JP5063076B2 - IDENTIFICATION METHOD, IDENTIFICATION SYSTEM, IDENTIFICATION PROGRAM, AND RECORDING MEDIUM THEREOF - Google Patents

Description

  The present invention is a method for identifying an arbitrary bright pigment by creating a database of image feature amounts of a bright pigment constituting a colored layer in a metallic / pearl-based coating or printing imparting a special design effect to an industrial product, The present invention relates to a system, a program, and a recording medium thereof.

  When repair coating is performed on the outer panel of an automobile, the repaired portion needs to have the same color as the other portions. However, the composition of the paint used at the time of manufacturing the new car is not disclosed, and it may be necessary to analyze the composition of the paint color.

  As a method for analyzing the composition of paint colors whose composition is unknown, a technique of computer toning (computer color matching, hereinafter referred to as “CCM”) for predicting the composition from data obtained by performing color measurement has been developed. It has been put into practical use.

  CCM is a method for predicting the color materials (color pigments and dyes) used for solid paint colors (paint colors whose appearance does not change depending on the viewing angle) from spectral reflectance data. It uses a combination of Munch's two-beam theory, Duncan's color mixing theory, and Saunderson's surface reflectance correction theory.

Further, in Patent Document 1, feature parameters of a glitter material used for a paint color whose composition is unknown are visually extracted, an image database is searched using the extracted feature parameters as a keyword, and the closest single item glitter material is searched. A method is disclosed in which an image is displayed on a monitor, and an image of an unknown paint color is compared with an image of a single article bright material to identify the type of the bright material used for the unknown paint color.
JP 2001-265786 A Japanese Patent Application No. 2005-58753 Supervised by Mikio Takagi and Yoshihisa Shimoda, “New Image Processing Handbook”, first edition, published by the University of Tokyo, September 10, 2004.

  However, a paint color CCM containing a luster pigment (metal powder such as aluminum flake, mica flake, plate-like iron oxide, etc.) has not been put into practical use yet. The reason is that, in the highlight part near the specular reflection light, the additivity between the light reflected from the glitter pigment (reflected light and interference light) is established, while the light absorption by the color pigment also occurs at the same time. On the other hand, in the shade part that is not affected by the specularly reflected light, light absorption caused mainly by the bright pigment and the colored pigment causes color development. That is, it is impossible to theorize the coloring characteristics of the color material contained in the coating film containing the bright pigment in a wide range from the highlight portion to the shade portion.

  Therefore, conventionally, in order to identify the paint pigments coated on industrial products, visual observation or observation using a microscope is performed, and identification is performed based on the characteristics of the glitter pigment stored by the observer. Things were going on. Therefore, in order to identify correctly, the observer needs experience and skill, and the identification result depends on the individual ability.

In recent years, new types of glitter pigments have been developed and put on the market. Even only aluminum flake pigment in the flaky metallic pigments, which precipitated the iron oxide surface with a or the CVD method that a surface coated with a organic pigments, those made by vapor deposition, a sheet ruber Dollar Type More than 100 types have been put on the market when grades of particle size are added in addition to manufacturing methods and color development techniques, such as those with a smooth surface. Further, there are more than 300 types of bright pigments other than aluminum flakes. It is very difficult for an observer to memorize the characteristics of all these bright pigments.

  Further, in Patent Document 1, only the particle size and color development characteristics of the glitter pigment are used as the characteristic parameters, and the type of the glitter pigment can be roughly determined, but the brand of the glitter pigment is specified. I can't.

  Therefore, the present invention has been researched to solve the above-mentioned problems of the prior art, and information such as image feature amounts and brands of glittering pigments imparting special design effects to industrial products. An identification method, an identification system for identifying an arbitrary bright pigment by creating an associated database in advance and checking the database based on an image feature amount extracted from an image obtained by imaging an arbitrary bright pigment, An object is to provide an identification program and a recording medium thereof.

In order to solve the above-described problems, a glitter pigment identification method (1) according to the present invention includes a first step of capturing image data by capturing a glitter pigment, and background processing for the acquired image data. A second step of extracting the image data of the vicinity including the glitter pigment particles as processing target image data, a third step of extracting an image feature amount from the processing target image data, and a plurality of steps The first step to the third step are performed for each of the types of glitter pigments, and a database in which information about the glitter pigments and the extracted image feature quantities are stored in correspondence with each other is created. And a fifth step of extracting the image feature amount of the bright pigment of the identification target by performing the processes of the first step to the third step for the identification target. And a sixth step of identifying the photoluminescent pigment of the identification target from the database based on the image feature of the photoluminescent pigment of the extracted photoluminescent pigment of the identification target, and the image feature but the bright feature quantity that represents the color of the pigment, and saw including a feature value representing the surface condition, the feature representing the color of the bright pigment is representative colors sine and cosine values of the hue angle, as well as the The brightness and saturation of the representative color, and the hue angle of the representative color is the mode of the histogram of the hue angle in the HSI color space of the pixel in the partial image corresponding to one glitter pigment particle, The lightness of the representative color is an average lightness of pixels having the hue angle of the representative color, and the saturation of the representative color is an average saturation of pixels having the hue angle of the representative color . .

The glitter pigment identification method ( 2 ) according to the present invention is the above-described glitter pigment identification method ( 1 ), wherein the feature amount representing the surface state of the glitter pigment includes the number of constituent colors, brightness average gray level, and chroma. The average gray level, and the number of constituent colors is the number of colors that occupy a predetermined ratio or more with respect to the partial image corresponding to the one subtractive glitter pigment particle, and the lightness average gray level is It is an average value of pixel data in an image obtained by converting a partial image corresponding to one glitter pigment particle into a light grayscale image and performing edge extraction processing of the grayscale image, and In the image obtained by converting a partial image corresponding to one glitter pigment particle with a gray scale average gray level into a grayscale image of saturation and performing edge extraction processing of the grayscale image It is the average value of the pixel data.

