amount to be extracted is at least one of the mean value, median value, mode value, maximum value, minimum value, quartile, percentile, standard deviation, skewness, and kurtosis of the pixel data constituting the image. Including. The extraction unit 52 may use a convolutional neural network or a recursive neural network to extract the image feature amount.

(High-brightness image extraction processing)
The extraction unit 52 may further select the existence range E of the brilliant pigment by the high-luminance image extraction process to extract the image feature amount. The extraction unit 52 is acquired from, for example, a specific range specified within the 10th percentile from the minimum distance from the specific RGB, HSV, and Lab coordinates in the pixel set of the existence range E of the brilliant pigment (A). At least one color value selected from the RGB value, brightness, saturation, hue angle, Lab value, and Fourier conversion value of the pixel or pixel set, and (B) the average value, median value, most frequent value, and maximum value of the above color values. At least one of at least one statistic selected from value, minimum value, quadrant, percentile, variance, standard deviation, skewness, and sharpness is extracted as an image feature amount.

For example, the extraction unit 52 extracts at least one of pixels and / or a set of pixels acquired from a specific range specified within the 10th percentile from the minimum distance from red or blue in RGB coordinates as an image feature amount. To do. By selecting RGB, HSV, Lab coordinates and a reference color in which the characteristics of the brilliant pigment to be determined are likely to appear, pixels and / or pixel sets in which the characteristics of the brilliant pigment are likely to appear can be suitably obtained. Further, by extracting the image feature amount from the pixels selected from the existence range E and / or the pixel set, the characteristics of the brilliant pigment are reflected as compared with the case where the image feature amount is directly extracted from the existence range E. The amount of image features can be preferably obtained.

(Clustering process)
The extraction unit 52 may further select the existence range E of the bright pigment by the clustering process and extract the image feature amount. The extraction unit 52 uses, for example, at least one of the RGB values, brightness, saturation, hue angle, Lab value, and Fourier conversion value of the pixel set in the existence range E of the bright pigment, and the presence of the bright pigment. Clustering of pixel sets in range E is performed, and the clusters sorted by clustering are ranked according to any of RGB value, brightness, saturation, hue angle, Lab value, Fourier conversion value, and distance from a specific color coordinate. RGB value, brightness, saturation, hue angle, Lab value, average value of Fourier transform value, median value, most frequent value, maximum value, minimum value, quadrant, percentile, dispersion, standard deviation, distortion of sorted cluster At least one of degree and sharpness is extracted as an image feature amount.

For example, the extraction unit 52 performs a clustering process using K-means. The clustering process may be a single link method, a complete link method, a group averaging method, a Ward method, a center of gravity method, a weighted averaging method, a median method, DBSCAN, OPTICS, a Gaussian mixture, a Birch method, or the like. The extraction unit 52 clusters the pixel sets in the existence range E of the bright pigment, selects, for example, the cluster having the largest RGB value from the clusters sorted by the clustering, and extracts the image feature amount. By extracting the image feature amount from the cluster in which the bright pigment is likely to be present, the image feature amount reflecting the characteristic of the bright pigment can be suitably obtained.

The determination unit 53 determines the glitter pigment using the "learned model M" which is the learning result of learning the relationship between the glitter pigment and the image feature amount based on the image feature amount extracted for each existence range E. To do. In the present embodiment, the trained model M is composed of a decision tree (Gradient Boosting Decision Tree) using gradient boosting. The input of the trained model M is the image feature amount extracted by the extraction unit 52. The output of the trained model M is the type of bright pigment determined. The determined type of bright pigment is output to the output unit 3.

The trained model M is provided as a program module of a part of the glitter pigment determination program executed by the computer 2 of the glitter pigment determination device 100. The computer 2 may have a dedicated logic circuit or the like for executing the trained model M.

[Generation of trained model M]
The trained model M is generated by prior learning based on the teacher data described later. The trained model M may be generated by the computer 2 of the glitter pigment determination device 100, or may be generated by using another computer having a higher computing power than the computer 2.

The trained model M is generated by supervised learning by a well-known technique such as a gradient descent method, and the weighting coefficient of the decision tree is updated.

In the present embodiment, the teacher data is a reflection image obtained by preliminarily capturing a coating film in which the contained bright pigment is known. In the following description, the reflection image captured in advance for learning is referred to as a “learning reflection image”. Specifically, the teacher data is a combination of the reflection image for learning and the bright pigment type code included in the coating film.

It is desirable to prepare as diverse teacher data as possible by changing the imaging conditions of the reflection image for learning. In particular, by preparing teacher data under various imaging conditions, a trained model M that has high S / N discrimination ability against noise generated under various conditions and can determine a robust bright pigment is generated. can do.

