Method and system for image processing x-ray transmission image of macromolecular material product including metal member

Abstract

To provide a method and a system for image processing an X-ray transmission image of a macromolecular material product that can reduce noise caused by a metal member, without impairing information of a structure of the macromolecular material product in the X-ray transmission image.SOLUTION: In an adjustment step, X-ray transmission image data D of a rubber product R having a metal member M is acquired using an X-ray CT apparatus 2, and parameters are adjusted and input to an arithmetic processing unit 7 from an input unit 10. In a generation step, output image data D1 for minimizing an error relative to the image data D is generated using the arithmetic processing unit 7, and output to an output unit 11, under the condition of reducing a feature amount that indicates a prescribed feature of the image data D on the basis of the parameters. The adjustment step and the generation step are repeated to specify parameters with which a pixel value of each pixel corresponding to an object element Ra including the rubber product R in the output image data D1, is higher than a pixel value of each pixel corresponding to a space other than the object element Ra, and then, the output image data D1 generated based on the specified parameters is output to the output unit 11 as reconstituted image data D2 in which a noise element Z has been reduced.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to an image processing method and system for X-ray transmission images of polymer material products having metal members. More specifically, the present invention relates to an image processing method and system for X-ray transmission images of polymer material products having metal members that can reduce noise caused by metal members without degrading the information on the structure of the polymer material products in the X-ray transmission images.

Background Art

[0002] X-ray transmission images obtained by an X-ray CT apparatus or the like can be used in various industrial fields including the medical field because the internal state of an object can be grasped nondestructively. For example, in the medical field, various image processing methods for reducing noise in medical images have been proposed (see Patent Document 1). Since the noise sources in medical images are few and limited, as proposed in Patent Document 1, a medical image with reduced noise can be obtained by simply detecting the positions of the noise and performing predetermined image processing.

[0003] On the one hand, in rubber products such as tires, conveyor belts, and rubber hoses, metal members such as steel cords exist in a wide area and / or multiple areas. Thus, metal members are used for purposes such as reinforcement in products made of polymer materials such as rubber. In the X-ray transmission image of a polymer material product having a metal member, noise (for example, radial metal artifacts) caused by the metal member occurs. Along with this, halation occurs around the metal member, causing the polymer material product and the like in the vicinity to become darker, or noise occurs in the air portion, making the shape of the metal member and its surroundings unclear. The method proposed in Patent Document 1 cannot be simply applied to an X-ray transmission image in which various noises caused by metal members occur. Also, when image processing such as a Gaussian filter is performed on such an X-ray transmission image, the contour of the polymer material product becomes blurred and the structure of the polymer material product cannot be accurately grasped. Therefore, there is room for improvement in obtaining an X-ray transmission image that can accurately grasp the structure of the polymer material product while reducing the noise caused by the metal member without losing the information on the structure of the polymer material product in the X-ray transmission image.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] An object of the present invention is to provide an image processing method and system for an X-ray transmission image of a polymer material product having a metal member that can reduce noise caused by the metal member without losing information on the structure of the polymer material product in the X-ray transmission image.

Means for Solving the Problems

[0006] The image processing method for X-ray transmission images of a polymer material product having a metal member of the present invention for achieving the above object is an image processing method for X-ray transmission images of a polymer material product having a metal member, which acquires X-ray transmission image data of the polymer material product having a metal member, and outputs a reconstructed image with reduced noise by reconstructing the X-ray transmission image data by an arithmetic unit. In the method, an adjustment step of adjusting parameters used for reconstructing the X-ray transmission image data by the arithmetic unit, and a generation step of generating and outputting output image data that minimizes the error with the X-ray transmission image data under the condition of reducing a feature amount indicating a predetermined feature of the X-ray transmission image data based on the parameters are repeated. Thereby, parameters are specified such that the pixel value of each pixel corresponding to an object element including the polymer material product in the output image data is higher than the pixel value of each pixel corresponding to the space other than the object element, and the output image data generated based on the specified parameters is output as reconstructed image data.

[0007] An image processing system for X-ray transmission images of polymer material products having a metal member according to the present invention includes an X-ray computed tomography apparatus that acquires X-ray transmission image data of a polymer material product having a metal member, and an arithmetic unit that performs data processing for reconstructing the X-ray transmission image data and outputs a reconstructed image with reduced noise. In the image processing system for X-ray transmission images of polymer material products having a metal member, the arithmetic unit includes an input unit for inputting parameters used for reconstructing the X-ray transmission image data, an arithmetic processing unit for generating output image data that minimizes the error with the X-ray transmission image data under the condition of reducing a feature amount indicating a predetermined feature of the X-ray transmission image data based on the parameters, and an output unit for outputting the generated output image. An adjustment step of adjusting and inputting the parameters by the input unit, and a generation step of generating the output image by the arithmetic processing unit and outputting it to the output unit are repeatedly performed, and the parameters are specified such that the pixel values of the respective pixels corresponding to the object elements including the polymer material product in the output image data are higher than the pixel values of the respective pixels corresponding to the space other than the object elements, and the output image data generated based on the specified parameters is output to the output unit as reconstructed image data.

Effect of the Invention

[0008] According to the present invention, by adjusting the parameter to adjust the degree of reduction of the feature amount indicating the characteristics of the noise (metal artifact) caused by the metal member in the output image data, the feature amount indicating the characteristics of the object element including the polymer material product is relatively increased. Then, the adjustment step of adjusting the parameter and the generation step of generating and outputting the output image data are repeated, and the pixel value of each pixel corresponding to the object element including the polymer material product becomes higher than the pixel value of each pixel corresponding to the space (background) other than the object element. Identify the parameters. In the reconstructed image data generated based on the identified parameters, the noise is reduced while the boundary between the object element and the space becomes clearer. Therefore, in the X-ray transmission image, the noise is effectively reduced without significantly damaging the information on the structure of the polymer material product.

Brief Description of Drawings

[0009]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Mode for Carrying Out the Invention

[0010] Hereinafter, an image processing method and system for an X-ray transmission image of a polymer material product having a metal member of the present invention will be described based on the embodiments shown in the drawings.

[0011] Using the embodiment of the image processing system 1 illustrated in FIG. 1, reconstructed image data D2 with reduced noise caused by the metal member M is obtained without degrading the information on the structure of the rubber product R in the X-ray transmission image (X-ray transmission image data D) of the rubber product R having the metal member M. Examples of the rubber product R include a tire, a conveyor belt, and a hose formed of vulcanized rubber. The metal member M is, for example, a bead wire in the bead portion or a steel cord in the belt layer when the rubber product R is a tire, and a reinforcing cord (steel cord) or the like when the rubber product R is a conveyor belt or a hose. Further, the metal member M is a sensor (for example, a pneumatic sensor) mounted on the rubber product R. The present invention is not limited to the rubber product R having the metal member M, and targets are those in which a metal having a specific gravity of 2 or more exists in a wide area and / or a plurality of areas in a polymer material such as rubber or resin.

