JP6198242B2 - Boundary detection method - Google Patents

Description

  The present invention relates to a boundary detection method for detecting a boundary between stacked different materials with respect to a member formed by stacking different materials.

The ratio of each layer occupying the cross section in the stacking direction in a metal member configured by stacking different types of metal materials or a resin member configured by stacking different types of resin materials (hereinafter referred to as stacking ratio) ) Has been developed.
Measurement of the lamination rate in a sample such as a metal member or a resin member includes measurement of the lamination rate by measuring the cross-sectional shape along the lamination direction of the sample, and lamination rate by irradiating fluorescent X-rays from the outside of the sample. The two methods of measuring are mainly adopted.

The method of irradiating fluorescent X-rays has the advantage that the stacking ratio can be measured without destroying the sample. However, the method of irradiating fluorescent X-rays has a limitation that a thick member cannot be measured and the member to be measured needs to contain a specific element. Therefore, the measurement of the lamination rate by measuring the shape of a cross section is widely performed.
Patent Document 1 discloses a measuring apparatus, a measuring program, and a measuring method for measuring a stacking ratio by measuring the shape of a cross section.

  The measurement apparatus disclosed in Patent Document 1 includes an image acquisition unit that captures a cross-sectional image related to a measurement target member in which a coating layer is formed on the surface of a core material, and a change in pixel value in the thickness direction of the cross-sectional image based on a change in pixel value. Boundary determining means for obtaining two opposing boundaries, and calculating means for calculating a stacking ratio of the coating layer by calculating a distance between the boundaries.

JP 2007-178375 A

  In Patent Document 1, an attempt is made to recognize a boundary between a core material and a coating layer by binarizing an image based on luminance. However, in a binarized image (binarized image), for example, the boundary is often not clearly shown due to, for example, variations in luminance. Therefore, it is difficult to recognize the boundary based on the binarized image. . In addition, in the binarized image of the image in which the alloy layer is imaged, there is an adverse effect such as recognizing the boundary of the precipitate due to the influence of the precipitate included in the alloy layer. A boundary recognition method corresponding to the above is required.

  SUMMARY OF THE INVENTION Accordingly, an object of the present invention is to provide a boundary detection method that can accurately detect a boundary between stacked different materials in an image of a cross section along a stacking direction of members configured by stacking different types of materials. And

In order to achieve the above-described object, the present invention takes the following technical means.
The boundary detection method according to the present invention is a boundary detection method for detecting a lamination boundary that is a boundary between the plurality of different materials with respect to a lamination material in which a plurality of different materials are laminated, and includes the lamination boundary. An image capturing step for capturing an image of the laminated material, and a feature amount detecting step for detecting a feature amount expressing a variation in luminance due to a difference in particle size distribution of the laminated material from the luminance distribution of the image captured in the imaging step If, have a, a laminated boundary detection step of detecting the laminated boundary in the captured image based on the feature quantity detected by said feature quantity detecting step, the feature quantity detecting step, as the feature quantity, A standard deviation and / or a peak count value of the luminance distribution in a direction along the stacking boundary is detected .

Here, before Symbol laminated boundary detection step, the calculated standard deviation and / or the feature amount similarity in the detected direction perpendicular to peak count value, the laminate based on the difference of the calculated similarity It is good to detect the boundary.

Further, the stack boundary detection step includes adding a mark indicating the stack boundary to a position where a feature amount indicating the difference is detected in the image based on the obtained difference in similarity. Should be detected.
Furthermore, the feature amount detection step may detect a differential value of the luminance distribution in a direction intersecting the stack boundary as the feature amount.

  According to the boundary detection method of the present invention, it is possible to accurately detect a boundary between stacked different materials in an image of a cross section along a stacking direction of members formed by stacking different types of materials.