In the pre-learning, the computer 2 selects the existence range E of the brilliant pigment by the threshold processing based on the brightness difference as the image feature amount to be extracted from the reflection image for learning, as in the processing of the sorting unit 51. In the pre-learning, the computer 2 extracts the image feature amount from the existence range E in the same manner as the processing of the extraction unit 52. The type of image feature amount extracted in the pre-learning is the same as that extracted by the extraction unit 52.

The computer 2 receives the image feature amount of the learning reflection image extracted in the pre-learning as an input, and generates a trained model M that outputs a bright pigment type code by supervised learning.

[Operation of Bright Pigment Judgment Device 100]
A method for determining a glitter pigment using the glitter pigment determination device 100 will be described.

The image input unit 1 images the reflective surface of the coating film (target to be determined) F having different aspecular angles with a general digital camera (imaging step).

The computer 2 selects the existence range E of the brilliant pigment as the extraction target of the image feature amount from the region of the reflection surface in the reflection image captured in the imaging step by the threshold processing based on the brightness difference. The computer 2 extracts the image feature amount from the existence range E of the selected bright pigment (extraction step). The computer 2 may further select the existence range E of the brilliant pigment and extract the image feature amount by a high-luminance image extraction process or a clustering process, if necessary.

The computer 2 determines the type of the glitter pigment using the "learned model M" that has learned the relationship between the glitter pigment and the image feature amount based on the image feature amount extracted from the existence range E (determination step). ). At this time, the trained model M may be retrained based on the correct answer rate of the output of the trained model M.

The output unit 3 displays the type of the determined bright pigment.

According to the brilliant pigment determination device 100 according to the present embodiment, since a general imaging device such as an area image sensor such as a CCD or CMOS or a digital camera is used, an extra large space and preparation time are required. Since the image can be easily carried out without any problem and the actual field size of the captured image is sufficiently larger than that of the microscope, it is possible to accurately determine the brilliant pigment.

According to the bright pigment determination device 100 according to the present embodiment, the presence range E of the bright pigment whose image feature amount is to be extracted is selected by the threshold processing based on the brightness difference. The existence range E of the bright pigment to be extracted from the image feature amount can be easily specified without obtaining the hue distribution of the entire reflected image.

According to the brilliant pigment determination device 100 according to the present embodiment, the presence range E of the brilliant pigment is further selected and the image feature amount is extracted by a high-luminance image extraction process or a clustering process as needed. By extracting the amount of image features from pixels and clusters in which the features of the glitter pigment are likely to appear, the glitter pigment contained in the coating film can be determined more easily and with higher accuracy than before.

The bright pigment determination program is recorded on a computer-readable recording medium. The program recorded on this recording medium is read into the computer system and executed. The term "computer system" as used herein includes hardware such as an OS and peripheral devices. Further, the "computer-readable recording medium" refers to a portable medium such as a flexible disk, a magneto-optical disk, a ROM, or a CD-ROM, or a storage device such as a hard disk built in a computer system. Further, a "computer-readable recording medium" is a communication line for transmitting a program via a network such as the Internet or a communication line such as a telephone line, and dynamically holds the program for a short period of time. It may also include a program that holds a program for a certain period of time, such as a volatile memory inside a computer system that serves as a server or a client in that case. Further, the above program may be for realizing a part of the above-mentioned functions, and may be further realized for realizing the above-mentioned functions in combination with a program already recorded in the computer system. It may be realized by using a programmable logic device such as FPGA (Field Programmable Gate Array).

Although the first embodiment of the present invention has been described in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes design changes and the like within a range that does not deviate from the gist of the present invention. .. In addition, the components shown in the above-described first embodiment and the modifications shown below can be appropriately combined and configured.

(Modification example 1)
For example, in the above embodiment, the trained model M is composed of a decision tree using gradient boosting, but the mode of the trained model is not limited to this. If the trained model is a model that determines the type of bright pigment from the image features and outputs it, for example, linear regression, logistic regression, simple perceptron, MLP, neural network, support vector machine, random forest, Gaussian process, etc. It may be composed of a Bayesian network, a k-nearest neighbor method, or other models used in machine learning.

(Modification 2)
For example, in the above embodiment, the type of the bright pigment is used in the determination step, but the mode of the determination step is not limited to this. The determination step may also be performed, for example, to determine the content of the bright pigment. In this case, the input of the trained model is the image feature amount extracted by the extraction unit 52, and the output of the trained model is the determined type of bright pigment and the content of the bright pigment. In the trained model, the combination of the type code of the bright pigment and the content of the bright pigment contained in the reflection image for learning and the coating film is learned as teacher data.

(Second Embodiment)
The brilliant pigment determination device 100B according to the second embodiment of the present invention will be described. In the following description, the same reference numerals will be given to the configurations common to those already described, and duplicate description will be omitted. The image input unit of the glitter pigment determination device 100B is different from that of the first embodiment.

[Bright pigment determination device 100B]
The glitter pigment determination device 100B includes an image input unit 1B, a computer 2 capable of executing a program, and an output unit 3 such as an LCD monitor.

