Calculation device and program

Abstract

A calculation device according to the present invention comprises: an acquisition unit that acquires information of a component group constituted by a plurality of components; and a calculation unit that calculates, on the basis of the information of the component group, a quantitative value pertaining to the overall circumferentiality of the component group.

Description

Technical Field

[0001] The present invention relates to a calculation apparatus and a program.

Background Art

[0002] In image processing technology, techniques for extracting contours and the like of elements included in an image are known. For example, Patent Document 1 discloses a technique for extracting line segments by performing Hough transform.

Prior Art Literature

Patent Literature

[0003]

Patent Document 1

Summary of the Invention

Problem to be Solved by the Invention

[0004] Incidentally, the quality of extracted elements (components) extracted from an image in image processing is generally basically determined based on the perceptual evaluation of an observer. Therefore, there is a demand for a technique that calculates an index with which such components can be appropriately evaluated.

[0005] An object of the present invention is to provide a calculation apparatus and a program that can calculate an index with which components can be appropriately evaluated.

Means for Solving the Problem

[0006] The calculation apparatus according to the present invention includes: an acquisition unit that acquires information of a component group composed of a plurality of components; and a calculation unit that calculates a quantitative value related to the overall circularity of the component group based on the information of the component group, . 、 The aforementioned component is capable of identifying its spatial distribution, The aforementioned quantitative value is the first point distance between a predetermined point surrounded by each component of the component group and one of the components, or the reciprocal of the first point distance. .

[0007] The program according to the present invention is It is a program, On the computer, A process to retrieve information about a group of components consisting of multiple components, A calculation process to calculate a quantitative value relating to the periphery of the component group based on the information of the component group, Let's execute it 、 The aforementioned component is capable of identifying its spatial distribution, The aforementioned quantitative value is the first point distance between a predetermined point surrounded by each component of the component group and one of the components, or the reciprocal of the first point distance. . [Effects of the Invention]

[0008] According to the present invention, it is possible to calculate an index that allows for appropriate evaluation of components. [Brief explanation of the drawing]

[0009] [Figure 1] This is a block diagram showing an evaluation apparatus according to the first embodiment of the present invention. [Figure 2A] This is a diagram to explain the calculation of quantitative values. [Figure 2B] This is a diagram to explain the calculation of quantitative values. [Figure 2C] This is a diagram to explain the calculation of quantitative values. [Figure 3] This flowchart shows an example of the operation of the calculation control of the evaluation device. [Figure 4] This flowchart shows an example of the operation of the calculation control of the evaluation device according to the second embodiment. [Figure 5A] This figure shows an example of a group of components. [Figure 5B] This figure shows an example of a group of components. [Figure 6] This figure shows an example of a persistent diagram. [Figure 7A] This figure shows an example of a group of components. [Figure 7B]It is a diagram illustrating an example of a component group. [Figure 8A] It is a diagram illustrating an example of a persistent diagram. [Figure 8B] It is a diagram illustrating an example of a persistent diagram. [Figure 9] It is a diagram illustrating an example of plot assignment for a component group in an original image. [Figure 10] It is a flowchart illustrating an operation example of calculation control of the evaluation device according to the third embodiment. [Figure 11] It is a block diagram illustrating a learning device according to the fourth embodiment. [Figure 12] It is a flowchart illustrating an operation example of machine learning control of the learning device according to the fourth embodiment. [Figure 13] It is a diagram illustrating an example of a component group. [Figure 14A] It is a diagram illustrating a component that is a line. [Figure 14B] It is a diagram illustrating an enlarged state of the component of FIG. 14A. [Figure 14C] It is a diagram illustrating a component that is a region. [Figure 14D] It is a diagram illustrating an enlarged state of the component of FIG. 14C. Description of Embodiments

[0010] Hereinafter, a first embodiment of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram illustrating an evaluation apparatus 1 according to the first embodiment of the present invention.

[0011] As illustrated in FIG. 1, the evaluation apparatus 1 is an apparatus for evaluating whether components included in an image are acceptable, for example, in a PC (Personal Computer) or the like. The evaluation apparatus 1 calculates a quantitative value that can be used for evaluating a component group configured of a plurality of components and that enables evaluation of the global property of the component group. The evaluation apparatus 1 corresponds to the "calculation apparatus" of the present invention.

[0012] < / think_never_used_51bce0c785ca2f68081bfa7d91973934> Circumference is an indicator of a predetermined level of reliability for the contour formed when each component of a group of components is enclosed within it.

[0013] A group of components is a distribution of multiple components that arise from applying image processing to elements contained in an image.