[0012] The image processing system 1 includes an X-ray computed tomography apparatus 2 (hereinafter referred to as the X-ray CT apparatus 2) and an arithmetic unit 6. The X-ray CT apparatus 2 acquires X-ray transmission image data D by detecting, with an X-ray detector 4, the X-rays irradiated from an X-ray generator 3 onto an object (rubber product R). The X-ray generator 3 and the X-ray detector 4 are arranged such that the object (rubber product R) can be placed therebetween. The X-ray CT apparatus 2 can adopt various known specifications.

[0013] By using this X-ray CT apparatus 2, the structure (internal structure and cross-sectional contour shape) of the rubber product R can be grasped nondestructively. However, the X-ray transmission image data D contains noise (so-called metal artifacts) caused by the metal member M embedded in the rubber product R. And the metal member M exists in a wide range and / or a plurality of ranges of the rubber product R. Therefore, the noise caused by the metal member M is effectively reduced by using this image processing system 1.

[0014] When the rubber product R is a tire, a support 5 for supporting the tire is required. In this embodiment, the support 5 has a rim 5a on which the tire R is mounted and a support body 5b that rotatably supports the rim 5a. The tire R may be mounted on the rim 5a and inflated to a specified air pressure to acquire the X-ray transmission image data D, or the X-ray transmission image data D may be acquired in a state where the tread of the inflated tire R is grounded and a load is applied. By using this support 5, the X-ray transmission image data D of the cross section of the tire R at an arbitrary position moved in the circumferential direction around the tire axis can be acquired. Since the main part of the support 5 including the rim 5a is made of metal, noise also occurs due to the metal support 5. The metal artifacts caused by the support 5 can also be effectively reduced by using the image processing system 1.

[0015] The arithmetic unit 6 performs various data processes using the stored data and the input data. In this embodiment, it performs a data process for reconstructing the X-ray transmission image data D and outputs reconstructed image data D2 with reduced noise. Various known computers can be used for the arithmetic unit 6. The arithmetic unit 6 includes an arithmetic processing unit (CPU) 7, a main storage unit (memory) 8, an auxiliary storage unit (e.g., HDD) 9, an input unit (keyboard, mouse) 10, and an output unit (display) 11. The auxiliary storage unit 9 stores the X-ray transmission image data D acquired by the X-ray CT apparatus 2 and the reconstructed image data D2 obtained by reconstructing the X-ray transmission image data D. The input unit 10 is used to adjust parameters (λ1, λ2, λ3, f) described later. The output unit 11 outputs output image data D1 and reconstructed image data D2 generated based on the X-ray transmission image data D. Note that the storage destination of the reconstructed image data D2 is not limited to the auxiliary storage unit 9 included in the arithmetic unit 6, and may be, for example, an external auxiliary storage device. The data process by the arithmetic unit 6 is not limited to being performed by the arithmetic processing unit 7, and may be configured to be performed by, for example, an image arithmetic processing unit (GPU) that causes the output unit 11 to display an image.

[0016] Next, an example of a procedure for reconstructing the original X-ray transmission image data D using the image processing system 1 to obtain reconstructed image data D2 with reduced noise caused by the metal member M without significantly damaging the information on the structure of the rubber product R will be described.

[0017] As illustrated in FIG. 2, in this procedure, first, the original X-ray transmission image data D of the rubber product R is acquired by the X-ray CT apparatus 2 (S110). Next, the adjustment step (S120) of adjusting the parameters and the generation step (S130) of generating the output image data D1 are repeated to specify the parameters (S140). Finally, the output image data D1 generated based on the specified parameters is output as the reconstructed image data D2 (S150). The details of each step (S110 to S150) will be described below.

[0018] In step (S110), X-ray CT apparatus 2 acquires X-ray transmission image data D of rubber product R. In this step (S110), rubber product R can also be rotated to acquire a plurality of consecutive X-ray transmission image data D.

[0019] In the adjustment step (S120), a parameter is adjusted by input unit 10 of arithmetic unit 6. Specifically, data processing is executed by arithmetic unit 6 to output a parameter adjustment screen to output unit 11, and the parameter is adjusted by input unit 10 on the adjustment screen.

[0020] The parameter is used in data processing for reconstructing X-ray transmission image data D in the generation step (S130) described later. By adjusting the parameter, the degree of noise reduction in output image data D1 generated in the generation step (S130) is adjusted. More specifically, the parameter indicates the degree of reducing a feature amount indicating a predetermined feature. The greater the parameter, the greater the degree of change in the feature amount indicating a predetermined feature between X-ray transmission image data D and output image data D1, and the smaller the parameter, the smaller the degree of change in the feature amount. Specific examples of the parameter will be described later.

[0021] Adjustment step (S120) is repeatedly executed together with the generation step (S130) described later. In the first iteration of the repetition, adjustment step (S120) is executed based on the original X-ray transmission image data D output to output unit 11, and in subsequent iterations, it is executed based on output image data D1 output to output unit 11.

[0022] In the generation step (S130), data processing for reconstructing X-ray transmission image data D by arithmetic unit 6 is executed based on the adjusted parameter. By this data processing, output image data D1 is generated and the generated output image data D1 is output to output unit 11. In this data processing, output image data D1 with the minimum error from X-ray transmission image data D is generated under the condition of reducing the feature amount indicating a predetermined feature of X-ray transmission image data D based on the parameter.

[0023] In the original X-ray transmission image data D, there are noise elements such as metal artifacts caused by object elements such as rubber products R and rims 5a, space (background), and metal members M. In the feature amount indicating a predetermined feature of the X-ray transmission image data D, the feature amount indicating the object element is only slightly included, and the feature amount indicating the noise element occupies most of it. Therefore, in this generation step (S130), a condition is given to reduce the feature amount indicating a predetermined feature of the X-ray transmission image data D based on a parameter. As a result, the feature amount indicating the noise element that occupies most decreases more, and the feature amount indicating the object element that is only slightly included relatively increases. That is, the generated output image data D1 approximates an ideal X-ray transmission image without noise, the noise element is reduced, and the boundary between the object element and the space becomes clear. As a result, the internal structure and cross-sectional contour shape of the object element become clearer.

[0024] The predetermined feature is, for example, the difference in pixel values between adjacent pixels, the sum of the absolute values of the pixel values of each pixel (L1 norm), and the frequency component in the power spectrum obtained by performing a Fourier transform on the image data. The predetermined feature is not limited to one, and a plurality of them can also be adopted. For example, when three features, namely the difference in pixel values between adjacent pixels, the sum of the absolute values of the pixel values of each pixel, and the frequency component in the power spectrum obtained by performing a Fourier transform on the image data, are adopted as the predetermined feature, in this generation step (S130), the feature amount indicating each feature becomes smaller based on the parameter. The predetermined feature is not limited to the exemplified features, and for example, the total number of pixels whose pixel values are non-zero (L0 norm) and the sum of the squared values of the pixel values of each pixel (L2 norm) can also be adopted. Also, as the predetermined feature, the frequency component and intensity in the power spectrum obtained by performing a wavelet transform on the image data can be adopted.