The glitter pigment identification method ( 3 ) according to the present invention is the feature pigment in any one of the glitter pigment identification methods ( 1 ) to ( 2 ) described above, wherein the image feature amount represents a particle shape of the glitter pigment. Is further included.

The glitter pigment identification method ( 4 ) according to the present invention is the same as the glitter pigment identification method ( 3 ) described above, wherein the feature amount representing the particle shape of the glitter pigment has a particle size, a circularity, a contour state, and a notch. In the two-dimensional contour profile obtained by plotting the distance from the center of gravity of the partial image corresponding to one glitter pigment particle to the contour pixel in order along the contour. And the number of notches is the number of deep valleys in the two-dimensional contour profile.

The glitter pigment identification method ( 5 ) according to the present invention outputs information on the glitter pigment using the image feature amount as an input unit in any of the glitter pigment identification methods (1) to ( 4 ) described above. It is characterized in that the bright pigment to be identified is identified using a neural network as a unit.

The bright pigment identification method ( 6 ) according to the present invention is the above-described bright pigment identification method ( 5 ), wherein the output value of the output unit is a real value, and the output value of the output unit depends on the output value of the output unit. A predetermined number of glitter pigments are selected from the database.

The glitter pigment identification system (1) according to the present invention includes an image acquisition device that captures an image of a glitter pigment and acquires image data, and performs background processing on the image data input from the image acquisition device. After extracting an image of a portion obtained by imaging one of the glitter pigment particles, various image feature amounts of the glitter pigment while performing image processing on the extracted partial image representing the glitter pigment particle. A feature amount detection unit that calculates the image, and a database that stores information relating to the bright pigment and the image feature amount calculated by the feature amount detection unit in association with each other for each of a plurality of types of the bright pigments. and a recording unit for storing the image feature amount, the feature amount representing the color of the bright pigment, and includes a feature value representing the surface condition, the feature representing the color of the bright pigment is, A sine value and a cosine value of the hue angle of the color and the brightness and saturation of the representative color, and the hue angle of the representative color is an HSI table of pixels in the partial image corresponding to the one glitter pigment particle. It is the mode of the histogram of the hue angle in the color space, the brightness of the representative color is the average brightness of the pixels having the hue angle of the representative color, and the saturation of the representative color is the hue angle of the representative color The feature amount detection unit obtains the image feature amount of the identification target glitter pigment from the image data of the identification target pigment input from the image acquisition device. Based on the calculated image feature amount of the identification target glitter pigment, the identification target glitter pigment is identified from the database.

Further, the glitter pigment identification program (1) according to the present invention performs background processing on the image data of the glitter pigment inputted to the computer, and the vicinity of the bright pigment including one particle of the glitter pigment. A first function for extracting image data as processing target image data, a second function for extracting an image feature amount from the processing target image data, the first function for each of a plurality of types of the glitter pigments, and A third function that performs processing according to the second function and creates a database that stores information relating to the glitter pigment and the extracted image feature amount in correspondence with each other, and the first target for the identification target A fourth function for performing processing according to the function and the second function, and extracting an image feature amount of the identification target glitter pigment; and a glitter of the extracted identification pigment Based on the image feature amount of the pigment, a fifth function for identifying the bright pigment to be identified from the database is realized, the image feature amount is a feature amount representing the color of the bright pigment , and viewed including the feature value representing the surface condition, the feature representing the color of the bright pigment is, sine and cosine values of the hue angle of the representative colors, as well as a lightness and saturation of the representative color, the representative color Is the mode value of the histogram of hue angles in the HSI color space of the pixels in the partial image corresponding to one glitter pigment particle, and the brightness of the representative color is the hue angle of the representative color And the saturation of the representative color is the average saturation of the pixel having the hue angle of the representative color .

The computer readable recording medium according to the present invention (1) is characterized in that to record the bright pigment identification program (1).

  According to the present invention, it is possible to automatically and easily identify an unknown glitter pigment from the database of brand information and image feature quantity of the glitter pigment prepared in advance based on the image feature quantity.

  In addition, when identifying using a method such as a neural network, it is possible to select a small number of approximate bright pigments in the upper group that have a high degree of coincidence as the result of automatic identification calculation. An engineer can confirm an image from a candidate while identifying an unknown bright pigment accurately and efficiently. This can also improve the accuracy of identification over complete automatic identification.

  In addition, image features such as hues of representative colors calculated from images of glittering pigments, the number of constituent colors in the particles, and the number of cuts match the human sense, and the images of glittering pigments can be grasped intuitively. Therefore, it is very effective as reference information for identifying an approximate bright pigment.

  The present invention not only identifies the luster pigment itself, but also includes a dispersion paste in which the luster pigment is suspended in a solvent, a paint containing the luster pigment, a coating film with a metallic paint, a film, a printed matter, and luster. It is also possible to identify plastics and cosmetics in which pigments are kneaded. In addition, it is useful for quality control of purchased glitter pigments, and furthermore, by analyzing the paint film pieces and identifying the paint, it is possible to narrow down the cars that were painted with that paint, so in the criminal investigation field Can also be applied.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

  The glitter pigment which is the subject of the present invention means a pigment used for developing a metallic color. As described above, the metallic color means a color whose appearance changes depending on the observation angle.