It is a figure which shows the process sequence of the boundary detection method by this embodiment. It is a figure which shows the image which removed the noise from the image of the cross section of a laminated material. It is a figure which shows the graph showing a Mahalanobis distance and a boundary position, (A) shows the change of the Mahalanobis distance along the vertical direction (Y-axis direction) of an image, (B) is the vertical direction (Y-axis direction) of an image. The position of the flag raised along is shown. It is a figure which shows the image displayed with the boundary line superimposed on the cross section of the laminated material. It is a figure explaining the detection method of the lamination | stacking boundary by a prior art, (a) shows the image for a detection, (b) shows the image which binarized the image for a detection.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. In addition, embodiment described below is an example which actualized this invention, Comprising: The structure of this invention is not limited only to the specific example. Therefore, the technical scope of the present invention is not limited only to the disclosed contents of the present embodiment.
First, the outline of the boundary detection method according to the present embodiment will be described with reference to FIG. FIG. 1 is a diagram illustrating a processing procedure of the boundary detection method according to the present embodiment.

  The boundary detection method described below is a method of laminating a plurality of different types of different materials such as a metal member configured by stacking different types of metal materials and a resin member configured by stacking different types of resin materials. This is a technique for detecting a boundary (lamination boundary) between a plurality of different dissimilar materials laminated with respect to the laminated material formed. Specifically, in the boundary detection method, a cross section along a stacking direction of a member (hereinafter referred to as a stacking material) configured by stacking different types of materials is captured with an area camera or the like, and the layered images are stacked. The boundary between the different types of materials is detected using a personal computer (PC) or a dedicated image processing apparatus.

As shown in FIG. 1A, in the boundary detection method, the cross-sectional image 1 of the laminated material is such that the laminated surface that is the boundary between the laminated different materials is parallel to the horizontal direction (X-axis direction). Acquired and noise of the acquired image 1 is removed.
Next, as shown in FIG. 1B, the luminance distribution for each line in the horizontal direction (X-axis direction) along the laminated surface is obtained from the image 2 (detection image 2) of the cross section from which noise is removed. Extracted.

As shown in FIG. 1C, when the luminance distribution 3 is extracted, a peak count C (Yn) is detected as a feature amount representing the luminance variation in the luminance distribution 3 of each line.
Subsequently, when the peak count C (Yn) of each line is detected as shown in FIG. 1D, the Mahalanobis distance for indicating the similarity between the detected peak counts C (Yn) is detected. Is calculated for each peak count C (Yn). When the Mahalanobis distance is calculated, the calculated Mahalanobis distance is arranged in the order of corresponding positions along the vertical direction (Y-axis direction) of the image as shown in FIG.

Further, as shown in FIG. 1 (E), the boundary detection method of the present invention detects and detects a position where the Mahalanobis distance arranged in order along the vertical direction (Y-axis direction) of the image changes across the threshold. Flags (signs) with signs a to l, o and p are set at the positions. After the flag is raised, for example, in the image 2 of the cross section shown in FIG. 1B, the boundary (lamination surface) between different materials is along the horizontal direction (X-axis direction) at the position where the flag is raised. It is detected that it exists, and a boundary line is displayed in an overlapped manner in the image 2 as shown in FIG. In FIG. 1 (F), each of the boundary lines is provided with a corresponding flag code a to l, o, and p.

  The above is the outline of the boundary detection method according to the present embodiment. In the following, each step shown in FIGS. 1A to 1F will be described in detail with reference to FIGS. FIG. 2 is a diagram showing an image 2 obtained by removing noise from the cross-sectional image 1 of the laminated material. 3A and 3B are graphs showing the Mahalanobis distance and the boundary position. FIG. 3A shows a change in the Mahalanobis distance in the vertical direction (Y-axis direction) of the image, and FIG. 3B shows a vertical direction of the image (Y-axis). (Direction) indicates the positions of the flags a to l, o, and p. FIG. 4 is a diagram showing an image displayed with the boundary line superimposed on the cross section of the laminated material.