[0025] Under the above-described conditions, the data processing in this generation step (S130) also simultaneously minimizes the error between the output image data D1 and the X-ray transmission image data D. The error between image data means the difference between the pixel values of pixels at the same pixel coordinates in each image data. As the parameter increases, the influence of the condition for reducing the feature amount becomes greater, and the error between the output image data D1 and the X-ray transmission image data D becomes greater. Therefore, in the generation step (S130), by simultaneously minimizing this error, it is possible to reduce the error between the output image data D1 and the X-ray transmission image data D while suppressing the influence of the condition for reducing the feature amount. In particular, under the above-described conditions, the difference between the pixel values of pixels at the same pixel coordinates of the output image data D1 and the X-ray transmission image data D changes. Therefore, to minimize the error between image data, the change in pixel coordinates should be minimized. The minimum change in pixel coordinates means that the change in the cross-sectional contour shape and its dimensions of the rubber product R is minimized, and the information on the structure of the rubber product R in the original X-ray transmission image data D1 can be generally retained.

[0026] The details of the data processing in the generation step (S130) will be described. Hereinafter, five specific examples will be described using mathematical formulas. In the following mathematical formulas, E represents an evaluation function, and (A - A 0 ) represents the error between the output image data D1 and the X-ray transmission image data D. |∇u| represents the differential value of the pixel value of the pixel in the output image data D1. The differential value |∇u| is the difference between adjacent pixel values, and represents the sum of the vertical difference and the horizontal difference in the image data. ||u|| represents the L1 norm of the output image data D1. The L1 norm ||u|| is the sum of the absolute values of the pixel values of each pixel in the output image data D1. (u - u 0 ) represents the error between the output image data D1 and the X-ray transmission image data D in the state of Fourier transform. F represents the inverse Fourier transform. In the inverse Fourier transform F, the frequency domain obtained by sampling the frequency domain of the power spectrum obtained by Fourier-transforming the X-ray transmission image data D at the sampling ratio f is used. λ 1 , λ 2 , λ 3 each represents a weight coefficient. The weight coefficients λ 1 , λ 2, λ 3 The parameter λ and the sampling ratio f are adjusted in the above-described step (S120). All the parameters are positive numbers.

[0027] Data processing using the difference between pixel values of adjacent pixels as a predetermined feature reconstructs output image data D1 from X-ray transmission image data D by solving the optimization problem of the following mathematical formula (1). The first term on the right side of the mathematical formula (1) represents the squared error between the X-ray transmission image data D and the output image data D1. By the first term, output image data D1 that is clearly different from the X-ray transmission image data D is excluded. The second term on the right side of the mathematical formula (1) represents regularization regarding the difference between pixel values of adjacent pixels, so-called total variation regularization.

[0028]

Equation

[0029] In the X-ray transmission image data D, generally there is a strong correlation between adjacent pixels, so most of the difference values take values close to zero, but only when there is an edge around a certain pixel, large values are taken for the differences in the vertical and horizontal directions. In the X-ray transmission image data D, the edge corresponds to the contour of an object element such as the rubber product R or the contour of a noise element. By obtaining a solution under the condition of this regularization, the difference between pixel values of adjacent pixels becomes smaller than that of the X-ray transmission image data D, and the roughness of the image is reduced. That is, output image data D1 with reduced radial metal artifacts existing in the X-ray transmission image data D is obtained.

[0030] Data processing using the sum of the absolute values of the pixel values of each pixel as a predetermined feature reconstructs output image data D1 from X-ray transmission image data D by solving the optimization problem of the following mathematical formula (2). The first term on the right side of the mathematical formula (2) is the same as the first term on the right side of the above-described mathematical formula (1). The second term on the right side of the mathematical formula (1) represents regularization of the sum of the absolute values of the pixel values of each pixel of the X-ray transmission image data D, so-called L1 norm regularization.

[0031]

Number

[0032] In the X-ray transmission image data without noise, the pixel values of each pixel corresponding to the space other than the object elements become 0. On the other hand, in the X-ray transmission image data D in which noise elements exist, due to the influence of the noise elements, the pixel values of some pixels corresponding to the space become values other than 0. That is, the X-ray transmission image data D in which noise elements exist has a larger sum of the absolute values of the pixel values of each pixel than the X-ray transmission image data without noise. By obtaining a solution on the condition of regularizing this sum, the sum of the absolute values of the pixel values of each pixel becomes smaller than that of the X-ray transmission image data D, and the darkness of the image increases. That is, the output image data D1 in which the noise with small pixel values existing in the X-ray transmission image data D is reduced is obtained.

[0033] Data processing using the frequency components in the power spectrum obtained by Fourier-transforming the image data as a predetermined feature reconstructs the output image data D1 from the X-ray transmission image data D by solving the optimization problem of the following mathematical formula (3). The right side of the mathematical formula (3) is the result of performing the Fourier inverse transform F on the squared error between the output image data D1 and the X-ray transmission image data D in the Fourier-transformed state and integrating it. In the Fourier inverse transform F, the frequency region obtained by sampling the frequency region of the power spectrum obtained by Fourier-transforming the X-ray transmission image data D at the sampling ratio f is used. Therefore, the output image data D1 obtained by solving the optimization problem of the mathematical formula (3) can be regarded as the image data obtained by performing the Fourier inverse transform on the frequency region obtained by sampling the frequency region of the power spectrum obtained by Fourier-transforming the X-ray transmission image data D at the sampling ratio f. This right side localizes the frequency components in the power spectrum by filtering the frequency region using the sampling ratio f with the frequency components in the power spectrum obtained by Fourier-transforming the image data as a feature amount. Also, this right side shows the minimization of the difference between the output image data D1 and the X-ray transmission image data D at the same time.

[0034]

Number

[0035] In the power spectrum obtained by Fourier-transforming the X-ray transmission image data D, the pixel values of the pixels corresponding to the object elements and the noise elements respectively appear as the intensities (strengths) of various frequency components. In the power spectrum obtained by Fourier-transforming the image data, generally, high-frequency components indicate linear elements and local elements, while low-frequency components indicate curved elements and rough elements. That is, in the X-ray transmission image data D, the frequency components of object elements such as the rubber product R are low, and the frequency components of noise elements with high regularity such as metal artifacts are high. Therefore, by obtaining a solution under the condition of frequency filtering, the high-frequency components in the X-ray transmission image data D are removed and the low-frequency components remain. As a result, the output image data D2 with reduced high-regularity noise existing in the X-ray transmission image data D is obtained.

[0036] Data processing using, as predetermined features, the differences in pixel values between adjacent pixels and the frequency components in the power spectrum obtained by Fourier-transforming the image data reconstructs the output image data D1 from the X-ray transmission image data D by solving the optimization problem of the following mathematical formula (4). The first term on the right side of mathematical formula (4) is the second term on the right side of the above-mentioned mathematical formula (1), and the second term on the right side of mathematical formula (4) is the right side of the above-mentioned mathematical formula (3).