  Specific examples of the bright pigment include scaly metal pigments such as aluminum, copper, nickel alloy, and stainless steel, scaly metal pigments whose surfaces are coated with metal oxides, and chemisorption or bonding of colored pigments to the surface. Scaly metal pigments, scaly aluminum pigments with an aluminum oxide layer formed by causing an oxidation reaction on the surface, white mica pigments or interference mica pigments coated with a metal oxide such as titanium dioxide, and colored pigments on the surface Colored mica pigments that have been chemically adsorbed or coated with a metal oxide such as iron oxide, reduced mica pigments obtained by reducing interference mica pigments, and interference graphite pigments that have a graphite surface coated with metal oxides such as titanium dioxide , Silica flake or alumina flake pigment coated with metal oxide such as titanium dioxide, plate-like iron oxide face , Mention may be made of glass flake pigment whose surface is coated with a metal or metal oxide, hologram pigments, cholesteric liquid crystal polymer pigment, the bismuth oxychloride pigment as an example.

  In addition, the object of identification in the present invention is the glitter pigment itself, a dispersion paste obtained by suspending the glitter pigment in a solvent, a paint containing the glitter pigment, a coating film with a metallic color, Inks, films, printed materials, plastics with bright pigments, cosmetics, and the like are included.

  FIG. 1 is a block diagram showing a schematic configuration of a glitter pigment identification system according to an embodiment of the present invention. The glitter pigment identification system includes a glitter pigment identification device (hereinafter also referred to as an identification device) 1, an image acquisition device 2, and a display device 3. Further, the glitter pigment identification apparatus 1 controls each part and executes a predetermined process to be described later, a memory 12, and image feature quantities such as colors and particle sizes of a plurality of kinds of glitter pigments and respective brands. A recording unit 13 that stores a database including data that associates information, an operation unit 14 that receives external instructions, and an interface unit (hereinafter referred to as an I / F unit) that serves as an interface between the operation unit 14 and external devices 15) and a data bus 16 for transmitting data between the respective units. The glitter pigment identification device 1 acquires the image data captured by the image input device 2 via the I / F unit 15 and causes the display device 3 to display the image data and the like.

  As the image input device 2, for example, an electron microscope, a device in which a digital camera is connected to an optical microscope, a video microscope, or the like can be used.

  The CPU 11 temporarily stores the image data of the glitter pigment transmitted from the image input device 2 in the memory 12, and performs background processing to extract a portion where one glitter pigment particle is imaged by a method described later. Further, image features such as color and particle size are extracted by further image processing, and based on the extracted image feature amounts, a bright pigment that matches or approximates from the database recorded in advance in the recording unit 13 Are determined using a neural network. It should be noted that the CPU 11 can carry out various processes by providing specific instructions on its own, and can also perform predetermined processes such as image processing by installing specific software.

  The display device 3 is an image display device capable of full color display, for example, and the CPU 11 displays the identification result of the glitter pigment in a predetermined format on the display device 3 via the I / F unit 15, Information obtained in the processing stage can be displayed.

  The outline of the glitter pigment identification system according to the present embodiment will be described as follows. First, an image of a bright pigment is captured using the image input device 2 and recorded in the memory 12, and the recorded image is subjected to image processing to calculate an image feature amount. Next, using the calculated image feature amount, a bright pigment whose brand is known from the database is predicted by a neural network. Finally, the predicted brand and product number of the glitter pigment are displayed on the display device 3.

  Hereinafter, database creation will be described first with reference to FIG.

  FIG. 2 is a flowchart showing a database creation procedure.

  In the following description, the processing performed by the CPU 11 will be described unless otherwise specified.

  As shown in FIG. 2, in the first stage (step S1) of creating a database, the image acquisition device 2 is controlled to apply a paint containing a bright pigment, specifically, a bright pigment painted on the surface of a plate or the like. A film (hereinafter referred to as a coating film) is imaged, and the obtained image data is captured and stored in the memory 12 of the glitter pigment identification apparatus 1. In order to extract an image feature amount, it is preferable to use a high-magnification microscope as the image acquisition device 2 and capture an image in a range of 100 to 3000 magnifications. For database creation, it is preferable to image with a uniform magnification. However, the particle size of the glitter pigment is not always the same, so adjust the magnification according to the type of glitter pigment to be imaged. be able to. In this case, it is necessary to simultaneously store the captured magnification and standard magnification images. Further, from the viewpoint of accuracy, it is preferable to adjust the observation magnification so that the size of the image data of one glitter pigment particle is 20 × 20 pixels or more. Furthermore, the illumination method at the time of imaging is not particularly limited, but since reflected light needs to be observed, a known epi-illumination, reflection illumination, epi-illumination horizontal mix method, or ring illumination method is desirable.

  In addition, when it is not a coating film composed only of a bright pigment, such as a metallic coating film, that is, when imaging a coating film in which a plurality of bright pigments or colored pigments are mixed, other bright pigments And images a portion that is not easily affected by color materials (color pigments, etc.). Specifically, an operator can pick up an image by visually selecting a glitter pigment present on the surface of the metallic coating film.

  Furthermore, in order to create a database, it is preferable to take an image of a coating film obtained by applying a clear paint containing one kind of glitter pigment on a black substrate and drying and curing it. This is because it is less affected by reflected light from the ground if painted on the black substrate, and it is not affected by absorption or scattering of other coloring materials unless the clear paint contains other coloring materials. It is because it is obtained. In addition, several types of color clear paints, in which transparent pigments and dyes are dispersed and mixed in advance with the clear paint, are prepared, and a paint in which one kind of glitter pigment is mixed with the color clear paint is similarly applied to the black substrate. You may image the coating film obtained by doing.

  The image captured as described above is taken into the memory 12 of the glitter pigment identification device 1 as image data. The format of the image to be captured is not limited to the RGB bitmap format, and a compressed data format such as JPEG may be adopted, and is not particularly limited.