[0014] A component is an element that can identify the spatial distribution of points, lines, regions, shades, etc., in an image, and is generated, for example, by applying a predetermined image processing to the image. In this embodiment, a component is, for example, a point. The object that is the target of a component is an object that can show some kind of contour shape depending on the distribution of each component, and may be, for example, the contour shape of an object corresponding to an element included in the image, or it may be multiple attached objects attached to the periphery of a predetermined object.

[0015] The specified image processing can be any processing that allows for the representation of components in an image, such as, for example, binarizing a grayscale image.

[0016] The evaluation device 1 is equipped with a CPU (Central Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), etc. The CPU reads a program corresponding to the processing content from ROM, loads it into RAM, and works in cooperation with the loaded program to centrally control the operation of each block of the evaluation device 1.

[0017] Furthermore, the evaluation device 1 includes an image processing unit 10, a calculation unit 20, and an evaluation unit 30. The image processing unit 10 performs the predetermined image processing described above on the image to be evaluated. Note that the image processing unit 10 does not need to be provided if the system is configured to acquire the processed image from another image processing device or the like.

[0018] The calculation unit 20 acquires information about a group of components, which consists of multiple components extracted by predetermined image processing in the image processing unit 10. The calculation unit 20 then calculates a quantitative value relating to the circumferentiality of the component group based on the information about the component group. The calculation unit 20 corresponds to the "acquisition unit" and "calculation unit" of the present invention.

[0019] The calculation unit 20 calculates a quantitative value related to circumferentiality based on the positional relationship of each component in the component group. Specifically, the calculation unit 20 calculates a quantitative value based on the first point distance between a predetermined point surrounded by each component in the component group and one of the components. This calculation method may be based on a method related to persistent homology, for example, as shown below.

[0020] The predetermined point is the intersection of at least two components when the gaps between the enlarged components are filled as each component is enlarged. Enlarging each component means enlarging each component by a constant rate. For example, if a component is a point, each component is enlarged by a constant rate such that the circle centered on the component is always equal in diameter.

[0021] For example, suppose we have a first component C1, a second component C2, a third component C3, and a fourth component C4, as shown in Figure 2A.

[0022] The first component C1 is the component located in the upper left of the four components in Figure 2A. The second component C2 is the component located in the upper right of the four components in Figure 2A. The third component C3 is the component located in the lower right of the four components in Figure 2A. The fourth component C4 is the component located in the lower left of the four components in Figure 2A.

[0023] As each component is enlarged, two of the four components that are relatively close to each other will intersect. Then, as shown in Figure 2B, all the components connect, and a ring enclosed by all the components is generated. Of the two adjacent components in each component, the intersection point of the two components that are last connected when the ring is generated becomes the Birth point.

[0024] In Figure 2B, etc., the dashed lines C11, C21, C31, and C41 represent enlarged versions of the first component C1, the second component C2, the third component C3, and the fourth component C4, respectively.

[0025] In Figure 2B, C31, which corresponds to the third component C3, and C41, which corresponds to the fourth component C4, are connected at the end, and this intersection B is the berth point.

[0026] As each component is enlarged, the gap enclosed by all the components gradually shrinks until it is completely filled, as shown in Figure 2C. The intersection of at least two components when this gap is filled becomes the Death point, or predetermined point.

[0027] In Figure 2C, the connection between C11, which corresponds to the first component C1, and C31, which corresponds to the third component C3, completely fills the gap enclosed by all the components. The intersection point D of C11 and C31 at this point is the death point. Note that in the example shown in Figure 2C, the death point is the intersection of two components, but depending on the distribution of each component, the death point may be the intersection of three or more components.

[0028] The calculation unit 20 determines the first point-to-point distance to be the point at which the distance between the death point and the center point of each component is minimized. In the example shown in Figure 2C, the death point is the intersection point D of C11 and C31, and the intersection point D lies on the circumference of C11 and C31, outside of C21 and C41. Since the radii of C11, C21, C31, and C41 are the same, the first point-to-point distance is the distance between the first component C1 or the third component C31 and the intersection point D.

[0029] The calculation unit 20 uses the reciprocal of this first point-to-point distance as a quantitative value related to circumferentiality. Alternatively, the first point-to-point distance may be used as the quantitative value related to circumferentiality. Furthermore, the first point-to-point distance is not limited to the distance between the death point and the center point of each component that minimizes the distance between them. The first point-to-point distance may be appropriately selected according to the contour shape of the component group. For example, the first point-to-point distance may be the distance between the death point and the center point of each component that maximizes the distance between them, or an intermediate distance among all point-to-point distances. Alternatively, the component to which the first point-to-point distance applies may be predetermined, and the distance between that component and the death point may be used as the first point-to-point distance.