The image input unit 1B is an imaging device capable of capturing images from multiple angles, and examples thereof include colorimeters such as MAT6 and T12 manufactured by x-rite. For example, when MAT6 manufactured by x-rite is used, the angle of the light receiving portion is 15 degrees from the normal direction N, and there are seven types of imaging angles (aspercular angle −45 degrees, aspercular angle -30 degrees, a. It is possible to take an image at a specular angle of -15 degrees, an aspergular angle of 15 degrees, an assicular angle of 45 degrees, an asspicular angle of 80 degrees, and one type of diffused light irradiation image). The image input unit 1B is not limited to a colorimeter, and may be an imaging device that images a reflecting surface by appropriately changing the imaging angle with a built-in digital camera. The digital camera built in the image input unit 1B is equipped with the image sensor 12 of the first embodiment, and the actual field size of the reflected image to be captured is 0.01 cm ² or more. The image input unit 1B saves the reflected image obtained by capturing the reflecting surfaces having different aspecular angles as an RGB image and transfers it to the computer 2.

A plurality of reflected images captured by the image input unit 1B are input to the sorting unit 51 of the computer 2, and the existence range E is selected by threshold processing based on the brightness difference. The following steps are the same as in the first embodiment.

According to the brilliant pigment determination device 100B according to the present embodiment, as in the case of the brilliant pigment determination device 100 of the first embodiment, the presence range E of the brilliant pigment whose image feature amount is to be extracted by the threshold processing based on the luminance difference. Select. The existence range E of the bright pigment to be extracted from the image feature amount can be easily specified without obtaining the hue distribution of the entire reflected image. If necessary, the presence range E of the bright pigment is further selected by high-luminance image extraction processing or clustering processing to extract the image feature amount. By extracting the amount of image features from pixels and clusters in which the features of the glitter pigment are likely to appear, the glitter pigment contained in the coating film can be determined more easily and with higher accuracy than before.

According to the bright pigment determination device 100B according to the present embodiment, since the colorimeter is used as the image input unit 1B, the reflection image for each incident angle is compared with the bright pigment determination device 100 of the first embodiment. Can be obtained with high accuracy, and a more accurate image feature amount can be extracted and used for determination.

Although the second embodiment of the present invention has been described in detail with reference to the drawings, the specific configuration is not limited to this embodiment and includes design changes and the like within a range that does not deviate from the gist of the present invention. .. In addition, the components shown in the second embodiment and the modification described above can be appropriately combined and configured.

(Third Embodiment)
The brilliant pigment determination device 100C according to the third embodiment of the present invention will be described. In the following description, the same reference numerals will be given to the configurations common to those already described, and duplicate description will be omitted. The glitter pigment determination device 100C is different from the second embodiment in that the spectral reflectance is used.

[Bright pigment determination device 100C]
The glitter pigment determination device 100C includes an image input unit 1C, a computer 2 capable of executing a program, and an output unit 3 such as an LCD monitor.

The image input unit 1C further includes a colorimeter or a spectrocolorimeter capable of measuring the spectral reflectance from a plurality of color measurement angles with respect to the image input unit 1B of the second embodiment. A colorimeter or spectrophotometer (sometimes called a spectrophotometer) is a device that can measure colors at multiple viewing angles within the visible region of the spectrum when used to identify color characteristics. There are no particular restrictions. The image input unit 1C including the colorimeter or the spectrophotometer is preferably a portable device. As in the second embodiment, the image input unit 1C images the reflecting surface by the image sensor 12 of the built-in digital camera. In addition, the image input unit 1C acquires the spectral reflectance by a light receiver built in the colorimeter or the spectrophotometer. As the image input unit 1C, for example, a colorimeter such as MAT6 or T12 manufactured by x-rite can be used. For example, when MAT6 manufactured by x-rite is used, the angle of the receiver is 45 degrees from the normal direction N, and the color measurement angle is 6 angles (spectral angle -15 degrees, spectroscopic angle 15 degrees, a spectroscopic angle). Spectral reflectance can be measured by a specular angle of 25 degrees, an aspecular angle of 45 degrees, an aspecular angle of 75 degrees, and an asspecular angle of 110 degrees).

In the present embodiment, the image input unit 1C stores the reflected image captured at the imaging angle as an RGB image, and further measures the spectral reflectance at the color measurement angle corresponding to the imaging angle at which the reflecting surface is imaged ( Measurement process). The image input unit 1C may simultaneously capture the reflected image and measure the spectral reflectance, or may separately perform the imaging. Further, the imaging of the reflected image and the measurement of the spectral reflectance may be performed by separate devices. For example, the imaging of the reflected image may be performed using the image input unit 1A of the first embodiment or the image input unit 1B of the second embodiment.

The image input unit 1C transfers the spectral reflectance corresponding to the RGB image to the computer 2. The storage format of the captured reflection image is not limited to the RGB image format, and may be another image format. The image input unit 1C may capture a monochrome image or a grayscale image instead of the RGB image.