  In the second stage of creating the database (step S2), the image data stored in the memory 12 is cut out of an image that includes at least one bright pigment particle, binarization of the image, labeling, And background processing including a series of processing including mask processing. The background processing for the image data fetched into the memory 12 is actually performed in the following flow by instructing the CPU 11 via the operation unit 14 while the operator actually observes the image on the display device 3. (See the image analysis handbook of Non-Patent Document 1).

  FIG. 3 is a diagram for explaining background processing of an image. 3 and the following FIGS. 4 and 6 to 9 show gray scale images, they are actually full color images.

As shown in FIG. 3, first, a region including one glitter pigment particle to be noticed is cut out from the image captured in step S1 (I). For example, using an image processing software, it can be selected as a range of the rectangle. In addition, the range to cut out is determined visually by the operator so that one glitter image particle becomes sufficiently large.

Next, in order to specify the portion of the bright pigment particles from the cut image (i), that is, to clarify the region and outline of the bright pigment particles, a binarization process is performed and extracted. The desired region is white (for example, data 1) and the background is black (for example, data 0) (II) (see the method described on page 1519 of the image analysis handbook of Non-Patent Document 1). The threshold value in the binarization process is set while the operator looks at the image displayed on the display device 3. Specifically, for example, a value determined by operating the operation unit 14 while comparing the original image before processing and the processed image simultaneously displayed by the operator can be set as the threshold value.

  In the binarized image (ii), the glitter pigment particle region is white. Since this white region (a set of continuous white pixels) is not usually one, labeling is performed for each white region (III) (see page p1526 of the image analysis handbook of Non-Patent Document 1). (iii) is a labeled image. In (iii), there are two lump portions a and b, which are labeled and displayed in different colors. After labeling, a white region (a in FIG. 3) occupying the largest area in the image is selected and specified as one region of glitter pigment particles. If there is a white area with a hole, labeling is performed after filling the hole.

  A mask image (iv) is created by setting the background color (black) in addition to the region of one bright pigment particle specified as described above (IV), and one region of the specified bright pigment particle Then, the original image (i) is synthesized (V) (see page 1740 of the image analysis handbook of Non-Patent Document 1), and so-called mask processing is performed. As a result, it is possible to obtain an image (v) in which only the region of one glittering pigment particle is full color and the other region (background) has a lightness / darkness level of zero. The image on which background processing has been performed in this manner is stored in the database of the recording unit 13 in association with the bright pigment brand.

  In the third stage of creating the database (step S3), the image feature amount of the image that has undergone the background processing is extracted.

  The image feature amount of the glitter pigment targeted in the present invention can be classified into three types, that is, a color that will be described in detail later, a particle shape, and a surface state. Hereinafter, properties of each type of image feature amount and how to obtain them will be described in order.

(color)
Color is important when characterizing glitter pigments. In addition, all three color attributes (hue, lightness, and saturation) can be used as image feature amounts. When a two-dimensional image is obtained by imaging the bright pigment, the color varies depending on the pixel in one bright pigment particle region (for example, c in FIG. 3). Therefore, the representative colors of the pixels in the area can be determined, and the saturation, brightness, and hue of the representative colors can be used as image feature amounts.

  In the present embodiment, the representative color is obtained by first using the colorimetric HSI that is similar to the human sense from the colorimetric RGB data used in computers and the like for all pixels in the region of the glitter pigment particles in the image. Conversion to data (hue, saturation, and brightness) is performed (see page 1897 of the image analysis handbook of Non-Patent Document 1). A conversion method from the RGB color system to the HSI color system is not particularly limited, but in this embodiment, a hexagonal pyramid color model is used. Next, as shown in FIG. 4, a hue histogram that is a distribution of the number of pixels for each hue is obtained, and the mode value is set as the hue of the representative color. Further, the average value of the saturation of the pixels having the hue of the obtained representative color is obtained and used as the saturation of the representative color.

  Further, in the HSI space data obtained by the conversion using the hexagonal pyramid color model, the hue is represented by an angle (a positive integer of 0 to 359 °) and thus becomes discontinuous between 359 ° and 0 °, which will be described later. In this embodiment, a sine value and a cosine value of a hue angle as shown in FIG. 5 are obtained and are used as image feature amounts. FIG. 5 shows an example of an interference blue pigment, the hue angle (hue) is 215 °, and the image feature amounts are cos (215 °) = − 0.57, sin (215 °) = − 0. 82.

  Further, in addition to the saturation and hue angle sine and cosine values, brightness can be used as an image feature amount. As for the lightness, the average value of the lightness of the pixels having the hue angle of the obtained representative color is obtained in the same manner as the saturation, and is used as the lightness of the representative color.

(Particle shape)
When characterizing a luster pigment, the shape of the particles is important as well as the color. As the image feature amount representing the shape of the particle, a particle size representing the size of the particle, a circularity representing the roundness, a contour state representing the smoothness of the contour, and the number of cuts representing a deep chip can be selected. Each image feature amount of these particle shape types can be obtained as follows.

  First, as shown in FIG. 6, contour tracing of an image showing one glitter pigment particle after background processing in step S <b> 2 is performed, and pixels corresponding to the contour (hereinafter referred to as contour pixels) are obtained. Next, the center of gravity of the area of the glitter pigment particles is obtained. Next, the geometric distance between the center pixel and the contour pixel is calculated for all the contour pixels and averaged, that is, the average distance between the center and the contour is calculated, and the particle size r which is the size of the particle corresponding to the radius is calculated. And The particle size r can be calculated by the following mathematical formula.

Here, (a, b) represents the coordinates of the center of gravity, (x _i , y _i ) represents the coordinates of the contour pixel, and n represents the number of contour pixels.