[0030] The evaluation unit 30 evaluates the circumferential properties of the component group based on quantitative values ​​related to circumferential properties. For example, the evaluation unit 30 evaluates circumferential properties such that the larger the quantitative value, the higher the evaluation of circumferential properties.

[0031] The above evaluation criteria are merely examples and can be modified as appropriate depending on the type of object being evaluated. Furthermore, the evaluation by the evaluation unit 30 can be any evaluation method as long as it allows for the determination of the quality of the omnidirectional performance. For example, it may be indicated by a score (number) that shows the level of quantitative values ​​in stages, or by using symbols such as "○", "△", or "×".

[0032] By doing so, it becomes possible to calculate metrics that allow for appropriate evaluation of the components.

[0033] An example of the operation of the calculation control of the evaluation device 1 described above will now be explained. Figure 3 is a flowchart showing an example of the operation of the calculation control of the evaluation device 1. The process in Figure 3 is executed as appropriate, for example, when information of the image to be evaluated is input to the evaluation device 1.

[0034] As shown in Figure 3, the evaluation device 1 performs image processing on the image to be evaluated (step S101). Next, the evaluation device 1 calculates the first point distance in the group of components generated by the image processing (step S102).

[0035] After calculating the distance between the first points, the evaluation device 1 calculates a quantitative value related to the circumferential properties (step S103). Then, the evaluation device 1 evaluates the circumferential properties based on the quantitative value (step S104). After that, the control process ends.

[0036] According to this embodiment configured as described above, quantitative values ​​related to omnidirectionality can be calculated from multiple components in an image. As a result, an index that can appropriately evaluate the components can be calculated.

[0037] Furthermore, since quantitative values ​​are calculated, it becomes easier to establish consistent evaluation criteria.

[0038] For example, if quantitative values ​​are not calculated, the quality of a group of components will be judged based on the observer's subjective evaluation. This means that the evaluation criteria will vary depending on the observer, potentially making it impossible to perform an appropriate evaluation.

[0039] In contrast, in this embodiment, it becomes easier to establish certain evaluation criteria based on the calculation of quantitative values, thus enabling appropriate evaluation of the component group.

[0040] Furthermore, since quantitative values ​​are calculated based on the distance between points (distance between components), quantitative values ​​can be easily calculated without complex calculations.

[0041] Next, a second embodiment will be described.

[0042] In the first embodiment described above, the quantitative value was calculated based on the Death point described above, but in the second embodiment, for example, the quantitative value may be calculated based on the Barth point described above.

[0043] Specifically, the calculation unit 20 calculates a quantitative value based on the distance between the second points of two adjacent components in the contour formed when each component of the component group is enclosed.

[0044] The second point distance is the maximum distance between two adjacent components. The Birth point is, as described above, the intersection point of two adjacent components when they are finally connected, after a ring enclosed by all components has been generated.

[0045] Since the radius of each component remains the same when each component is enlarged, the two adjacent components that connect at the end of the ring generation will be the furthest apart components in the group.

[0046] Therefore, the distance between the two components corresponding to the berth point becomes the second point distance. For example, in the example shown in Figure 2B, C31, which corresponds to the third component C3, and C41, which corresponds to the fourth component C4, are connected at the end, and their intersection point B is the berth point, so the distance between the third component C3 and the fourth component C4 becomes the second point distance.

[0047] Furthermore, the calculation unit 20 uses this distance between the second point as the quantitative value for the circumferential measurement. Alternatively, the reciprocal of the distance between the second point may be used as the quantitative value for the circumferential measurement.

[0048] An example of the operation of the calculation control of the evaluation device 1 according to the second embodiment will be described. Figure 4 is a flowchart showing an example of the operation of the calculation control of the evaluation device 1 according to the second embodiment. The processing in Figure 4 is executed as appropriate, for example, when information of the image to be evaluated is input to the evaluation device 1.

[0049] As shown in Figure 4, the evaluation device 1 performs image processing on the image to be evaluated (step S201). Next, the evaluation device 1 calculates the distance between the second point in the group of components generated by the image processing (step S202).

[0050] After calculating the distance between the two points, the evaluation device 1 calculates a quantitative value related to the circumferential properties (step S203). Then, the evaluation device 1 evaluates the circumferential properties based on the quantitative value (step S204). After that, the control process ends.

[0051] According to the second embodiment configured as described above, similar to the first embodiment, quantitative values ​​related to omnidirectionality can be calculated from multiple components in the image. As a result, an index that can appropriately evaluate the components can be calculated.

[0052] Next, a third embodiment will be described.

[0053] In the above embodiments, quantitative values ​​related to circumferentiality were calculated based on either the Death point or the Barth point. However, in the third embodiment, quantitative values ​​related to circumferentiality are calculated based on both the Death point and the Barth point.