The reflected image captured by the image input unit 1C is used for extracting the image feature amount as in the first embodiment. High-luminance image extraction processing and clustering processing are performed on the reflected image as needed.

The spectral reflectance measured by the image input unit 1C is used for extracting the spectral reflection feature amount. The spectral reflection feature amount is a feature amount exemplified below. The spectral reflection feature amount may be a combination of two or more feature amounts.
(1) Its value, maximum, minimum, median, quadrant or percentile of spectral reflectance or reflectance (2) Maximum, minimum, median, quadrant, or percenttile of spectral reflectance or reflectance (3) Average, dispersion, standard deviation, sharpness, or distortion of the reflectance or reflectance of the spectrum (4) Obtained by a convolution filter, Fourier transform, or wavelet transform on the reflectance or reflectance of the spectrum. Value (5) Peak wavelength of reflectance or reflectance of the spectrum, or difference or ratio between peaks of reflectance or reflectance of the spectrum

The determination unit 53 has a relationship between the bright pigment, the image feature amount, and the spectral reflection feature amount based on the image feature amount extracted for each existence range E and the spectral reflection feature amount extracted from the corresponding spectral reflectance. The bright pigment is determined using the "learned model M2" which is the learning result of learning (determination step). The spectral reflection feature amount may be the spectral reflectance itself.

In the present embodiment, the trained model M2 is composed of a decision tree (Gradient Boosting Decision Tree) using gradient boosting. The inputs of the trained model M2 are the image feature amount and the spectral reflection feature amount extracted by the extraction unit 52. The output of the trained model M2 is the type of bright pigment determined. The determined type of bright pigment is output to the output unit 3.

The trained model M2 is generated by prior learning based on the teacher data. The trained model M2 may be generated by the computer 2 of the glitter pigment determination device 100, or may be generated by using another computer having a higher computing power than the computer 2.

In the present embodiment, the teacher data is a reflection image obtained by preliminarily capturing a coating film in which the contained bright pigment is known. Specifically, the teacher data is a combination of the reflection image for learning and the bright pigment type code included in the coating film.

In the pre-learning, the computer 2 selects the existence range E of the brilliant pigment by the threshold processing based on the brightness difference as the extraction target of the image feature amount and the spectral reflection feature amount from the reflection image for learning, as in the processing of the sorting unit 51. To do. In the pre-learning, the computer 2 extracts the image feature amount and the spectral reflection feature amount from the existence range E in the same manner as the processing of the extraction unit 52. The types of image features and spectral reflection features extracted in the pre-learning are the same as those extracted by the extraction unit 52.

The computer 2 inputs the image feature amount and the spectral reflection feature amount of the learning reflection image extracted in the pre-learning, and generates a trained model M that outputs a glitter pigment type code by supervised learning.

According to the glitter pigment determination device 100C according to the present embodiment, since the image input unit 1C has a colorimeter or a spectrophotometer, it is compared with the glitter pigment determination device 100 of the first embodiment. , The reflected image and the spectral reflectance for each incident angle can be obtained, and the spectral reflectance can be extracted in addition to the image feature amount and used for the determination.

Hereinafter, the present invention will be described in detail based on examples, but the technical scope of the present invention is not limited to these examples.

<Example 1>
(1) Imaging process
As the coating film F to be judged for the bright material pigment determination, a solid coating color and a metallic coating color were selected from the existing toning-designed automobile top coating color reserve coating plate. Here, the metallic coating color means a coating color containing a scaly pigment (bright material pigment) such as aluminum flakes and mica flakes, and the solid coating color means a coating color that does not contain the bright material as a pigment. .. The back coat plate is a test plate obtained by coating a base material having a predetermined size with a paint adjusted so as to obtain the above-mentioned solid paint color and metallic paint color, and then applying a clear paint.

The image field of view of the image input unit 1 used is 30 mm × 5 mm, and the actual field of view size is 0.01 cm ² or more. The image input unit 1 images the reflective surface of the coating film F having an aspecular angle of about 15 degrees. The image input unit 1 saves the captured reflection image as an RGB image.

(2) Image conversion step In Example 1, the image conversion step is not carried out.

(3) Extraction Step In Example 1, the existence range E is not selected by the threshold value processing. Image features are extracted from the entire captured reflected image.

(4) Judgment Step The number of features used for judgment in Example 1 is eight, and the maximum value and standard deviation of the R value of the single-angle overall image, and the median and maximum value of the G value of the single-angle overall image. It is a combination of the standard deviation, the maximum value, the minimum value, and the standard deviation of the G value of the entire single-angle image (feature amount combination No. 2-1).

Further, the trained model M used in Example 1 is composed of a decision tree (Gradient Boosting Decision Tree) using gradient boosting (machine learning model No. 1-3).

<Example 2>
Only the parts that are different from those of Example 1 are described.