[0037]

Number

[0038] Data processing using, as respective predetermined features, the difference in pixel values between adjacent pixels, the sum of the absolute values of the pixel values of each pixel, and the frequency components in the power spectrum obtained by Fourier-transforming the image data reconstructs output image data D1 from X-ray transmission image data D by solving the optimization problem of the following mathematical formula (5). The first term on the right side of mathematical formula (5) is the second term on the right side of mathematical formula (1) described above, the second term on the right side of mathematical formula (5) is the second term on the right side of mathematical formula (2) described above, and the third term on the right side of mathematical formula (5) is the right side of mathematical formula (3) described above.

[0039]

Number

[0040] The optimization problems of the above-described mathematical formulas (1) to (5) can be solved using known proximal splitting optimization algorithms such as the proximal gradient method, the alternating direction method of multipliers, and the primal-dual proximal splitting method, using the parameters (weight coefficients λ1, λ2, λ3, sampling ratio f) adjusted in step (S120). Specific algorithms are, for example, the ADMM (Alternating Direction Method of Multipliers) algorithm and the Chambolle-Pock algorithm.

[0041] In the generation step (S130), when adopting a plurality of features as predetermined features, data processing for each feature can be summarized into one formula and performed simultaneously as in the above-described mathematical formula (4) or mathematical formula (5). Also, in the generation step (S130), when adopting a plurality of features, data processing for each feature can also be performed individually in a predetermined order. For example, noise reduction can be performed by individually performing the respective data processing using mathematical formula (1), mathematical formula (2), and mathematical formula (3) in order. When performing a plurality of data processings individually, the order thereof is not particularly limited.

[0042] In step (S140), the above-described steps (S120, S130) are repeated until the pixel value of each pixel corresponding to the object element including the rubber product R in the output image data D1 becomes higher than the pixel value of each pixel corresponding to the space other than the object element. Then, the parameter at the time when it becomes higher is specified.

[0043] The fact that the pixel value of each pixel corresponding to the object element becomes higher than the pixel value of each pixel corresponding to the space means that the boundary between the object element and the space becomes clear. The clarity of the boundary between the object element and the space can be arbitrarily set, but it is preferable that the cross-sectional contour shape of the rubber product R existing in the output image data D1 can be digitized. Being digitizable means that the cross-sectional contour shape of the rubber product R can be extracted by processing the output image data D1 by the arithmetic unit 6. This data processing is, for example, data processing for binarizing the output image data D1 and extracting the cross-sectional contour shape of the rubber product R by various known contour extraction methods such as contour tracing and edge detection. Note that, depending on the setting of the parameter, the cross-sectional contour shape of the metal member M existing in the output image data D1 can also be digitized.

[0044] The X-ray transmission image data D of the rubber product R in which the metal member M exists in a wide area and / or a plurality of areas inside the vulcanized rubber is the X-ray transmission image data D at the same position of the same rubber product R obtained under the same imaging conditions, but the occurrence state of metal artifacts caused by the metal member M is different. Also, in the rubber product R in which the metal member M exists in a wide area and / or a plurality of areas inside the vulcanized rubber, various metal artifacts irregularly occur everywhere in the X-ray transmission image data D. Thus, the X-ray transmission image data D of the rubber product R has a specific problem of the rubber product R that noise such as various metal artifacts irregularly occurs over the entire area of the image with respect to medical image data in which the metal member M hardly exists in the object. Therefore, in order to effectively remove the noise according to the occurrence state of the noise in the X-ray transmission image data D of the rubber product R and obtain the output image data D1 in which the boundary between the object element and the space is clearer, the above-described steps (S120, S130) are repeated, and the method of appropriately adjusting the parameters each time is the best.

[0045] In step (S150), data processing is executed to output the output image data D1 based on the parameters specified by the arithmetic unit 6 as the reconstructed image data D2. By performing data processing on a plurality of consecutive reconstructed image data D2 obtained from the same rubber product R using a known tomography method by the arithmetic unit 6, X-ray CT image data can also be obtained.

[0046] Hereinafter, the reconstructed image data D2a to D2d obtained by the above-described procedure of FIG. 2 will be described in detail using the image data Dm obtained by simulating the X-ray transmission image data D of the rubber product R having the metal member M illustrated in FIG. 3.

[0047] The image data Dm illustrated in FIG. 3 is data created by mimicking the X-ray transmission image data D. The image data Dm has an annular object element Ra arranged at the center of the image and an L-shaped noise element Z arranged at the lower right of the image. The object element Ra mimics the rubber product R that is the object in the X-ray transmission image data D. The noise element Z mimics the metal artifact in the X-ray transmission image data D.

[0048] The reconstructed image data D2a illustrated in FIG. 4 was reconstructed by solving the optimization problem of the above-mentioned formula (1) for the image data Dm of FIG. 3. The weight coefficient λ1, which is a parameter, was adjusted to 40. In the reconstructed image data D2a, the noise element Z is thinner than in the image data Dm.

[0049] The reconstructed image data D2b illustrated in FIG. 5 was reconstructed by solving the optimization problem of the above-mentioned formula (2) for the image data Dm of FIG. 3. The weight coefficient λ2, which is a parameter, was adjusted to 60. In the reconstructed image data D2b, there is no change around the edge of the object element Ra, and there is generally no change in the noise element Z with a relatively large pixel value, but the noise with a small pixel value other than the noise element Z is reduced.

[0050] The reconstructed image data D2c illustrated in FIG. 6 was reconstructed by solving the optimization problem of the above-mentioned formula (3) for the image data Dm of FIG. 3. The sampling ratio f, which is a parameter, was adjusted to 50% (0.5). In the reconstructed image data D2c, the highly regular noise element Z has disappeared.

[0051] The reconstructed image data D2d illustrated in FIG. 7 was reconstructed by solving the optimization problem of the above-mentioned formula (5) for the image data Dm of FIG. 3. The weight coefficient λ1 was adjusted to 50, the weight coefficient λ2 was adjusted to 200, the weight coefficient λ3 was adjusted to 1, and the sampling ratio f was adjusted to 30% (0.3), respectively.

[0052] The reconstructed image data D2d has pixel values of each pixel corresponding to the object element Ra that are smaller and pixel values of each pixel corresponding to the space that are larger, compared to the image data Dm. That is, as the feature amount decreases such that the pixel values of each pixel corresponding to the object element Ra approach the pixel values of each pixel corresponding to the space, the noise element Z is reduced. Also, in the reconstructed image data D2d, since the pixel values of each pixel corresponding to the object element Ra are higher than the pixel values of each pixel corresponding to the space, the edge of the object element Ra is sharp.