  In addition, the peripheral length of an image showing one glitter pigment particle is calculated by contour processing, and the circularity is calculated from this peripheral length and the calculated particle size r (No. 224 of the image analysis handbook of Non-Patent Document 1). Page). Specifically, the circumference is calculated from the particle size r corresponding to the calculated radius, and the ratio of the circumference length to the circumference length is the circularity representing the roundness of the image.

  Next, as shown in FIGS. 7A and 7B, a two-dimensional graph (FIG. 7B) of the contour profile in which the distance L from the image centroid to the contour is developed for each contour pixel is created, and the maximum point of the distance L in the contour profile is created. And a minimum point is detected and a peak and a valley are determined (refer patent document 2). The number of peaks and valleys in the contour profile is used as an index (contour state) representing the smoothness of the contour. In FIG. 7B, the horizontal axis is a contour pixel row along the contour starting from a certain contour pixel (specifically, arranged in a counterclockwise direction from the portion A in FIG. 7A), and the vertical axis is the image The distance L from the center of gravity to each contour pixel. 7A and 7B, A, B, and C indicate peaks, and a, b, and c indicate valleys.

Furthermore, if there is an extremely deep valley in the contour profile, that is, if there is a local dent or chip in the glitter pigment, this is regarded as a notch, and the notch is determined on a predetermined basis, for example, the center of gravity of adjacent peaks and valleys. the difference L ₁ of the distance from the 40 pixels or more, and the distance L ₂ between the peaks of adjacent peaks is counted by the criterion of 40 pixels or less, and the image feature amount of the number of the notches. For example, in FIGS. 7A and 7B, a is a notch.

(Surface condition of glitter pigment)
The surface condition is also important when characterizing glitter pigments. The surface state can be indirectly observed in a two-dimensional image because it is represented by a change in light and shade levels. Specifically, the surface state of the glitter pigment particles is represented by two colors: the color constituting the particles and the unevenness of the particle surface. In particular, in the case of an optically captured image, brightness and saturation appear to be different due to the unevenness of the glitter pigment base material and the manufacturing method and the difference in the thickness of the coating layer. Therefore, in this embodiment, the color non-uniformity (intra-particle) in the HSI space data from the number of different colors (number of constituent colors in the particle) in the image representing one glitter pigment particle and the change in color density. The surface state is indirectly expressed by calculating (smoothness).

  First, the “number of constituent colors in the particle” representing the surface state, that is, the number of colors in the image representing one glitter pigment particle is obtained by the following procedure and used as an image feature amount. Specifically, as shown in FIG. 8, the full-color image data was obtained by performing discoloration processing by the uniform quantization method (see page 1209 of the image analysis handbook of Non-Patent Document 1). For example, in a 27-color image, a color having an occupancy ratio of 1.0% or more with respect to the image is set as the color constituting the particle, and the total number is obtained as the number of component colors in the particle. An example is shown in FIG.

  Regarding the “smoothness within the particle” representing the surface state, since humans feel unevenness when the brightness and saturation of the particle surface in the captured image are different, in this embodiment, the region where the density level in the image changes The image feature amount is extracted by paying attention to. Specifically, as shown in FIG. 9, a gray scale image of brightness in the HSI color system is first created based on an image representing one glitter pigment particle, and then edge extraction processing or filtering (image analysis handbook) Page 1229). The average gray level of the obtained image is determined as a feature amount representing the smoothness of the surface with respect to lightness, that is, “lightness average gray level”. Similarly, with respect to saturation, a grayscale image of saturation is created based on the original image, edge extraction processing is performed, the average gray level of the obtained image is obtained, and surface smoothness related to saturation is expressed. It is determined as an image feature amount, that is, “saturation average gray level”.

If each pixel value (gray level) of the image after the edge extraction processing is g ₁ , g ₂ ,..., G _n , the average gray level G is obtained by the following equation.

Further, the edge extraction process is not particularly limited, but a known method such as a first-order differential process such as a Sobel filter may be used.

  In this way, for the glittering pigment, four color-related image feature amounts including the sine value, the cosine value, the saturation of the representative color, and the brightness of the representative color, the particle size, and the circular shape of the representative color. Four particle shape-related image features including degree, contour state, and number of cuts, and three surface shape-related image features including number of constituent colors in the particle, lightness average gray level, and saturation average gray level In total, 11 types of image feature amounts can be extracted.

  In order to create the database, the extraction of each image feature amount for the one glitter pigment particle is performed at least once, preferably 10 times or more for one kind of glitter pigment.

  In the fourth stage of creating the database (step S4), brands of multiple types of glitter pigments, product numbers, manufacturers, other various data (particle size, specific surface area, surface treatment, etc.), type (material, Input of color etc.), price, etc. is received via the operation unit 14, for example, and stored in the memory 12.

  In the fifth stage of creating the database (step S5), each image feature amount extracted from the image in step S3 is stored in the recording unit 13 in association with the brand of the glitter pigment obtained in step S4. In addition to the image feature amount, other information obtained at each stage of image processing including the image itself can be associated and stored in the recording unit 13. A known method may be used as the data association and storage method.

  Various methods including statistical calculation, neural network, and the like can be used as an identification method for identifying a brand of an unknown bright pigment using the database obtained as described above. In the present embodiment, an identification method using a hierarchical neural network in which the image feature amount is input information and the brand (or type) of the bright pigment is output information is described (image analysis of Non-Patent Document 1). (See page 1615 of the handbook).

  Therefore, in the sixth stage (step S6) of creating the database, a neural network is constructed, and learning by the error back propagation method is performed using the data relating the image feature amount and the brand stored in the database as learning data, Determine bond weight data (synaptic weight), etc.