(1) Imaging process
The image input unit 1 captures an image of the reflective surface of the coating film F having an asspecular angle of about 15 degrees, 25 degrees, 45 degrees, 75 degrees, and 110 degrees, and saves each as an RGB image as a reflected image.

(4) Judgment Step The number of features used for judgment in Example 2 is 200, which is a combination of the maximum value for each angle of the B value of the entire image and the average value for each angle of the G value of the entire image (features). Quantity combination No. 2-2).

<Example 3>
Only the parts that are different from those of Example 2 will be described.

(3) Extraction Step In Example 3, the existence range E is selected by the threshold processing, and the image feature amount is extracted from the existence range E of the bright pigment.

(4) Judgment Step The number of features used for judgment in Example 3 is 200, which is a combination of the maximum value for each G value angle of the image below the threshold value and the FFT amplitude for each G value angle of the entire image (FFT amplitude). Feature combination No. 2-3).

<Example 4>
Only the parts that are different from those of Example 3 are described.

(2) Image Conversion Step In Example 4, an image conversion step is carried out, and the RGB image is converted into an HSV image or a Lab image. In addition, FFT (Fast Fourier Transform) for RGB images is also performed.

(3) Extraction Step In Example 4, the existence range E is not selected by the threshold value processing. Image features are extracted from the entire captured reflected image.

(4) Judgment step The number of features used for judgment in Example 4 is 300, the third quartile for each angle of the S value of the entire image, the maximum value for each angle of the a value of the entire image, and the entire image. It is a combination of the third quartile of the b value (feature amount combination No. 2-4).

<Example 5>
Only the parts that are different from those of Example 4 are described.

(3) Extraction Step In Example 5, the existence range E is selected by the threshold processing, and the image feature amount is extracted from the existence range E of the bright pigment.

(4) Judgment Step The number of features used for judgment in Example 5 is 900, and the ratio between the angles of the maximum a value of the image below the threshold, the minimum value for each angle of the a value of the image below the threshold, and the threshold. The ratio between the angles of the average V (brightness) values of the image, the ratio of the G values of the entire image between the angles of the third quartile, and the ratio of the average V (brightness) values of the entire image between the angles. , (Characteristic amount combination No. 2-5).

<Example 6>
Only the parts that differ from Example 5 are described.

(4) Judgment Step The number of features used for judgment in Example 6 is 500, which is the maximum value for each angle of the a value of the image above the threshold, the minimum value for each angle of the a value of the image above the threshold, and the image below the threshold. It is a combination of the average value for each angle of the a value and the ratio between the angles of the minimum value of the L value of the image above the threshold value (feature amount combination No. 2-6).

<Example 7>
Only the parts that differ from Example 6 are described.

(4) Judgment Step The number of features used for judgment in Example 7 is 600, and it is between the first quartile for each angle of the G value of the image above the threshold and the angle of the maximum value of the a value of the image below the threshold. It is a combination of the ratio, the ratio between the angles of the average value of the b values of the image above the threshold value, and the maximum value for each angle of the B value of the image below the threshold value (feature amount combination No. 2-7).

<Example 8>
Only the parts that differ from Example 7 are described.

(4) Judgment Step The number of features used for judgment in Example 8 is 500, and the average value for each angle of the L value of the image above the threshold value, the ratio between the angles of the maximum value of the a value of the image below the threshold value, and the threshold value. It is a combination of the standard deviation for each angle of the b value of the less than image and the first quartile for each angle of the a value of the image above the threshold (feature amount combination No. 2-8).

<Example 9>
Only the parts that are different from those of Example 8 are described.

(4) Judgment Step The number of features used for judgment in Example 9 is 800, and the ratio between the minimum value for each angle of the a value of the image below the threshold and the maximum value of the a value of the image below the threshold, as a whole. A combination of the ratio between the angles of the first quartile of the R value of the image, the third quartile for each angle of the a value of the image above the threshold, and the ratio of the standard deviation of the a value of the image above the threshold. Yes (feature amount combination No. 2-9).

<Example 10>
Only the parts that are different from those of Example 9 are described.

(3) Extraction Step In Example 9, a clustering process is performed in the extraction step to classify into 20 clusters. The HSV value of the cluster having the maximum average saturation of the pixels included in each cluster is selected.

(4) Judgment Step The number of features used for judgment in Example 10 is 660, which is the ratio between the angles of the maximum a values of images below the threshold and the first quartile of the R value of images above the threshold. The ratio between angles, the minimum value of the a value of the image below the threshold for each angle, the maximum value of the H (hue) value of the entire image for each imaging angle, and the images above the threshold are clustered into 20 clusters by the Ward method to each cluster. It is a combination of the HSV value of the cluster in which the average value of the saturation of the included pixels is the maximum (feature amount combination No. 2-10).

<Example 11>
Only the parts that are different from those of Example 9 are described.