[0053] When comparing the respective reconstructed image data D2a to D2d shown in FIGS. 4 to 7, the boundary between the object element Ra and the space in the reconstructed image data D2d shown in FIG. 7 is the clearest, and the object element Ra is the sharpest. In the above-described generation step (S130), depending on the occurrence of noise such as metal artifacts in the X-ray transmission image data D, the above-described mathematical formulas (1) to (5) may be appropriately selected. However, it can be seen that it is more effective to adopt a plurality of features as predetermined features of the X-ray transmission image data D. Therefore, in the generation step (S130), it is preferable to execute data processing for solving an optimization problem adopting a plurality of features, and it is more preferable to execute data processing for solving the optimization problems of mathematical formulas (4) and (5).

[0054] As described above, according to this embodiment, by adjusting the parameters (weight coefficients λ1, λ2, λ3, sampling ratio f) to adjust the reduction degree of the feature amount indicating the characteristics of the noise elements in the output image data D1, the feature amount indicating the characteristics of the object elements such as the rubber product R is relatively increased. Also, by minimizing the error between the output image data D1 and the X-ray transmission image data D, the information on the structure of the object element is generally retained. Then, by repeating the adjustment of the parameters, the generation, and the output of the output image data D1, the parameters are specified such that the pixel values of the respective pixels corresponding to the object element are higher than the pixel values of the respective pixels corresponding to the space (background). In the reconstructed image data D2 based on the specified parameters, noise such as metal artifacts is effectively reduced, and the boundary between the object element and the space becomes clearer. Therefore, in the reconstructed image data D2, the metal artifacts caused by the metal member M are effectively reduced without significantly damaging the information on the structure of the rubber product R. As a result, the structure of the rubber product R can be grasped more accurately, which greatly contributes to ensuring the quality of the rubber product R and the development and improvement of the rubber product R.

[0055] Also, according to this embodiment, the reconstructed image data D2 can be data-processed by an arithmetic device 6 or the like, and the cross-sectional contour shape of the rubber product R existing in the reconstructed image data D2 can be digitized by data processing. As a result, the structure of the rubber product R can be analyzed and analyzed with high precision, which is advantageous for the development and improvement of the rubber product R.

[0056] This embodiment outputs reconstructed image data D2 of one or a plurality of limited regions in one X-ray transmission image data D. When outputting the reconstructed image data D2 of a plurality of limited regions, the adjustment step (S120) and the generation step (S130) may be repeated for each limited region to specify parameters. The plurality of limited regions are divided based on the occurrence condition of noise. The size of each limited region may be different for each limited region. When the rubber product R is a tire, for example, it is divided into regions where the bead core and the rim 5a exist, regions where the belt layer exists, and the like. Different parameters can be specified for each of the plurality of limited regions in one X-ray transmission image data D, and noise such as metal artifacts with different occurrence conditions can be more effectively reduced for each limited region.

[0057] Compared with machine learning that learns a large number of image data, the data processing in the generation step (S130) does not require the analysis of a large amount of image data and can be processed with only one X-ray transmission image data D. Therefore, the computational load on the arithmetic device 6 can be significantly reduced. In addition, this data processing does not significantly change the object elements compared with image processing using a known noise removal filter such as a Gaussian filter. Therefore, by using the reconstructed image data D2 obtained by this data processing, the internal structure and cross-sectional contour shape of the rubber product R can be accurately grasped.

[0058] All the specified parameters are positive numbers. The parameters are set to appropriate values according to the occurrence situation of noise in the X-ray transmission image data D. By repeatedly implementing this embodiment, it is also possible to set the range of parameters according to the type of the rubber product R. When the rubber product R is a tire, for example, the sampling ratio f is preferably a value within the range of 20% or more and less than 80%, and more preferably a value within the range of 30% or more and 60% or less. When using the above formula (5), the weighting coefficients λ1, λ2, and λ3 are preferably based on the ratio of the values of each term on the right side of the formula (5). For example, the ratio of the value of the first term to the value of the second term on the right side of the formula (5) and the ratio of the value of the first term to the value of the third term are each preferably more than 0 times and 1000 times or less, and more preferably more than 0 times and 100 times or less.

[0059] Next, another embodiment of an image processing method and system for an X-ray transmission image of a polymer material product having a metal member will be described. In this embodiment, the initial values of the parameters are set using the learning data LD stored in advance in the auxiliary storage unit 9 of the arithmetic unit 6.

[0060] In the procedure illustrated in FIG. 8, another step (S210, S220) is added to the procedure of FIG. 2 described above. That is, in the procedure of this embodiment, the initial values of the parameters are set by performing data processing by the arithmetic unit 6 based on the learning data LD stored in advance (S210). Then, after storing the reconstructed image data D2 (S150), the learning data LD is updated (S220). Therefore, in this embodiment, every time the X-ray transmission image data D is reconstructed, new data is added to and expands the learning data LD, and the expanded learning data LD is used for setting the initial values of the parameters in subsequent times.

[0061] The learning data LD shows the relationship between a large number of X-ray transmission image data D (samples) acquired in the past and the parameters specified when reconstructing the X-ray transmission image data D. The learning data LD is created by accumulating a large number of the data obtained by the above procedure. The learning data LD has a small number of samples in the initial stage where the number of times the above procedure is performed is small, but the number of samples increases every time the number of times the above procedure is performed increases. The learning data LD is stored in advance in the auxiliary storage unit 9 before acquiring the X-ray transmission image data D of the desired rubber product R in step (S110).

[0062] In the learning data LD, parameters identified when reconstructing a number of samples are aggregated for those samples. The learning data LD may indicate the relationship between the samples and the parameters using feature quantities indicating the features of the samples. Such features are not limited to features directly possessed by the samples such as histograms, and the features of the rubber product R that is the object of the sample can also be used. Examples of the features of the rubber product R include information (specifications) of the rubber product R, such as the type of the rubber product R like a tire, a conveyor belt, a hose, etc., the shape and dimensions of the rubber product R. Details of the learning data LD will be described later.

[0063] In step (S210), based on the learning data LD, data processing is executed by the arithmetic unit 6 to set an initial value of a parameter used for reconstructing the new X-ray transmission image data D acquired in step (S110). In this step (S210), it is only necessary to set an initial value as a rough guideline for adjusting the parameter in the adjustment step (S120), and it is only necessary to ensure that an obviously inappropriate value is not set as the parameter used for reconstructing the new X-ray transmission image data D. "Inappropriate" means that the noise existing in the new X-ray transmission image data D is not removed at all, or the information on the structure of object elements such as the rubber product R is greatly damaged. For X-ray transmission image data D with similar noise generation situations, since feature quantities such as the information of the rubber product R and the histogram are approximated, the parameters identified when reconstructing each X-ray transmission image data D are approximated. Therefore, by using the learning data LD, the initial value of the parameter used for reconstructing the new X-ray transmission image data D can be obtained from the relationship between each sample and the parameter that has been grasped in advance.