[0054] Specifically, the calculation unit 20 calculates the distance between the first point and the distance between the second point, similar to the embodiments described above.

[0055] The calculation unit 20 then calculates the difference between the distance between the second point and the distance between the first point, and sets this difference as a quantitative value.

[0056] The evaluation unit 30 evaluates the component group as having good omnidirectional properties if the quantitative value calculated by the calculation unit 20 is greater than a predetermined threshold.

[0057] For example, suppose an element in an image has a shape that approximates a perfect circle, and multiple deposits are present around it. Then, suppose these deposits are generated as components. As shown in Figure 5A, suppose the group of components generated from this element are arranged almost uniformly along the contour shape. Such a shape is one that we would like to evaluate as having good omnidirectional properties.

[0058] In this case, the distance between two adjacent components tends to remain constant across multiple components, and the aforementioned ring is generated relatively quickly when the components are enlarged.

[0059] Therefore, the distance between the two points tends to be relatively small.

[0060] Furthermore, with this shape, the position where the gaps surrounded by each component are filled tends to be close to the center of the component group, so the distance between the first points tends to be relatively large.

[0061] Therefore, the difference between the distance between the first point and the distance between the second point tends to be a relatively large value.

[0062] Furthermore, as shown in Figure 5B, suppose that the group of components generated from the above elements has some areas where the distance between components is wide (areas enclosed by dashed lines). Such a shape should be evaluated as having poor circumferential properties.

[0063] In this case, the distance between the two components corresponding to the area enclosed by the dashed line is greater than the distance between the other two components. Therefore, when the component is enlarged, the generation of the ring described above will be slower compared to the shape shown in Figure 5A.

[0064] Therefore, the distance between the two points tends to be relatively large.

[0065] Furthermore, since the position where the gaps surrounded by each component are filled tends to be close to the center of the component group, the distance between the first points will be approximately the same as in Figure 5A.

[0066] As a result, the difference between the distance between the first point and the distance between the second point tends to be a relatively small value.

[0067] Thus, a difference is likely to occur between the difference between the first and second point distances in the component group in Figure 5A and the difference between the first and second point distances in the component group in Figure 5B. Therefore, it becomes easier to evaluate the quality of the circumferential coverage according to the predetermined threshold mentioned above.

[0068] In other words, in the third embodiment, since the quantitative value is calculated by considering both the distance between the first point and the distance between the second point, the evaluation index for circumferentiality can be made even more accurate.

[0069] Furthermore, the calculation unit 20 may, for example, as shown in Figure 6, use a two-dimensional coordinate system (persistent diagram) with the parameter corresponding to the first point distance on the vertical axis and the parameter corresponding to the second point distance on the horizontal axis, and the points corresponding to the calculated first and second point distances as quantitative values.

[0070] In this case, defining a specific area on the persistent diagram as an area with good circumferential properties makes it easier to evaluate circumferential properties. For example, if the area where the distance between the second point is B1 or less and the distance between the first point is D1 or more (the area indicated by the dots) is defined as an area with good circumferential properties, then the evaluation of circumferential properties can be determined simply by checking whether or not the object falls within that area.

[0071] Furthermore, when multiple component groups exist, the persistent diagram can display points representing each component group, making it easier to judge the overall omnipresence of the image.

[0072] For example, suppose the contour shape of an element in an image is distorted, with a part of it being concave towards the center, and multiple attached objects are present around it. Let's say we generate these attached objects as components. As shown in Figure 7A, we want to evaluate a shape as having good circumferential properties if components are present even in the concave area (dashed line), and as shown in Figure 7B, we want to evaluate a shape as having poor circumferential properties if components are not present in the concave area (dashed line).

[0073] In the case of Figure 7A, the distance between two adjacent components tends to be constant across multiple components, and the above-mentioned ring is generated relatively quickly when the components are enlarged.

[0074] Therefore, the distance between the two points tends to be relatively small.

[0075] Furthermore, because components also exist in the recessed areas, when the components are enlarged, the gaps between each component are filled completely relatively quickly, and the distance between the first points tends to become relatively small.

[0076] In contrast, in the case of Figure 7B, the distance between the two components flanking the dashed line is wider compared to other cases, which means that the birth point occurs later, and the distance between the second point tends to be relatively large.

[0077] Furthermore, because there are no components in the recessed areas, when a component is enlarged, the gaps between each component are filled completely relatively late. In this case, the death point tends to be closer to the center of the component group, so the distance between the first points tends to be relatively large.