(3) Extraction Step In Example 11, a high-luminance image extraction process is performed in the extraction step. The pixel and / or pixel set closest to the primary color having the maximum hue angle of 270 degree saturation is acquired, and the image feature amount is extracted from the acquired pixel and / or pixel set.

(4) Judgment Step The number of features used for judgment in Example 11 is 760, which is the ratio between the angles of the maximum a values of images below the threshold and the first quartile of the R value of images above the threshold. It is a combination of the ratio between the angles, the minimum value for each angle of the a value of the image above the threshold value, and the ratio between the angles of the maximum H (hue) value of the entire image (feature amount combination No. 2-11).

<Comparative example 1>
(1) Imaging process
In Comparative Example 1, the reflected image is an image obtained by capturing the coating film F to be determined using an electron microscope. The reflected image obtained using an electron microscope has an actual visual field size of less than ^{0.01 cm 2.}

<Comparative example 2>
Only the parts that are different from those of Example 2 will be described.

The trained model M used in Comparative Example 2 is composed of a convolutional neural network (CNN) (machine learning model No. 1-2). The trained model M of Comparative Example 2 composed of a convolutional neural network includes a convolutional layer and a pooling layer for extracting image features from a reflected image.

Tables 1 and 2 show the details and determination results of Examples 1 to 11 and Comparative Examples 1 to 2.
Comparing Example 1 with Examples 2 to 11, the correct answer probability is higher in Examples 2 to 11. It has been shown that it is possible to determine the brilliant pigment contained in the coating film with higher accuracy by capturing the reflecting surface of the coating film F having a plurality of different aspecular angles to create a reflected image.

Comparing Example 2 and Example 3, the probability of correct answer is higher in Example 3. Further, when comparing Example 4 and Example 5, the probability of correct answer is higher in Example 5. It has been shown that the brilliant pigment contained in the coating film can be determined with higher accuracy by performing the threshold processing.

Comparing Example 2 and Example 4, the probability of correct answer is higher in Example 4. It has been shown that it is easier to set the optimum image feature amount and the bright pigment contained in the coating film can be determined with higher accuracy when the image conversion process is performed.

Comparing Example 10 with Examples 5 to 9, Example 10 has a higher probability of correct answer. It has been shown that the brilliant pigment contained in the coating film can be determined with higher accuracy by performing the clustering treatment.

Comparing Example 11 with Examples 5 to 9, the probability of correct answer is higher in Example 11. It has been shown that the bright pigment contained in the coating film can be determined with higher accuracy by performing the high-luminance image extraction process.

In Comparative Example 1, the actual field size of the reflected image acquired by using an electron microscope was ^{less than 0.01 cm 2} , and it was not possible to extract a sufficient amount of image features for determination in the extraction step.

Comparing Comparative Example 2 with Examples 1 to 11, the correct answer probability is higher in Examples 1 to 11. In Comparative Example 2, the extraction of the image feature amount is performed in the convolutional layer or the pooling layer of the convolutional neural network. It is considered that the convolutional neural network is good at edge detection of images and not good at extracting image features of bright pigments observed as points.

Next, Examples 12 to 13 related to the third embodiment will be described.
<Example 12>
(1) Imaging process
Similar to Example 1, as the coating film F to be judged for the bright material pigment determination, a solid coating color and a metallic coating color were selected from the existing toning-designed automobile top coating color reserve coating plate.

The image field of view of the image input unit 1C used is 30 mm × 5 mm, and the actual field of view size is 0.01 cm ² or more. The image input unit 1C images the reflective surface of the coating film F having an aspecular angle of about 15 degrees and 45 degrees. The image input unit 1C saves the captured reflection image as a grayscale image.

(2) Image conversion step In Example 12, an image conversion step is carried out, and the grayscale image is converted into an L image. In addition, FFT (Fast Fourier Transform) for grayscale images is also performed.

(3) Extraction Step In Example 12, the selection of the existence range E by the threshold processing is performed, and the image feature amount is extracted from the existence range E of the bright pigment. Further, in Example 12, a high-luminance image extraction process is performed in the extraction step. The pixel and / or pixel set closest to the primary color having the maximum hue angle of 270 degree saturation is acquired, and the image feature amount is extracted from the acquired pixel and / or pixel set. In Example 12, the measurement of the spectral reflectance and the extraction of the spectral reflection feature amount from the spectral reflectance are not carried out.