[0064] The arithmetic unit 6 executes data processing for setting the initial values of the parameters by comparing the new X-ray transmission image data D and the learning data LD. Alternatively, the arithmetic unit 6 predicts the parameters specified when reconstructing the new X-ray transmission image data D based on the prediction model constructed by machine learning using the learning data LD and the new X-ray transmission image data D, and executes data processing for setting the predicted parameters as the initial values. The initial values of the parameters set in this step (S210) are not limited to one, and there may be a plurality of cases. The details of the data processing in this step (S210) will be described later.

[0065] The initial values of the set parameters are output to the output unit 11 together with the X-ray transmission image data D in the above-described adjustment step (S120). Therefore, in the first adjustment step (S120), the parameters are adjusted using the initial values of the set parameters. Note that the initial values of the set parameters can also be used in the generation step (S130) without adjustment.

[0066] In step (S220), data processing for updating the learning data LD is executed by the arithmetic unit 6. Specifically, the arithmetic unit 6 executes data processing for adding, as a new sample, the new X-ray transmission image data D acquired in step (S110) and the parameters specified when reconstructing the new sample to the learning data LD. In this way, each time the reconstructed image data D2 is acquired, the learning data LD is expanded, which is advantageous for improving the accuracy of the initial values of the parameters set in subsequent times.

[0067] The details of setting the initial values of the parameters in step (S210) will be described below.

[0068] The learning data LDa illustrated in FIG. 9 shows the relationship between each sample and the parameters using a histogram for each sample. Specifically, for each sample (1 to n) in the leftmost column of the table in the learning data LDa, the histogram and the parameters are aggregated.

[0069] When using the learning data LDa, in the above-described step (S210), the acquired new X-ray transmission image data D is processed by the arithmetic unit 6 to extract the histogram of the new X-ray transmission image data D. Then, the extracted histogram is compared with the learning data LDa, and a sample having a histogram approximated to the histogram extracted from the learning data LDa is selected, and the parameter specified when reconstructing the selected sample is set as the initial value. Specifically, the arithmetic unit 6 compares the histogram of the new X-ray transmission image data D with each histogram of the learning data LDa to determine whether it is within the allowable range set as approximated. The allowable range can be arbitrarily set. For example, for the new X-ray transmission image data D, the Pearson correlation coefficient between the histograms is 0.2 or more.

[0070] The learning data LDb illustrated in FIG. 10 shows the relationship between each sample and the parameter using the information (specifications) of the rubber product R that is the object of each sample. More specifically, the rubber product R is a pneumatic tire, and as information on the pneumatic tire, for example, the tire size and the type of rim to be assembled are adopted. The information of the rubber product R can be arbitrarily selected and may be selected according to the shape and dimensions of the rubber product R, the specifications of the support 5, and the like. In the learning data LDb shown in FIG. 10, for each sample (1 to n) in the leftmost column, the tire size, the type of rim, and the parameter are integrated. As the tire size, in addition to the tire width [mm], the aspect ratio [%], and the rim diameter [in], a load index, a speed symbol, and the like can also be adopted. As the type of rim, the material, the structure (one-piece structure, two-piece structure, three-piece structure, etc.) can be adopted.

[0071] When using the learning data LDb, in the above-described step (S210), the information of the rubber product R is input into the arithmetic unit 6 by the input unit 10. Then, the input tire size and rim type are compared with the learning data LDb, and a sample approximated to the input tire size and rim type is selected from the learning data LDb, and the parameter specified when reconstructing the selected sample is set as the initial value. Specifically, the arithmetic unit 6 compares the information of the rubber product R, which is the object of the new X-ray transmission image data D, with the respective information of the learning data LDb, and determines whether it is within the allowable range set as approximated. The allowable range can be arbitrarily set. For example, it is about ±3% with respect to the information of the rubber product R in the new X-ray transmission image data D.

[0072] Note that the learning data LDb may be, for example, the specifications (shape, dimensions, etc.) of each rubber product R integrated for each type of rubber product R such as pneumatic tires, non-pneumatic tires, conveyor belts, and hoses. In the case of such learning data LDb, as a feature of the rubber product R, the type of the rubber product R may be input.

[0073] It is also possible to use, as the learning data LD, the data obtained by integrating the learning data LDa and LDb. In this way, by adopting both the features directly possessed by the X-ray transmission image data D and the features possessed by the rubber product R as the object, as the features of the X-ray transmission image data D, the samples of the learning data LD can be further subdivided. The subdivision of the samples is advantageous for improving the fitting accuracy of the initial value of the parameter to be set.

[0074] In the learning data LDc illustrated in FIG. 11, the parameters for each sample (1 to n) in the leftmost column are integrated. When using the learning data LDc, in the above-described step (S210), the arithmetic unit 6 executes data processing for predicting the parameter specified when reconstructing the new X-ray transmission image data D based on the new X-ray transmission image data D acquired and the prediction model constructed by machine learning using the learning data LDc, and sets the predicted parameter as the initial value of the parameter.

[0075] The prediction model is a type of computer program that is constructed by machine learning using learning data LDc and predicts parameters specified when reconstructing new X-ray transmission image data D. The prediction model is constructed by supervised machine learning, with the feature amount of each sample as the explanatory variable and each parameter as the objective variable. As the machine learning, various known machine learnings such as random forest and deep neural network (DNN) can be used.

[0076] The occurrence condition of noise such as metal artifacts in the X-ray transmission image data D also varies depending on the imaging conditions of the X-ray transmission image data D. Therefore, as the learning data LD, the imaging conditions can also be adopted as additional features.

[0077] As described above, according to this embodiment, the initial value of the parameter set based on the learning data LD is set to a value corresponding to the occurrence situation of the noise in the acquired X-ray transmission image data D. Therefore, since the parameter can be adjusted using the initial value as a reference, the labor and time required for parameter adjustment can be significantly reduced.

[0078] The present invention is not limited to a specific embodiment, and various modifications and changes are possible within the scope of the gist of the present invention.

Example

[0079] The X-ray transmission image data D illustrated in FIG. 12 shows the structure of a pneumatic tire in cross section. The rubber product R that is the object of this X-ray transmission image data D is a pneumatic tire for trucks with a tire width of 285 mm, an aspect ratio of 85%, and a rim diameter of 22.5 inches. The X-ray transmission image data D has been acquired by a known X-ray CT apparatus 2 while this tire is mounted on a metal rim 5a and inflated to a specified air pressure. In this X-ray transmission image data D, noise elements Z due to various metal artifacts caused by metal members M such as the belt layer and bead core of the tire and the metal rim 5a are irregularly generated in a wide plurality of regions. Therefore, the object element Ra of the rubber product R has become unclear.

[0080] The reconstructed image data D2 illustrated in FIG. 13 is obtained by reconstructing the X-ray transmission image data D of FIG. 12 according to the procedure illustrated in FIG. 2, and the above-described mathematical formula (5) is used in this procedure. Each parameter was such that the weighting coefficient λ1 was 2, the weighting coefficient λ2 was 3, the weighting coefficient λ3 was 0.000001, and the sampling ratio f was 60% (0.6). The continued ratio of the values of each term on the right side of the mathematical formula (5) was the value of the first term: the value of the second term: the value of the third term = 1: 2: 9.