[0078] Here, we define a region with good omnidirectional properties in the persistent diagram as having a second point distance of B2 or less and a first point distance of D2 or more, as shown in Figure 8A. B2 can be appropriately set according to the second point distance considering the timing of the death point when a component is present in the recessed area of ​​Figure 7A. Specifically, B2 is set to be greater than the second point distance when a component is present in the recessed area as in Figure 7A, and smaller than the second point distance when a component is not present in the recessed area as in Figure 7B. D2 can be appropriately set according to the first point distance considering the timing of the death point when a component is present around a shape like that of Figure 7A.

[0079] For example, in the case where a component is present in the recessed area as shown in Figure 7A, the distance between the second point is less than or equal to B2, and the distance between the first point is greater than or equal to D2, so it can be evaluated as having good circumferential coverage. On the other hand, in the case where there is no component in the recessed area as shown in Figure 7B, the distance between the second point is greater than B2 at least, so it can be evaluated as having poor circumferential coverage.

[0080] Furthermore, the region in the persistent diagram that is considered to have good perimeter may be defined, for example, as shown in Figure 8B, where the distance between the second point is B2 or less and the distance between the first point is D3 or less. D3 can be set to be greater than the distance between the first point when a component exists in a recessed area as shown in Figure 7A, and smaller than the distance between the first point when no component exists in a recessed area as shown in Figure 7B.

[0081] By setting it in this way, for example, in the case where a component exists in the recessed area as shown in Figure 7A, the distance between the second point is B2 or less and the distance between the first point is D3 or less, so it can be evaluated as having good circumferential coverage. On the other hand, in the case where there is no component in the recessed area as shown in Figure 7B, the distance between the second point is greater than B2 and the distance between the first point is greater than D3, so it can be evaluated as having poor circumferential coverage.

[0082] Furthermore, the region considered to have good perimeter in the persistent diagram may be appropriately set to one of four regions composed of a threshold for the first point distance and a threshold for the second point distance, depending on the shape of the elements for which perimeter evaluation is performed. The threshold for the first point distance corresponds to the line demarcated by D1 in Figure 6, D2 in Figure 8A, and D3 in Figure 8B. The threshold for the second point distance corresponds to the line demarcated by B1 in Figure 6, B2 in Figure 8A, and B3 in Figure 8B.

[0083] For example, for shapes where the distance between the first point is large and the distance between the second point is small, the upper left region of the four regions composed of each threshold can be set as the region with good circumferential properties, as shown in Figures 6 and 8A. For shapes where the distance between the first point is small and the distance between the second point is small, the lower left region of the four regions composed of each threshold can be set as the region with good circumferential properties, as shown in Figure 8B. For shapes where the distance between the first point is small and the distance between the second point is large, the lower right region of the four regions composed of each threshold can be set as the region with good circumferential properties. For shapes where the distance between the first point is large and the distance between the second point is large, the upper right region of the four regions composed of each threshold can be set as the region with good circumferential properties.

[0084] Furthermore, the evaluation unit 30 may also plot the component group in the image if the quantitative value of circumferentiality falls within a predetermined range. The predetermined range is, for example, the range in which circumferentiality is judged to be good according to the evaluation criteria.

[0085] For example, as shown in Figure 6, when an image containing six component groups is evaluated, suppose that among the points representing these component groups, A1, A2, and A3 are located in a region with good omnipresence on the persistent diagram, while A4, A5, and A6 are outside that region.

[0086] In this case, as shown in Figure 9, the evaluation unit 30 assigns plot P to the component group corresponding to A1, A2, and A3 in the component group of the original image. Note that no plot P is assigned to the component group corresponding to A4, A5, and A6 in the image.

[0087] By doing this, it becomes easy to determine which group of components in an image has good omnidirectional properties.

[0088] The position of the plot in the image can be anywhere, as long as it can be determined that it corresponds to the group of components to which the plot is assigned. For example, the plot may be located at a position corresponding to the centroid of the contour shape, or at a position corresponding to the death point or birth point of the group of components.

[0089] Furthermore, the evaluation unit 30 may evaluate the overall omnipresence of the image according to the number of plots.

[0090] An example of the operation of the calculation control of the evaluation device 1 according to the third embodiment will be described. Figure 10 is a flowchart showing an example of the operation of the calculation control of the evaluation device 1 according to the third embodiment. The processing in Figure 10 is executed as appropriate, for example, when information of the image to be evaluated is input to the evaluation device 1.

[0091] As shown in Figure 10, the evaluation device 1 performs image processing on the image to be evaluated (step S301). Next, the evaluation device 1 calculates the distance between the first point and the distance between the second point in the group of components generated by the image processing (step S302).

[0092] After calculating the distance between the first and second points, the evaluation device 1 calculates a quantitative value related to the circumferential properties (step S303). Then, the evaluation device 1 evaluates the circumferential properties based on the quantitative value (step S304). After that, the control process ends.