First, a method for obtaining the cross-sectional image 1 of the laminated material shown in FIG.
A cross section when the laminated material is cut along the lamination direction is captured as a two-dimensional image using an area camera or the like using a CCD (Charge Coupled Device) or CMOS (Complementary Metal Oxide Semiconductor). An image of 1360 pixels × vertical direction (Y-axis direction) 1024 pixels is obtained. At this time, a cross section of the laminated material is imaged so that the laminated surface, which is a boundary (laminated boundary) between the laminated different materials, is parallel to the horizontal direction (X-axis direction), and the laminated material shown in FIG. An image 1 of a cross section of the material is acquired. The process of capturing the image 1 of the laminated material including the laminated boundary is referred to as an imaging process. Each pixel (pixel) of the acquired image 1 corresponds to each coordinate in the XY coordinate system.

Next, with reference to FIG. 1B and FIG. 2, a method for extracting the luminance distribution from the captured cross-sectional image will be described. FIG. 2 is a diagram showing an image 2 obtained by removing noise from the cross-sectional image 1 of the laminated material.
The image 1 acquired in the imaging process is subjected to preprocessing for reducing image-derived noise. For example, image processing using a 3 × 3 median filter is performed on the image 1 acquired in the imaging process, and an image 2 (detection image 2) with less noise is obtained while maintaining the contour of the cross section.

In the detection image 2 shown in FIG. 2, for example, a detection line A (Yn) along the horizontal direction (X-axis direction) such as 1360 pixels in the horizontal direction (X-axis direction) × 1 pixel in the vertical direction (Y-axis direction). Set. In FIG. 2, the detection line A (Yn) set from the pixel position (0, Y) to the pixel position (X, Y) is indicated by a broken line. That is, for the detection image 2, 1024 (n = 0 to 1023) detection lines A (Y ₀ ) to A (Y ₁₀₂₃ ) corresponding to the number of pixels in the vertical direction (Y-axis direction). The detection images are sequentially set.

For each detection line A (Yn) set in this way, the luminance distribution 3 in the horizontal direction (X-axis direction) along the detection line A (Yn), that is, along the laminated surface is extracted, and is vertically The luminance distribution 3 for each of 1024 detection lines A (Yn) corresponding to the number of pixels in the direction (Y-axis direction) is extracted. The step of extracting the luminance distribution 3 is referred to as a luminance distribution extraction step.
With reference to FIG. 1C, a method of detecting the peak count C (Yn) as a feature quantity expressing the luminance variation in the luminance distribution 3 of each detection line A (Yn) will be described. In FIG. 1C, a waveform representing the luminance distribution 3 of one detection line A (Yn) and a straight line 4 representing the average value of the luminance distribution 3 are shown.

The waveform of the luminance distribution 3 in FIG. 1C expresses the variation in luminance caused by the difference in the particle size distribution of the laminated material, and intersects the straight line 4 indicating the average value a plurality of times, for example, 14 times. . This intersection occurs when the luminance value increases or decreases across the average value, and the number of intersections (for example, a value of 14) is defined as the peak count C (Yn). In the present embodiment, the peak count C (Yn) is detected for each of the 1024 detection lines A (Y ₀ ) to A (Y ₁₀₂₃ ) arranged in the vertical direction (Y-axis direction). Similar to the number of lines A (Yn), 1024 peak counts C (Yn) are detected as peak counts C (Y ₀ ) to C (Y ₁₀₂₃ ). As described above, the step of detecting the peak count C (Yn) as the feature amount of the luminance distribution 3 is referred to as a feature amount detection step.

As can be seen from FIG. 1C, the waveform of the luminance distribution 3 having the same average value as the average value shown in FIG. 1C can exist in addition to the waveform shown in FIG. Therefore, even if the waveform of the luminance distribution 3 has the same average value, the number of times of crossing the straight line 4 indicating the average value may be different if the shape is different. That is, the peak count C (Yn) is more than the average value of the luminance distribution 3.
It can be said that the value is suitable for reflecting the characteristic (difference) of the luminance distribution 3.