[0081] It can be seen that the reconstructed image data D2 shown in FIG. 13 has significantly reduced noise elements Z compared to the original X-ray transmission image data D shown in FIG. 12, while the object element Ra of the rubber product R has become clear. That is, in the reconstructed image data D2, the noise caused by the metal member M is reduced without impairing the information on the structure of the rubber product R in the X-ray transmission image data D.

[0082] The present disclosure includes the following inventions. Invention 1: In an image processing method for an X-ray transmission image of a polymer material product having a metal member, wherein X-ray transmission image data of the polymer material product having the metal member is acquired, and the X-ray transmission image data is reconstructed by an arithmetic unit to output a reconstructed image with reduced noise. An adjustment step of adjusting parameters used for reconstruction of the X-ray transmission image data by the arithmetic unit, and a generation step of generating and outputting output image data that minimizes the error from the X-ray transmission image data under the condition that a feature amount indicating a predetermined feature of the X-ray transmission image data is reduced based on the parameters, are repeated, An image processing method for an X-ray transmission image of a polymer material product having a metal member, which specifies the parameters such that the pixel value of each pixel corresponding to an object element including the polymer material product in the output image data is higher than the pixel value of each pixel corresponding to a space other than the object element, and outputs the output image data generated based on the specified parameters as reconstructed image data. Invention 2: The image processing method for an X-ray transmission image of a polymer material product having a metal member according to Invention 1, wherein in the generation step, total variation regularization for reducing the difference in pixel values between adjacent pixels of the X-ray transmission image data is used, and a weight coefficient for the total variation regularization is used as the parameter. Invention 3: The image processing method for an X-ray transmission image of a polymer material product having a metal member according to Invention 1, wherein in the generation step, frequency filtering for reducing frequency components in a power spectrum obtained by Fourier transform is used, and sampling ratios before and after the frequency filtering are used as the parameters. Invention 4: The image processing method for an X-ray transmission image of a polymer material product having a metal member according to Invention 1, wherein in the generation step, total variation regularization for reducing the difference in pixel values between adjacent pixels of the X-ray transmission image data and frequency filtering for reducing frequency components in a power spectrum obtained by Fourier transform are used, and a weight coefficient for each of the total variation regularization and the frequency filtering and sampling ratios before and after the frequency filtering are used as the parameters. Invention 5: In the generation step, using total variation regularization for reducing the difference in pixel values between adjacent pixels of the X-ray transmission image data, L1 regularization for normalizing the sum of absolute values of all pixel values of the X-ray transmission image data, and frequency filtering for reducing frequency components in the power spectrum obtained by Fourier transform, as the parameters, the weight coefficients for each of the total variation regularization, the L1 regularization, and the frequency filtering, and the sampling ratios before and after the frequency filtering are used, which is an image processing method for an X-ray transmission image of a polymer material product having a metal member according to Invention 1. Invention 6: In the generation step, an image processing method for an X-ray transmission image of a polymer material product having a metal member according to Invention 5, wherein the total variation regularization, the L1 regularization, and the frequency filtering are individually performed in a predetermined order. Invention 7: In the generation step, an image processing method for an X-ray transmission image of a polymer material product having a metal member according to Invention 5, wherein the total variation regularization, the L1 regularization, and the frequency filtering are simultaneously performed using one formula. Invention 8: An image processing method for an X-ray transmission image of a polymer material product having a metal member according to any one of Inventions 1 to 7, wherein the polymer material product is a tire, and the X-ray transmission image data includes an image of at least one of a bead portion or a belt layer of the tire. Invention 9: Output the reconstructed image data of one limited region or a plurality of limited regions in one X-ray transmission image data. When outputting the reconstructed image data of a plurality of limited regions, for each of the limited regions, repeat the adjustment step and the generation step to specify the parameters, which is an image processing method for an X-ray transmission image of a polymer material product having a metal member according to any one of Inventions 1 to 8. Invention 10: Store in the arithmetic unit learning data showing the relationship between a large number of the X-ray transmission image data acquired in the past and the parameters specified when reconstructing the X-ray transmission image data. When reconstructing the new X-ray transmission image data, a method for image processing of an X-ray transmission image of a polymer material product having a metal member according to any one of Inventions 1 to 9, wherein an initial value of a parameter used for reconstructing the new X-ray transmission image data is set by data processing by the arithmetic unit based on the learning data. Invention 11: The learning data includes histograms of a large number of the X-ray transmission image data acquired in the past, When reconstructing the new X-ray transmission image data, an X-ray transmission image data having a histogram approximated to the histogram of the new X-ray transmission image data is selected from the learning data by the arithmetic unit, and the parameter specified when reconstructing the selected X-ray transmission image data is used as the initial value. A method for image processing of an X-ray transmission image of a polymer material product having a metal member according to Invention 10. Invention 12: The learning data includes information on the polymer material product that was the object of a large number of the X-ray transmission image data acquired in the past, When reconstructing the new X-ray transmission image data, information approximated to the information on the polymer material product that is the object of the new X-ray transmission image data is selected from the learning data by the arithmetic unit, and the parameter specified when reconstructing the X-ray transmission image data of the polymer material product of the selected information is used as the initial value. A method for image processing of an X-ray transmission image of a polymer material product having a metal member according to Invention 10 or 11. Invention 13: A method for image processing of an X-ray transmission image of a polymer material product having a metal member according to Invention 12, wherein the polymer material product is a tire, and the information on the polymer material product includes a tire size and a type of rim. Invention 14: When reconstructing the new X-ray transmission image data, based on the prediction model constructed by machine learning using the learning data and the new X-ray transmission image data, the parameter specified when the new X-ray transmission image data is reconstructed by the arithmetic unit is predicted, and the predicted parameter is used as the initial value. The image processing method for an X-ray transmission image of a polymer material product having a metal member according to Invention 10. Invention 15: In an X-ray transmission image processing system for a polymer material product having a metal member, including an X-ray computed tomography apparatus that acquires X-ray transmission image data of the polymer material product having a metal member, and an arithmetic unit that performs data processing for reconstructing the X-ray transmission image data and outputs a reconstructed image with reduced noise. The arithmetic unit includes an input unit for inputting a parameter used for reconstructing the X-ray transmission image data, an arithmetic processing unit that generates output image data that minimizes the error with the X-ray transmission image data under the condition of reducing a feature amount indicating a predetermined feature of the X-ray transmission image data based on the parameter, and an output unit that outputs the generated output image. An adjustment step of adjusting and inputting the parameter by the input unit, and a generation step of generating the output image by the arithmetic processing unit and outputting it to the output unit are repeatedly performed. The parameter is specified such that the pixel value of each pixel corresponding to an object element including the polymer material product in the output image data is higher than the pixel value of each pixel corresponding to the space other than the object element, and the output image data generated based on the specified parameter is output to the output unit as reconstructed image data. The X-ray transmission image processing system for a polymer material product having a metal member is configured as such.