[0093] According to the third embodiment configured as described above, since the quantitative value is calculated by considering both the distance between the first point and the distance between the second point, even for a group of components with an irregular contour shape, the quantitative value can be made a quantitative value that allows for the desired evaluation. As a result, the evaluation index for circumferentiality can be made even more accurate.

[0094] Furthermore, by indicating points representing quantitative values ​​on the persistent diagram, it becomes easier to evaluate circumferential coverage.

[0095] Furthermore, by adding plots to the parts of the image corresponding to the component groups that have been determined to have good omnidirectional properties, it becomes easier for users to determine which parts of the image have good omnidirectional properties.

[0096] Next, a fourth embodiment will be described.

[0097] As shown in Figure 11, the learning device 2 according to the fourth embodiment has a learning unit 40 in addition to the image processing unit 10 and calculation unit 20 in the configuration of each of the above embodiments. The learning device 2 may also have an evaluation unit. The learning device 2 corresponds to the "calculation device" of the present invention.

[0098] The learning unit 40 performs machine learning based on the quantitative values ​​calculated by the calculation unit 20. The machine learning method may be supervised learning or unsupervised learning. Furthermore, the machine learning method may be reinforcement learning or semi-supervised learning.

[0099] Examples of supervised learning methods include linear regression, deep learning, neural networks, support vector machines, LASSO regression, Ridge regression, random forests, logistic regression, support vector regression, naive Bayes, and k-nearest neighbors.

[0100] Examples of unsupervised learning methods include k-means, principal component analysis, non-negative matrix factorization, Gaussian mixture distributions, and t-SNE.

[0101] The learning unit 40 performs machine learning, using explanatory variables, for example, the generation conditions for the object to be evaluated, and the objective variable as a quantitative value, to create a prediction formula for generating the object to be evaluated. Examples of generation conditions for the object to be evaluated include temperature conditions and time conditions when generating the object to be evaluated. Note that the explanatory variables are not limited to the generation conditions for the object to be evaluated, but may include other conditions as well.

[0102] By doing so, it becomes easier to predict the optimal conditions for the explanatory variables in the subject being evaluated.

[0103] An example of the operation of the machine learning control of the learning device 2 according to the fourth embodiment will be described. Figure 12 is a flowchart showing an example of the operation of the machine learning control of the learning device 2 according to the fourth embodiment. The processing in Figure 12 is executed as appropriate, for example, when information of the image to be evaluated is input to the learning device 2.

[0104] As shown in Figure 12, the learning device 2 calculates a quantitative value related to omnidirectionality (step S401). The method for calculating the quantitative value is the same as the calculation method shown in Figures 3, 4, and 10.

[0105] Once the quantitative values ​​are calculated, the learning device 2 performs machine learning using the explanatory variables and the target variable (quantitative values) (step S402). Then, the learning device 2 generates a prediction formula (step S403). After that, this control process ends.

[0106] According to the fourth embodiment configured as described above, by performing machine learning using quantitative values ​​related to omnidirectionality, an effective learning effect that reflects these quantitative values ​​can be obtained.

[0107] In the above embodiment, the case where noise occurs inside or outside the component group was not specifically mentioned. For example, as shown in Figure 13, when deposits around a perfectly round object (solid line portion) are considered components, components may be detected as noise inside the solid line portion. In such cases, for example, if a group consisting of the component corresponding to the noise and the surrounding components is considered a component group (for example, the component enclosed by the dashed line), the circumferential nature of the component group may be evaluated. However, in the present invention, such components corresponding to noise may be excluded.

[0108] For example, components dealing with noise may experience burst points at different timings than normal components, or death points at relatively earlier timings.

[0109] Therefore, the first point distance for components corresponding to noise tends to be smaller than the first point distance for normal components.

[0110] Therefore, for example, the calculation unit 20 compares the calculated inter-point distance with a noise threshold, and if it is smaller than the noise threshold, it excludes the group of components related to that inter-point distance from the calculation of quantitative values. The noise threshold can be set appropriately according to the size of the object being evaluated for omnidirectional performance.

[0111] By doing so, it is possible to calculate quantitative values ​​related to circumferential properties after excluding the components inside the solid line shown in Figure 13.

[0112] Furthermore, the evaluation of omnidirectional performance may be performed by considering the components corresponding to the noise mentioned above. For example, in the components inside the solid line shown in Figure 13, the further you move from the solid line, the earlier the gaps between each component will be completely filled, and the distance between the first points will also decrease.