  Specifically, the neural network can be configured with the type (number) of image feature amounts as the number of input units and the type or brand (number) of glitter pigments as the number of output units. Note that it is not necessary to use all of the feature quantities extracted above as the image feature quantities used as the input unit, and the feature quantities relating to the color are essential, but other feature quantities can be appropriately selected. Further, the addition of an image feature amount related to the particle shape increases the accuracy of identification, and the addition of an image feature amount related to the surface state further improves the accuracy.

  FIG. 10 is a diagram illustrating a configuration example of a neural network. In FIG. 10, the input layer has 10 types of normalized image features other than the lightness of the representative hue as input units, the output layer has 30 types of glitter pigments as output units, and the hidden layers Has 10 units. The normalization of the image feature amount can be performed for each feature amount using, for example, the following equation.

Xi _{, j} = (Pi _{, j- Pmin, i} ) / ( _{Pmax, i- Pmin, i} )
Here, i represents the type of feature quantity, X _{i, j} is the value of the normalized feature quantity i obtained from the image j, P _{i, j} is the value of the feature quantity i obtained from the image j, P _{max, i} is the maximum value of the feature quantity i in the learning data _, and P _{min, i} is the minimum value of the feature quantity i in the learning data.

  Here, the data that associates the normalized image feature quantity with the brand of the bright pigment is used as learning data (teacher data), but 10 or more learning data are prepared for one type of bright pigment. Furthermore, it is preferable to use an image feature amount from which those having a strong correlation are excluded in order to improve the accuracy of identification. The number of learnings is preferably 10 million times or more, but can be adjusted while confirming the convergence state according to the image feature amount used, the brand of the bright pigment, and the learning data.

  In the seventh stage of creating the database (step S7), the neural network connection load data (synaptic weight) obtained through learning as described above is used to construct the neural network configuration information (number of layers, number of units, etc.), The normalized image feature amount and the maximum and minimum values of each feature amount are stored in the recording unit 13 to complete the creation of the database.

  Next, the identification method of the glitter pigment according to the present embodiment will be described with reference to the flowchart of FIG. 11 as the operation of identifying the glitter pigment of the identification system shown in FIG.

  Below, it demonstrates with the luster pigment mix | blended in the metallic / pearl coating color to which the design was provided by the luster pigment. In the following description, the processing performed by the CPU 11 will be described unless otherwise specified.

First, in step S11 of FIG. 11, as in the case of database creation, the image input device 2, and controls the microscope For example, it captures an image of the coating film containing the bright pigment. The captured image data is recorded as, for example, a full-color JPEG image in the memory 12 via the I / F unit 15. It should be noted here that when a plurality of types of glitter pigments are blended in the coating film, only the particles of the glitter pigment to be identified are carefully observed by a person at the time of microscopic observation. Is to select.

  Next, in step S12, in the same manner as when creating the database, in response to an instruction to the CPU 11 through the operation unit 14 as appropriate, one target bright pigment particle is cut out from the full-color image recorded in step S11, and the image Background processing including binarization, labeling, and mask processing is performed to generate image data of one glitter pigment particle that has undergone background processing.

  In step S13, a plurality of image feature quantities indicating colors, particle shapes, and surface states are obtained from the background-treated glitter pigment image obtained in step S12 via the operation unit 14 in the same manner as in the database creation. In response to an instruction to the CPU 11, extraction is performed.

  In step S14, the maximum value and the minimum value of each type of image feature amount are read from the database stored in the recording unit 13, and each type of image feature amount obtained in step S13 is normalized.

  In step S15, the brand of the glitter pigment is identified based on the extracted image feature amount. Although many statistical methods can be used for the identification method, in the present embodiment, the normalized image feature amount is used as input information, the brand of the glitter pigment is used as output information, and the error back propagation method is used. Using a neural network in which bond weight data has been determined in advance using an algorithm, the brand of the glitter pigment is identified (see page 1597 of the image analysis handbook of Non-Patent Document 1). Therefore, the image feature amount normalized in step S14 is input to the input unit of the neural network, and the identification pigment is identified by the neural network using the neural network connection load data in the database stored in the recording unit 13. Do.

  In step S <b> 16, information such as the brand of the bright pigment identified by the neural network is output to the display device 3 via the I / F unit 15.

  As described above, in the identification system of the glitter pigment according to the present invention, the feature amount is extracted from the image obtained by imaging the unknown glitter pigment, and the known glitter property is extracted based on the extracted image feature amount. By using a database previously constructed using the image feature amount of the pigment, the brand of the unknown bright pigment can be automatically and easily identified by the neural network.

  In addition, although the method for identifying the glitter pigment (metallic / pearl coating color) in the coating film has been described here, the identification method of the present invention is not limited in its use, and the film, plastic or other glitter It is also possible to identify a product in which an active pigment is kneaded, a cosmetic product, or a glitter pigment itself by the method of the present invention.

  Depending on the data related to the glitter pigment registered in the database, the range of data in the database to be identified may be limited in advance, or the range of neural network output data may be limited or expanded. It is also possible to do. That is, when the approximate type of the unknown bright pigment is known, it is efficient to limit the data range in the database to be identified to the predicted range. Also, in the neural network, instead of outputting a binary judgment result such as yes or no (1 or 0), the output data range is made a real number from 0 to 1 and sorted in descending order from the highest output value. It is also possible to select a predetermined number of similar bright pigment brands at the top as candidates.

  For example, if the output unit is an approximately known type of glitter pigment, the neural network is re-learned to obtain new connection weight data (synapse weight), and the new connection weight data is used to determine the unknown brightness. The brand of the pigment can be identified. In this case, the accuracy of identification is further increased.

  Also, when the type of glitter pigment cannot be identified in the form of yes or no, the output data range of the neural network is set to a real value of 0 to 1, and the superior pigment corresponding to, for example, five output numbers is unknown. It can also be presented as a bright pigment similar to the bright pigment.