  In order to detect the above-mentioned peak count C (Yn), the average value of the luminance distribution 3 is adopted, but the average value is not necessarily adopted, and another value may be adopted. For example, as in the evaluation of surface properties, instead of using the average of one value, the range of average value ± 2σ is adopted as a threshold, and the number of times this range is crossed from the upper limit to the lower limit or from the lower limit to the upper limit is peaked. It may be used as the count C (Yn).

In addition to these values, it is also conceivable to use an average value of peak-to-peak intervals (distance between peaks) in the waveform of the luminance distribution 3 and variations in peak positions. As described above, simple luminance variations such as the standard deviation of the luminance distribution 3, information such as the peak count equivalent value, the distance between peaks, and the like can be employed as the feature amount of the luminance distribution 3.
The processing in the feature amount detection process described above is summarized as follows.

  The feature amount detection step is performed from the image 1 captured in the imaging step (that is, based on the luminance distribution 3 of the detection image 2 obtained by removing noise from the image 1 captured in the imaging step). A feature amount expressing a variation in luminance caused by a difference in particle size distribution is detected, and the standard deviation of the luminance distribution 3 and / or the value of peak count C (Yn) in the direction along the stacking boundary is detected as the feature amount. This is a process for detecting.

  Next, a description will be given of a layer boundary detection step of detecting a layer boundary from the detection image 2 using the feature amount of the luminance distribution 3 exemplified by the above-described peak count C (Yn). In the layer boundary detection step, the layer boundary is given to the detection image 2 shown in FIG. 2 through the calculation of the Mahalanobis distance for the peak count C (Yn) and the provision of the boundary position flag based on the calculated Mahalanobis distance. The Mahalanobis distance here is a kind of distance used in statistics, and is a generally well-known concept.

With reference to FIG. 1D and FIG. 3A, calculation of the Mahalanobis distance with respect to the peak count C (Yn) will be described. FIG. 3A is a diagram showing a change in Mahalanobis distance along the vertical direction (Y-axis direction) of the image.
The above-described peak count C (Yn) is detected by obtaining an average value (average value ± 2σ) of each luminance distribution 3 for each detection line A (Yn). However, in the calculation of the Mahalanobis distance in the present embodiment, one layer is selected from a plurality of layers in the detection image 2 as a reference layer, a reference point and a variance are calculated, and all peaks are calculated based on this reference point. The Mahalanobis distance of the counts C (Y ₀ ) to C (Y ₁₀₂₃ ) is obtained.

The Mahalanobis distance in this embodiment is calculated | required as follows.
First, one layer is selected as a reference layer from a plurality of layers stacked in the detection image 2 shown in FIG. 2, and an average value of peak counts C (Yn) corresponding to the selected reference layer is set to an average value P. The average of absolute values of deviation is set as _0, and the average of absolute values of deviation is Ps. Then, a one-dimensional simplified Mahalanobis distance MD (Yn) is defined as the following equation (1), and is defined as the Mahalanobis distance.

MD (Yn) = (C (Yn) −P ₀ ) / Ps (1)
After selecting the reference layer from the detection image 2 shown in FIG. 2 and calculating the average value P ₀ and the absolute value average Ps, all the peak counts C (shown in FIG. The Mahalanobis distances of Y ₀ ) to C (Y ₁₀₂₃ ) are obtained, and the obtained Mahalanobis distances MD (Y ₀ ) to MD (Y ₁₀₂₃ ) are sequentially arranged along the Y-axis (Y coordinate) direction. For each of the Mahalanobis distances MD (Y ₀ ) to MD (Y ₁₀₂₃ ) arranged in order, an average of three points using the preceding and following Mahalanobis distance MD (Y _n-1 ) and Mahalanobis distance MD (Y _{n + 1} ) is calculated. By plotting the obtained value as the Mahalanobis distance MD (Yn), the change in the Mahalanobis distance MD (Yn) along the Y-axis direction as shown in FIG. 3A can be represented in a graph.