[0113] Therefore, for example, in the evaluation of omnidirectionality, the threshold for the distance between the first point may be adjusted to a value such that components located close to the solid line are considered to have good omnidirectionality. The threshold for the distance between the second point may also be adjusted in the same way. This would, for example, narrow the region in the persistent diagram that is considered to have good omnidirectionality by eliminating noise.

[0114] This approach helps to suppress the influence of noise on the omnidirectional evaluation, for example, when there are components that are clearly noise.

[0115] Furthermore, in this case, for components moving away from the solid line, the distance between the first points falls below the threshold range mentioned above, which affects the evaluation of omnidirectionality. In this case, the omnidirectionality can be evaluated as poor due to noise.

[0116] Furthermore, although the above embodiments included an evaluation unit 30, an evaluation unit is not required as long as quantitative values ​​can be calculated.

[0117] Furthermore, the objects of evaluation in each of the above embodiments, that is, the objects for which quantitative values ​​related to circumferential properties are calculated, vary. For example, if the electrode catalyst of a fuel cell is used as the object of calculation, quantifying the circumferential properties of the electrode catalyst makes it possible to improve output by the electrode catalyst and reduce costs by reducing the amount of platinum used.

[0118] Furthermore, if the calculation target is a metal nanoparticle catalyst for extracting hydrogen energy, it becomes possible to provide a service that optimizes the quality of the metal nanoparticle catalyst by optimizing its circumferential properties.

[0119] Furthermore, if the calculation target is the odor particles on the surface of the sensor part of the odor sensor, it becomes possible to quantify the degree of adsorption of those particles by using them as the calculation target.

[0120] Furthermore, by limiting the calculation to technologies for collecting marine microplastics, it becomes possible to detect the degree of algal coverage adsorbed onto the surface of microplastics, as well as the degree of microplastic adsorption onto porous algae.

[0121] Furthermore, if the calculation target is electronic granular material, it can be applied to the evaluation of its quality (porous products, surface-treated products).

[0122] Furthermore, if the calculation targets are related to CO2 conversion technology, it becomes possible to create a large number of finely arranged, uniform active sites (nickel metal) within a three-dimensional porous matrix to achieve high conversion efficiency.

[0123] Furthermore, by using organic / inorganic composite particles and inorganic hollow particles that utilize the catalytic activity of PDMAEMA as the calculation targets, it becomes possible to optimally impart functions such as anti-reflective materials and heat insulating materials using these particles.

[0124] Furthermore, if the calculation targets cancer cells related to the roughly cylindrical digestive system in the human body, such as colorectal cancer, small intestine cancer, and esophageal cancer, it becomes easier to confirm the progression of those cancer cells.

[0125] Furthermore, by using blood vessels as the target of the calculation, it becomes easier to check the degree of thrombosis in the blood vessels.

[0126] Furthermore, by including water pipes, gas pipes, etc., in the calculation, it becomes easier to check the degree of blockage caused by foreign objects in these essential infrastructures.

[0127] Furthermore, various other items may be included in the calculation, such as liposome formulations and other materials with a core-shell structure (nanoparticle materials for drug transport), toners with a core-shell structure, fertilizers, and pest control agents (insecticides, herbicides, fungicides).

[0128] Furthermore, while the above embodiments calculated quantitative values ​​related to circumferentiality for a group of components in a two-dimensional image, the present invention is not limited thereto, and quantitative values ​​related to circumferentiality may also be calculated for a group of components in a three-dimensional image.

[0129] Furthermore, while points were shown as an example of a component (something whose spatial distribution can be identified) in the above embodiment, quantitative values ​​can be calculated in the same way even when the component is a line or a region (a region of varying shades).

[0130] For example, if component C5 is a line as shown in Figure 14A, the component will be enlarged in a shape (for example, a rectangle) that is at a constant distance from component C5, as shown by the dashed line C51 in Figure 14B.

[0131] Furthermore, as shown in Figure 14C, if component C6 is a region, the component is enlarged in a shape that is at a constant distance from component C6, as shown by the dashed line C61 in Figure 14D. Note that in Figures 14A to 14D, the same applies to other components, so for convenience, only one of each component is shown.

[0132] Furthermore, while the above embodiments primarily calculated quantitative values ​​related to omnidirectionality for still images, the present invention is not limited thereto, and quantitative values ​​related to omnidirectionality may also be calculated for moving images.

[0133] Furthermore, in the embodiments described above, the components were elements generated by image processing in an image, but the present invention is not limited to this. For example, the components may be artificially created (created with computer graphics (CG) or by drawing dots with a stylus), or they may relate to the results of some processing, such as a simulation.

[0134] Furthermore, the components may be extracted using image recognition to determine points or regions, extracted using non-image measuring instruments such as 3D sensors or ultrasonic sensors, or extracted using information obtained from sensors attached to the object itself, such as RFID.