  In addition, the identification of the unknown bright pigment was performed by the identification method of the present invention, and when the obtained result was not the correct answer that the observer knew, the sample was particularly misrecognized in the database so that the correct answer was obtained. You can add samples and retrain the neural network. By performing this relearning, new combined weight data can be obtained, and the identification accuracy of the erroneously recognized sample can be improved. That is, adding correct data to the database makes subsequent identification more accurate.

  Next, an Example is given and this invention is demonstrated more concretely.

(1) Preparation of glitter pigments As shown in FIG. 12, the glitter pigments were roughly grouped, and 30 brands with different color development, particle size, and surface condition were selected from each group. Specifically, aluminum flake pigments (brands with different particle sizes), colored aluminum flake pigments (blue, green, red), colored pearl pigments (red), interference pearl pigments (interference colors are red, gold, yellow, green, blue) 30 types were selected from purple, natural mica, alumina flakes, brands with different particle sizes), silver pearl pigments (brands with different particle sizes), and multicolor pigments.

(2) Production of coating film containing glitter pigment Each of the glitter pigments selected in (1) with respect to 100 parts by mass of resin solid content of nitrocellulose-based clear paint (manufactured by Kansai Paint Co., Ltd., Aclick 2026GL) After blending to a solid content of 1 part by mass and stirring and mixing, the coating composition was adjusted by diluting to a viscosity suitable for coating. Each adjusted paint is put in No. on art paper previously painted in black. Coating was performed with 20 bar coaters to produce a coating film having a dry film thickness of about 15 μm.

(3) Imaging of bright pigment The coating film produced as described above using a video microscope as an image acquisition device is imaged with an imaging magnification of 2500 times and an epi-illumination method, and the obtained image data is identified by an identification device ( Stored in the computer's memory.

  For the same type of glitter pigment, 10 image data were acquired, and a total of 30 × 10 = 300 images were acquired.

(4) Image Background Processing, Feature Extraction, and Normalization As described in the creation of the database, the stored image is processed using an identification device, and 10 sets of image feature values (representative) 10 types of feature values other than color brightness) were extracted. Then, 30 × 10 = 300 samples of each type of extracted image feature amount were normalized.

(5) Confirmation of correlation between feature quantities As a result of calculating a correlation matrix between 10 types of image feature quantities each having 300 samples, a correlation coefficient shown in FIG. 13 was obtained. It was confirmed that not all cross-correlation coefficients were 0.9 or more, that is, the correlation between different feature quantities was low. Therefore, it can be said that these feature quantities are appropriate.

(6) Creation of database Data relating the image feature amount and the brand of the bright pigment normalized in (4) above, an image showing one bright pigment particle, the maximum value of each type of image feature amount, The minimum value was stored as a database in the memory of the identification device.

(7) Learning of the neural network For identification using the neural network, learning by the error back propagation method was performed using the data associating the normalized image feature quantity with the brand of the bright pigment as learning data. The neural network connection weight data (synaptic weight) obtained by learning was also stored in the database. As shown in FIG. 10, the configuration of the neural network at this time is 10 input units, 30 output units, 1 hidden layer, 10 hidden layer units, 1 bias input unit, and function type. Is a sigmoid function, the number of learning is 10 million, the learning rate is 0.8, and the allowable error is 0.1.

(8) Acquisition of glitter pigment image for verification of identification accuracy In order to verify the accuracy of the identification method, 2 newly for each of the glitter pigments, similar to the images obtained in (1) to (3) above. Two images were captured to give a total of 2 × 30 = 60 images.

(9) Background processing of image and extraction and normalization of feature amount After performing background processing for each of the above 60 images, the image feature amount is extracted, and the obtained image feature amount is stored in the database in (6). Normalized with the maximum and minimum values corresponding to each stored in.

(10) Identification of glitter pigment The glitter pigment was identified using the neural network configured in (7) above for the image feature value normalized in (9). As a result, the probability that the brand of the glitter pigment was correctly judged was 74.1%, and when five brands were listed as correct glitter pigment candidates, the probability that the correct pigment was included was 94.8%. there were. As described above, since the glitter pigment can be identified with very high accuracy, it can be said that the used image feature amount is very effective.

  In the above, the present invention has been described by way of specific embodiments and examples. However, the present invention is in no way limited thereto, and the glitter pigment identification device and the identification system according to the present invention are not limited to those of the present invention. Various modifications can be made without departing from the spirit of the invention. Further, the glitter pigment identification method and program according to the present invention can be variously modified without departing from the spirit of the present invention.

It is a block diagram which shows schematic structure of the luster pigment identification system which concerns on embodiment of this invention. It is a flowchart which shows the preparation procedure of a database. It is a figure explaining the background process of an image. It is a figure which shows the histogram of the hue angle in the image of an interference blue pigment. It is a figure which shows the sine value and cosine value of a hue angle. It is a figure which shows the outline tracking of an image which shows one glitter pigment particle | grain, a gravity center, and particle size. It is a figure which shows the state of the outline in the image of an interference green pigment. It is a figure which shows the outline profile of the image of the interference green pigment shown to FIG. 7A. It is a figure explaining calculation of the number of constituent colors in a particle image. It is a figure which shows preparation of a gray scale image, and its edge extraction. It is a figure which shows the structural example of a neural network. It is a flowchart which shows the identification method of the luster pigment which concerns on this Embodiment. It is a figure which shows grouping of the luster pigment in the Example of this invention. It is a figure which shows the correlation coefficient of 10 types of image feature-values obtained by the 30 types of luster pigment * 10 sample image in the Example of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Bright pigment identification apparatus 2 Image acquisition apparatus 3 Display apparatus 11 CPU
12 Memory 13 Recording unit 14 Operation unit 15 I / F unit 16 Data bus