Here, it should be noted that the horizontal axis of the graph shown in FIG. 3A corresponds to the Y axis (Y coordinate) that is the vertical axis of the inspection image 2 shown in FIG. That is, in the graph shown in FIG. 3A, features that appear along the horizontal axis direction correspond to features that appear along the vertical axis direction in the inspection image 2.
Referring to the graph shown in FIG. 3A, in the Y coordinate, the Mahalanobis distance MD (Yn) continuously changes around the value 10-11, and the interval continuously changes around the value 8-9. It can be said that the sections that continuously shift around the values 1 to 2 appear alternately. The section in which the Mahalanobis distance MD (Yn) transitions around the value 1 to 2 in the Y coordinate of FIG. 3A is the shortest distance from the reference point, and therefore is the same as the section corresponding to the reference layer or the reference layer It is considered that this is the section corresponding to the layer.

  In this way, by obtaining the Mahalanobis distance MD (Yn) with respect to the peak count C (Yn) that is the feature quantity of the luminance distribution 3, a section in which the Mahalanobis distance MD (Yn) shows a continuously close value is peaked. It is possible to extract as a section having high similarity of the count C (Yn). In other words, obtaining the Mahalanobis distance MD (Yn) is synonymous with obtaining the similarity of the peak count C (Yn). As described above, based on the obtained difference in similarity, a section having high similarity can be extracted and a stack boundary can be detected.

Hereinafter, a method for detecting a stacking boundary using a change in Mahalanobis distance MD (Yn) shown in FIG. 3A will be described with reference to FIGS. 1E, 3A, and 3B. To do. FIG. 3B is a diagram showing the position of the flag set up along the vertical direction (Y-axis direction) of the detection image 2.
In the change of the Mahalanobis distance MD (Yn) shown in FIG. 3A, for example, a flag (marker) with a value of 0 is set for each Y coordinate in a section where the Mahalanobis distance MD (Yn) is less than 3.5. Then, a flag (marker) with a value of 3 is set for each Y coordinate of the section where the Mahalanobis distance MD (Yn) is 3.5 or more.

  In addition, the flag is switched once from the value 0 to the value 3 or from the value 3 to the value 0 in the range of the continuous predetermined Y coordinate corresponding to the predetermined number of continuous pixels (about several pixels). In the range where only this occurs, the Y coordinate where the one-time switching has occurred is recognized as the stacking boundary, and the flag at that position is maintained at the value 3. In addition, in the range where the switching of the flag occurs a plurality of times, the barycentric positions of the plurality of Y coordinates where the switching has occurred are obtained, the Y coordinate corresponding to the obtained barycentric position is recognized as the stacking boundary, and the flag at that position is set as the value. 3 is maintained. Thus, after recognizing the stacking boundary, a flag at a position other than the position of the stacking boundary (boundary position) is set to 0.

Through the above processing, a graph shown in FIG. 3B in which a flag of value 3 is given only to the recognized boundary position is obtained.
Thus, based on the magnitude (that is, the difference) of the Mahalanobis distance MD (Yn) indicating the similarity of the peak count C (Yn) that is the feature quantity, the Mahalanobis distance MD is detected in the detection image 2 shown in FIG. A flag (label) indicating the position (boundary position) of the stacking boundary can be added to the position where the peak count C (Yn) indicating the magnitude (difference) of (Yn) is detected, and the stacking boundary can be detected. it can.

With reference to FIG. 1 (F) and FIG. 4, a method for giving a layer boundary to the detection image 2 shown in FIG. 2 will be described. FIG. 4 is a diagram showing an image displayed with the boundary line superimposed on the cross section of the laminated material.
The flags a to l, o, and p shown in the graph of FIG. 3B indicate the position of the stacking boundary on the vertical axis (Y axis) of the detection image 2 shown in FIG. In the detection image 2 shown, a boundary line is given along the horizontal axis (X axis) to the position of the Y coordinate indicated by the flags a to l, o, and p. As a result, an image 5 in which boundary lines corresponding to the flags a to l, o, and p are added to the detection image 2 as shown in FIG. 4 can be obtained, and each layer stacked based on the boundary line intervals. Can be detected.