(4) Judgment Step The number of feature amounts used for judgment in Example 12 is 248, which is a combination of feature amounts consisting of (A) and (B) shown below (feature amount combination No. 2-12). ..
(A) Mean value, standard deviation, distortion, sharpness, maximum value of L value for the whole image of the reflected image taken at 15 degrees and 45 degrees, the image extracted only in the bright part, and the image extracted only in the dark part. , 95th percentile, 90th percentile, 75th percentile, 50th percentile, 25th percentile, 10th percentile, 5th percentile, minimum value, difference between maximum and minimum values, difference between 95th percentile and 5th percentile, and reflection image taken at 15 degrees. Difference and ratio of the above feature amount calculated from the reflected image taken at 45 degrees
(B) The wavelength of the power spectrum obtained by Fourier transforming the L values of the entire reflected image captured at 15 degrees and 45 degrees, the image extracted only in the bright part, and the image extracted only in the dark part, with a wavelength of 1 to 100 pixels. , 100-400 pixels, 400-1000 pixels, 1000-4000 pixels total value

Further, the trained model M used in Example 12 is composed of a decision tree (Gradient Boosting Decision Tree) using gradient boosting (machine learning model No. 1-3).

<Example 13>
Only the parts that differ from Example 12 are described.

(5) Measurement Step In Example 13, the image input unit 1C measures the spectral reflectance at the colorimetric angle corresponding to the imaging angle at which the reflecting surface is imaged.

(3) Extraction Step In Example 13, the existence range E is selected by the threshold processing, and the image feature amount is extracted from the existence range E of the bright pigment. Further, in Example 13, a high-luminance image extraction process is performed in the extraction step. The pixel and / or pixel set closest to the primary color having the maximum hue angle of 270 degree saturation is acquired, and the image feature amount is extracted from the acquired pixel and / or pixel set. Further, in Example 13, the spectral reflection feature amount is extracted from the measured spectral reflectance. The extracted spectral reflection features include features related to the reflectance of the spectrum.

The number of features used for the determination in Example 13 is 403, which is a combination of features (A) and (B) shown in Example 12 and (C) shown below (feature combination). No. 2-13).
(C) Spectral reflectance of 400 to 700 nm with angles from specular reflection of 15 degrees, 25 degrees, 45 degrees, 75 degrees, and 110 degrees with respect to 45 degrees incident light with specular reflection light as 0 degrees.

Table 3 shows the details of Examples 12 to 13 and the determination results.
Comparing Comparative Example 2 and Example 12, the probability of correct answer is higher in Example 12. By performing threshold processing and high-brightness image extraction processing on the grayscale image, the brilliant pigment contained in the coating film can be used as compared with Comparative Example 2 in which the convolutional neural network is used to extract the image feature amount of the RGB image. It has been shown that it can be judged with high accuracy.

Comparing Examples 1 to 2 with Example 12, the probability of correct answer is higher in Example 12 even though the grayscale image is used. It has been shown that even when a grayscale image is used, the brilliant pigment contained in the coating film can be determined with high accuracy by performing the threshold value processing and the high-luminance image extraction processing.

Comparing Example 12 and Example 13, the probability of correct answer is higher in Example 13. It has been shown that even when a grayscale image is used, the spectroscopic reflection feature amount related to the reflectance of the spectrum can be used in the extraction step to determine the bright pigment contained in the coating film with higher accuracy. ing.

The present invention can be applied to an apparatus for evaluating a painted coating film or the like.

100, 100B, 100C Bright pigment judgment device 1, 1B, 1C Image input unit 11 Light source 12 Image sensor 2 Computer 51 Sorting unit 52 Extraction unit 53 Judgment unit 3 Output unit

Claims (21)