[0135] Furthermore, while the quantitative values ​​in the above embodiments were calculated based on a method related to persistent homology, the present invention is not limited thereto, and quantitative values ​​may be calculated using calculation methods other than the said method. For example, quantitative values ​​may be calculated by calculating the required point-to-point distance for all combinations of three or more components from among multiple components.

[0136] Furthermore, in the quantitative evaluation of each of the above embodiments, for example, if there are multiple component groups, the higher the number of component groups that receive a good evaluation, the higher the evaluation may be, or it may be the number of component groups that receive an evaluation above a certain level. Also, the evaluation may change depending on the distribution of omnidirectionality within a predetermined region. The predetermined region may be the entire image, a local region within the image, or a region set in physical space.

[0137] Furthermore, the criteria for judging the quality of the circumferential evaluation can be appropriately modified depending on the subject being calculated.

[0138] Furthermore, the above embodiments are merely examples of how the present invention may be implemented, and the technical scope of the present invention should not be limited by them. In other words, the present invention can be implemented in various ways without departing from its gist or its main features.

[0139] All disclosures in the specification, drawings, and abstract contained in the Japanese application 2022-024029, filed on 18 February 2022, are incorporated herein by reference. [Explanation of symbols]

[0140] 1. Evaluation device 2 Learning device 10 Image Processing Unit 20 Calculation Section 30 Evaluation Department 40 Learning Department

Claims

1. An acquisition unit that acquires information about a group of components consisting of multiple components, A calculation unit that calculates a quantitative value relating to the circumferential properties of the component group based on the information of the component group, Equipped with, The aforementioned component is capable of identifying its spatial distribution, The aforementioned quantitative value is the first point distance between a predetermined point surrounded by each component of the component group and one of the components, or the reciprocal of the first point distance. Calculation device.

2. An acquisition unit that acquires information on a group of components composed of multiple components, A calculation unit that calculates a quantitative value relating to the circumferential properties of the component group based on the information of the component group, Equipped with, The aforementioned component is capable of identifying its spatial distribution, The aforementioned quantitative value is the distance between two adjacent points in the contour formed when each component of the component group is enclosed, or the reciprocal of the distance between those two points. Calculation device.

3. Further comprising an evaluation unit that evaluates the circumferential properties based on the quantitative value, The evaluation unit evaluates the circumferential properties such that the evaluation of the circumferential properties differs as the quantitative value increases. The calculation apparatus according to claim 1 or claim 2.

4. The aforementioned periphery is an indicator of predetermined reliability for the contour formed when each component of the component group is enclosed. The calculation apparatus according to claim 1 or claim 2.

5. The predetermined point is the intersection of at least two components when the gaps surrounding the enlarged components are filled as each component is enlarged. The calculation apparatus according to claim 1.

6. The distance between the second point is the maximum distance among the distances between the two adjacent components. The calculation apparatus according to claim 2.

7. The aforementioned group of components is a distribution of the multiple components that are generated by applying image processing to the elements contained in the image. The calculation apparatus according to claim 1 or claim 2.

8. The evaluation unit, when the evaluation value of the omnidirectional aspect falls within a predetermined range, assigns a plot to the group of components in the image. The calculation apparatus according to claim 3.

9. The system includes a learning unit that performs machine learning based on the aforementioned quantitative values. The calculation apparatus according to claim 1 or claim 2.

10. It is a program, On the computer, A process to retrieve information about a group of components consisting of multiple components, A calculation process to calculate a quantitative value relating to the circumferential properties of the component group based on the information of the component group, Make it run, The aforementioned component is capable of identifying its spatial distribution, The aforementioned quantitative value is the first point distance between a predetermined point surrounded by each component of the component group and one of the components, or the reciprocal of the first point distance. program.

11. A program, On the computer, A process to retrieve information about a group of components consisting of multiple components, A calculation process to calculate a quantitative value relating to the circumferential properties of the component group based on the information of the component group, Make it run, The aforementioned component is capable of identifying its spatial distribution, The aforementioned quantitative value is the distance between two adjacent points in the contour formed when each component of the component group is enclosed, or the reciprocal of the distance between those two points. program.

12. An acquisition unit that acquires information on a group of components composed of multiple components, A calculation unit that calculates a quantitative value relating to the circumferential properties of the component group based on the information of the component group, Equipped with, The aforementioned component is capable of identifying its spatial distribution, The aforementioned quantitative value is the difference between the distance between the first point and the distance between the second point. The first point distance is the distance between a predetermined point surrounded by each component of the component group and one of the components. The second point distance is the distance between two adjacent points in the contour formed when each component of the component group is enclosed. Calculation device.

Images