  The operation of the layer boundary detection step is as described above. For the boundary detection method including the above-described imaging step, luminance distribution extraction step, feature amount detection step, and layer boundary detection step, for example, a personal computer (PC) or a dedicated arithmetic device By executing the computer program, it is possible to recognize and detect a boundary that cannot be detected by the conventional technique simply using the average value of the luminance in the detection line A (Yn).

Here, in the boundary detection method described above, a method for determining the type of material by tabulating the feature value values, or calculating the difference of the feature value in the image vertical direction (Y-axis direction), and a position where there has been a significant change Can be applied. In addition, when considering variations in feature values, by creating a numerical index based on a plurality of parameters such as Mahalanobis distance in the space including other parameters such as the average value of luminance in addition to the feature value, and using a threshold value, A method for recognizing a specific stacking boundary can also be considered.

Here, with reference to FIG. 5, a conventional method for detecting a stacking boundary will be described as a reference.
5A and 5B are diagrams for explaining a conventional method for detecting a layer boundary, in which FIG. 5A shows a detection image 6 and FIG. 5B shows an image obtained by binarizing the detection image 7.
The detection image 6 shown in FIG. 5A is an image obtained by capturing a cross section of a laminated material of aluminum alloy and averaging the captured image for 40 pixels in the horizontal direction. The binarized image 7 shown in FIG. 5B was obtained by binarizing the detection image 6 in FIG. 5A using the average value + standard deviation of the entire image as a threshold value. In the binarized image 7 shown in FIG. 5B, it is difficult to recognize the layer boundary at the top of the image due to uneven brightness in the image.

Since it is difficult to stably detect the stack boundary based on only the brightness value and the average value of brightness as in the prior art, the brightness change and brightness distribution as in the boundary detection method of the present invention are difficult. If the feature amount is used, the stack boundary can be detected stably and reliably.
By the way, it should be thought that embodiment disclosed this time is an illustration and restrictive at no points. In particular, in the embodiment disclosed this time, matters that are not explicitly disclosed, such as operating conditions and measurement conditions, various parameters, dimensions, weights, volumes, and the like of a component deviate from a range that is normally implemented by those skilled in the art. Instead, values that can be easily assumed by those skilled in the art are employed.

  Specifically, in the feature amount detection step, a differential value that is a change amount of the luminance distribution in the direction intersecting the stack boundary may be detected and used as the feature amount.

1,5 image 2,6 image for detection 3 luminance distribution 4 straight line (average value)
7 Binary image

Claims (4)

A boundary detection method for detecting a lamination boundary, which is a boundary between the plurality of different materials, for a laminated material in which a plurality of different materials are laminated,
An imaging step of capturing an image of the laminated material including the laminated boundary;
A feature amount detection step for detecting a feature amount representing a variation in luminance due to a difference in particle size distribution of the laminated material from the luminance distribution of the image captured in the imaging step;
Have a a laminated boundary detection step of detecting the laminated boundary in the captured image based on the feature quantity detected by said feature quantity detecting step,
The feature detection step detects a standard deviation and / or a peak count value of the luminance distribution in a direction along the stack boundary as the feature. In the stacking boundary detection step, similarity of the feature amount is obtained in a direction perpendicular to the direction in which the standard deviation and / or peak count value is detected, and the stacking boundary is detected based on the obtained difference in similarity. The boundary detection method according to claim 1 . The stacking boundary detection step detects the stacking boundary by adding a label indicating the stacking boundary to a position where the feature amount indicating the difference is detected in the image based on the obtained difference in similarity. The boundary detection method according to claim 2 , wherein:   The boundary detection method according to claim 1, wherein the feature amount detection step detects a differential value of the luminance distribution in a direction intersecting with the stack boundary as the feature amount.