Explanation of Reference Numerals

[0083] 1 Image processing system 2 X-ray computed tomography apparatus (X-ray CT apparatus) 3 X-ray generator 4 X-ray detector 5 Support 5a Rim 5b Support 6 Computing device 7 Arithmetic processing unit 8 Main memory 9 Auxiliary memory 10 Input unit 11 Output unit R Rubber product (polymer material product) M Metal member D X-ray transmission image data (original image data) Dm Falsely created X-ray transmission image data D1 Output image data D2 (D2a, D2b, D2c, D2d) Reconstructed image data LD (LDa, LDb, LDc) Learning data Ra Object element Z Noise element

Claims

1. In an image processing method for an X-ray transmission image of a polymer material product having a metal member, which acquires X-ray transmission image data of the polymer material product having the metal member and outputs a reconstructed image with reduced noise by reconstructing the X-ray transmission image data using an arithmetic unit, a adjustment step of adjusting parameters used for reconstructing the X-ray transmission image data by the arithmetic unit, and a generation step of generating and outputting output image data that minimizes the error from the X-ray transmission image data under the condition of reducing a feature amount indicating a predetermined feature of the X-ray transmission image data based on the parameters are repeated, to identify the parameters such that the pixel value of each pixel corresponding to an object element including the polymer material product in the output image data is higher than the pixel value of each pixel corresponding to the space other than the object element, and output the output image data generated based on the identified parameters as reconstructed image data. An image processing method for an X-ray transmission image of a polymer material product having a metal member.

2. The image processing method for an X-ray transmission image of a polymer material product having a metal member according to claim 1, wherein in the generation step, total variation regularization for reducing the difference in pixel values between adjacent pixels of the X-ray transmission image data is used, and a weight coefficient for the total variation regularization is used as the parameter.

3. The image processing method for an X-ray transmission image of a polymer material product having a metal member according to claim 1, wherein in the generation step, frequency filtering for reducing frequency components in a power spectrum obtained by Fourier transform is used, and a sampling ratio before and after the frequency filtering is used as the parameter.

4. The image processing method for an X-ray transmission image of a polymer material product having a metal member according to claim 1, wherein in the generation step, total variation regularization for reducing the difference in pixel values between adjacent pixels of the X-ray transmission image data and frequency filtering for reducing frequency components in a power spectrum obtained by Fourier transform are used, and a weight coefficient for each of the total variation regularization and the frequency filtering and a sampling ratio before and after the frequency filtering are used as the parameters.

5. In the generation step, total variation regularization for reducing the difference in pixel values between adjacent pixels of the X-ray transmission image data, L1 regularization for normalizing the sum of absolute values of all pixel values of the X-ray transmission image data, and frequency filtering for reducing frequency components in the power spectrum obtained by Fourier transform are used. As the parameters, the weight coefficients for the total variation regularization, the L1 regularization, and the frequency filtering, respectively, and the sampling ratio before and after the frequency filtering are used. The image processing method for an X-ray transmission image of a polymer material product having a metal member according to claim 1.

6. In the generation step, the total variation regularization, the L1 regularization, and the frequency filtering are individually performed in a predetermined order. The image processing method for an X-ray transmission image of a polymer material product having a metal member according to claim 5.

7. In the generation step, the total variation regularization, the L1 regularization, and the frequency filtering are simultaneously performed using one formula. The image processing method for an X-ray transmission image of a polymer material product having a metal member according to claim 5.

8. The polymer material product is a tire, and the X-ray transmission image data includes an image of at least one of the bead portion or the belt layer of the tire. The image processing method for an X-ray transmission image of a polymer material product having a metal member according to any one of claims 1, 4, or 5.

9. Output the reconstruction image data of one limited region or a plurality of limited regions in one of the X-ray transmission image data. When outputting the reconstruction image data of the plurality of limited regions, for each of the limited regions, the adjustment step and the generation step are repeated to specify the parameters. The image processing method for an X-ray transmission image of a polymer material product having a metal member according to any one of claims 1, 4, or 5.

10. Store learning data showing the relationship between a large number of the X-ray transmission image data acquired in the past and the parameters specified when reconstructing the X-ray transmission image data in the arithmetic unit. When reconstructing new X-ray transmission image data, set an initial value of the parameters used for reconstructing the new X-ray transmission image data by performing data processing by the arithmetic unit based on the learning data. The image processing method for an X-ray transmission image of a polymer material product having a metal member according to any one of claims 1, 4, or 5.

11. The learning data includes histograms of a large number of the X-ray transmission image data acquired in the past, When reconstructing new X-ray transmission image data, an X-ray transmission image data having a histogram approximating to a histogram of the new X-ray transmission image data is selected from the learning data by the arithmetic unit, and the parameter specified when reconstructing the selected X-ray transmission image data is used as the initial value. The image processing method for an X-ray transmission image of a polymer material product having a metal member according to claim 10.

12. The learning data includes information on the polymer material product that was the object of a large number of the X-ray transmission image data acquired in the past, When reconstructing new X-ray transmission image data, information approximating to the information on the polymer material product that is the object of the new X-ray transmission image data is selected from the learning data by the arithmetic unit, and the parameter specified when reconstructing the X-ray transmission image data of the polymer material product of the selected information is used as the initial value. The image processing method for an X-ray transmission image of a polymer material product having a metal member according to claim 10.

13. The image processing method for an X-ray transmission image of a polymer material product having a metal member according to claim 12, wherein the polymer material product is a tire, and the information on the polymer material product includes tire size and type of rim.

14. When reconstructing new X-ray transmission image data, based on a prediction model constructed by machine learning using the learning data and the new X-ray transmission image data, the parameter specified when the arithmetic unit reconstructs the new X-ray transmission image data is predicted, and the predicted parameter is used as the initial value. The image processing method for an X-ray transmission image of a polymer material product having a metal member according to claim 10.

15. In an image processing system for an X-ray transmission image of a polymer material product having a metal member, comprising: an X-ray computed tomography apparatus for acquiring X-ray transmission image data of a polymer material product having a metal member; and an arithmetic unit for performing data processing for reconstructing the X-ray transmission image data and outputting a reconstructed image with reduced noise, The arithmetic unit includes an input unit for inputting parameters used for reconstructing the X-ray transmission image data, an arithmetic processing unit for generating output image data that minimizes the error with the X-ray transmission image data under the condition of reducing a feature amount indicating a predetermined feature of the X-ray transmission image data based on the parameters, and an output unit for outputting the generated output image. An adjustment step of adjusting and inputting the parameters by the input unit and a generation step of generating the output image by the arithmetic processing unit and outputting the output image to the output unit are repeatedly performed. An X-ray transmission image processing system for a polymer material product having a metal member configured such that parameters are specified in which pixel values of respective pixels corresponding to object elements including the polymer material product in the output image data are higher than pixel values of respective pixels corresponding to spaces other than the object elements, and the output image data generated based on the specified parameters is output to the output unit as reconstructed image data.

Images