It is a method of determining the bright pigment from the object to be judged containing the bright pigment.
An imaging process in which the reflective surface to be determined is imaged by an image sensor, and
An extraction process that extracts image features from the captured reflection image,
A determination step of determining the type and / or content of the brilliant pigment based on the image feature amount using a learned model in which the relationship between the image feature amount and the brilliant pigment is learned by machine learning.
With
The actual visual field size of the reflected image is 0.01 cm ² or more.
Glitter pigment determination method. The reflected image from which the image feature amount is extracted is an image containing the plurality of the glittering pigments.
The method for determining a bright pigment according to claim 1 . In the imaging step, the reflecting surface is imaged by the image sensor mounted on the digital camera.
The method for determining a bright pigment according to claim 1 or 2. The reflection image is an image obtained by capturing a reflection surface having a different horns.
The method for determining a bright pigment according to claim 1 or 2. The image feature amount is extracted from an RGB image, a monochrome image, a grayscale image, an HSV image, or a Lab image, or an image obtained by subjecting these images to Laplace transform, wavelet transform, or convolution filter processing.
The method for determining a bright pigment according to any one of claims 1 to 4. The existing range of the bright pigment is extracted by threshold processing based on the brightness difference, and the image feature amount is extracted from the pixel set of the existing range of the bright pigment.
The method for determining a bright pigment according to any one of claims 1 to 5. Among the pixel sets, the RGB values, brightness, saturation, hue angle, and Lab of the pixels or pixel sets acquired from a specific range specified within the 10th percentile from the minimum distance from the specific RGB, HSV, and Lab coordinates. A value, at least one color value selected from the RGB transform values, and
At least one statistic selected from the mean, median, mode, maximum, minimum, quartile, percentile, variance, standard deviation, skewness, and kurtosis of the color values.
At least one of them is used as the image feature amount.
The method for determining a bright pigment according to claim 6. Clustering of the pixel set is performed using at least one of the RGB values, brightness, saturation, hue angle, Lab value, and Fourier transform value of the pixel set.
The clusters sorted by the clustering are ranked according to any of RGB value, brightness, saturation, hue angle, Lab value, Fourier transform value, and distance from a specific color coordinate.
RGB value, brightness, saturation, hue angle, Lab value, average value of Fourier transform value, median value, mode value, maximum value, minimum value, quartile, percentile, variance, standard deviation, skewness of the cluster. , At least one of the kurtosis is used as the image feature amount,
The method for determining a bright pigment according to claim 6. A measurement step of measuring the spectral reflectance of the reflective surface to be determined with a colorimeter or a spectrophotometer , and a measurement step.
With more
Based on the image feature amount and the spectral reflectance corresponding to the image feature amount, the type and / or content of the bright pigment is learned, and the relationship between the image feature amount, the spectral reflectance and the bright pigment is learned by machine learning. Judgment process to judge using the trained model
To prepare
The method for determining a bright pigment according to any one of claims 1 to 8. A measurement step of measuring the spectral reflectance of the reflective surface to be determined with a colorimeter or a spectrophotometer , and a measurement step.
A step of extracting the spectral reflection feature amount from the measured spectral reflectance, and
With more
Machine learning of the type and / or content of the bright pigment, and the relationship between the image feature amount, the spectral reflection feature amount, and the bright pigment, based on the image feature amount and the corresponding spectral reflection feature amount. Judgment process to judge using the trained model learned by
To prepare
The method for determining a bright pigment according to any one of claims 1 to 8. The image feature amount is extracted from a monochrome image, a grayscale image, or an image obtained by subjecting these images to Laplace transform, wavelet transform, or convolution filter processing.
The method for determining a bright pigment according to claim 9 or 10. The trained model is used in linear regression, logistic regression, simple perceptron, MLP, neural network, support vector machine, decision tree analysis, random forest, Gaussian process, Bayesian network, k-nearest neighbor method, gradient boosting and other machine learning. Generated by one of the models,
The method for determining a bright pigment according to any one of claims 1 to 11. The object to be judged is a coating film.
The method for determining a bright pigment according to any one of claims 1 to 12. The glitter pigment is a vapor-deposited metal flake pigment, an aluminum flake pigment, or a photocoherent pigment.
The method for determining a bright pigment according to any one of claims 1 to 13. The determination step determines the type and content of the bright pigment.
The method for determining a bright pigment according to any one of claims 1 to 14. An image input unit that captures the reflective surface to be judged with an image sensor,
An extraction unit that extracts image features from the captured reflection image,
A determination unit that determines the type and / or content of the brilliant pigment based on the image feature amount using a learned model in which the relationship between the image feature amount and the brilliant pigment is learned by machine learning.
An output unit that displays the determination result of the determination unit, and
With
The actual visual field size of the reflected image is 0.01 cm ² or more.
Bright pigment determination device. The reflected image from which the image feature amount is extracted is an image containing the plurality of the glittering pigments.
The glittering pigment determination apparatus according to claim 16. The image input unit measures the spectral reflectance of the reflective surface of the object to be determined with a colorimeter or a spectrophotometer.
The extraction unit extracts the spectral reflection feature amount from the measured spectral reflectance, and then extracts the spectral reflection feature amount.
Based on the image feature amount and the spectral reflection feature amount corresponding to the image feature amount, the determination unit determines the type and / or content of the bright pigment, the image feature amount, the spectral reflection feature amount, and the bright pigment. Judgment of the relationship using a trained model learned by machine learning,
The glittering pigment determination apparatus according to claim 16 or 17. An image input unit having an image sensor having an actual field of view size of 0.01 cm ^{2 or more,}
Extractor and
Judgment part and
It is a program for controlling a bright pigment determination device provided with
The image input unit is made to image the reflection surface to be determined by the image pickup device.
The extraction unit is made to extract an image feature amount from the captured reflection image.
The determination unit is made to determine the type and / or content of the brilliant pigment based on the image feature amount by using a learned model in which the relationship between the image feature amount and the brilliant pigment is learned by machine learning. ,
Bright pigment judgment program. The reflected image from which the image feature amount is extracted is an image containing the plurality of the glittering pigments.
The bright pigment determination program according to claim 19. The image input unit is made to measure the spectral reflectance of the reflective surface to be determined with a colorimeter or a spectrophotometer.
The extraction unit is made to extract the spectral reflection feature amount from the measured spectral reflectance.
Based on the image feature amount and the spectral reflection feature amount corresponding to the determination unit, the type and / or content of the brilliant pigment is determined by the image feature amount, the spectral reflection feature amount, and the brilliant pigment. To determine the relationship using a trained model learned by machine learning,
The bright pigment determination program according to claim 19 or 20.

Images