Claims (9)

A first step of imaging the bright pigment and acquiring image data;
A second step of performing background processing on the acquired image data and extracting image data of the vicinity including one of the particles of the glitter pigment as processing target image data;
A third step of extracting an image feature amount from the processing target image data;
The processing of the first step to the third step is performed for each of the plurality of types of the luster pigment, and a database in which information about the luster pigment and the extracted image feature amount are stored in association with each other is created. The fourth step;
A fifth step of performing the processing of the first step to the third step for the bright pigment to be identified, and extracting an image feature amount of the bright pigment to be identified;
A sixth step of identifying the identification target glitter pigment from the database based on the extracted image feature amount of the identification target glitter pigment;
Wherein the image feature amount, the feature amount representing the color of the bright pigment, and saw including a feature value representing the surface condition,
The characteristic amount representing the color of the glitter pigment is the sine value and cosine value of the hue angle of the representative color, and the lightness and saturation of the representative color,
The hue angle of the representative color is the mode value of the histogram of hue angles in the HSI color space of the pixels in the partial image corresponding to the one glitter pigment particle,
The brightness of the representative color is an average brightness of pixels having the hue angle of the representative color,
The glitter pigment identification method , wherein the saturation of the representative color is an average saturation of pixels having a hue angle of the representative color . The characteristic amount representing the surface state of the glitter pigment is the number of constituent colors, lightness average gray level, and saturation average gray level,
The number of constituent colors is the number of colors that occupy a predetermined ratio or more with respect to a partial image corresponding to one glittering pigment particle that has undergone color reduction processing;
The lightness average gray level is obtained by converting a partial image corresponding to one glitter pigment particle into a lightness grayscale image, and performing edge extraction processing of the grayscale image. Average value,
Pixels in an image obtained by converting a partial image whose saturation average gray level corresponds to one glitter pigment particle into a grayscale image of saturation and performing edge extraction processing of the grayscale image The bright pigment identification method according to claim 1, which is an average value of data. Wherein the image feature amount, the bright pigment identification method according to claims 1 to 2, characterized by further comprising a feature quantity representing the particle shape of the bright pigment. The characteristic amount representing the particle shape of the glitter pigment is the particle size, the circularity, the contour state, and the number of cuts.
The contour state depends on the number of peaks and valleys in a two-dimensional contour profile obtained by plotting the distance from the center of gravity of a partial image corresponding to one glitter pigment particle to a contour pixel in order along the contour. Represented,
4. The glitter pigment identification method according to claim 3 , wherein the number of cuts is the number of deep valleys in the two-dimensional contour profile. Any the image features as an input unit, according to claim 1 to claim 4, wherein the identifying the bright pigment of the object to be identified object by using a neural network to an output unit information on the bright pigment The bright pigment identification method according to claim 1. 6. The glitter according to claim 5 , wherein the output value of the output unit is a real value, and a predetermined number of glitter pigments are selected from the database according to the output value of the output unit. Pigment identification method. An image acquisition device for acquiring image data by imaging a bright pigment;
After performing background processing on the image data input from the image acquisition device and extracting an image of a portion obtained by imaging the one glitter pigment particle, a partial image representing the one glitter pigment particle extracted. A feature amount detection unit for calculating various image feature amounts of the glitter pigment while performing image processing on
A recording unit that stores a database that stores information relating to the bright pigment and the image feature amount calculated by the feature amount detection unit in association with each other for each of the plurality of types of the bright pigment. ,
The image feature amount includes a feature amount representing a color of the glitter pigment , and a feature amount representing a surface state,
The characteristic amount representing the color of the glitter pigment is the sine value and cosine value of the hue angle of the representative color, and the lightness and saturation of the representative color,
The hue angle of the representative color is the mode value of the histogram of hue angles in the HSI color space of the pixels in the partial image corresponding to the one glitter pigment particle,
The brightness of the representative color is an average brightness of pixels having the hue angle of the representative color,
The saturation of the representative color is an average saturation of pixels having the hue angle of the representative color,
The feature amount detection unit calculates the image feature amount of the identification target glitter pigment from the image data of the identification target glitter pigment input from the image acquisition device, and calculates the calculated identification target object. A glitter pigment identification system, wherein the glitter pigment to be identified is identified from the database based on the image feature amount of the glitter pigment. On the computer,
A first function of performing background processing on the input image data of the bright pigment and extracting image data in the vicinity thereof including one particle of the bright pigment as processing target image data;
A second function of extracting an image feature amount from the processing target image data;
A database that stores the information related to the glitter pigment and the extracted image feature quantity in advance in correspondence with each other by performing processing using the first function and the second function for each of the plurality of glitter pigments. A third function to create,
A fourth function of performing processing by the first function and the second function for the identification target bright pigment and extracting an image feature amount of the identification target bright pigment;
Based on the extracted image feature amount of the identification target glitter pigment, the fifth function of identifying the identification target glitter pigment from the database,
Wherein the image feature amount, the feature amount representing the color of the bright pigment, and saw including a feature value representing the surface condition,
The characteristic amount representing the color of the glitter pigment is the sine value and cosine value of the hue angle of the representative color, and the lightness and saturation of the representative color,
The hue angle of the representative color is the mode value of the histogram of hue angles in the HSI color space of the pixels in the partial image corresponding to the one glitter pigment particle,
The brightness of the representative color is an average brightness of pixels having the hue angle of the representative color,
A bright pigment identification program in which the saturation of the representative color is an average saturation of pixels having the hue angle of the representative color . A computer-readable recording medium on which the bright pigment identification program according to claim 8 is recorded.