JP5413283B2 - Focus shift detection device, focus shift detection method and program - Google Patents

Description

  The present invention relates to a focus shift detection apparatus, a focus shift detection method, and a program, and more particularly, to a technique suitable for application to an optical system of an image pickup apparatus that picks up an image of a surface in order to detect a defect of an object to be inspected.

  Conventionally, various techniques (inspection processing) for inspecting defects on the surface of an object to be inspected are known. For example, the surface of the object to be inspected is imaged using an imaging device to acquire image data, and the difference between the value obtained by averaging the luminance of the acquired image data and the luminance of the original image data is calculated. A method for detecting a defect based on the comparison result between the threshold value and the threshold value has been implemented.

  In such a surface defect inspection apparatus that performs flaw detection on the surface of an object to be inspected, a defect is detected from an image captured by the imaging apparatus. Therefore, in order to detect a defect with high accuracy, the focus of the optical system of the imaging apparatus is covered. It is important to match the surface of the specimen. However, the mechanical vibration in the surface defect inspection device causes the lens of the imaging device to deviate from the in-focus position (the lens position when the focus of the imaging device matches the surface of the object to be inspected) or during maintenance. By moving the imaging apparatus, the lens position of the imaging apparatus may be moved from the in-focus position. For this reason, it is necessary to confirm whether or not the imaging apparatus is in focus, and when the focus is out of focus, it is necessary to perform focus adjustment work for the imaging apparatus.

  As described above, as an apparatus for detecting whether or not the imaging apparatus is in focus, for example, Patent Document 1 discloses that a predetermined inspection pattern is used to calculate a sum of absolute values of differences between adjacent pixels and an average value of pixels. An optical inspection apparatus that periodically determines whether or not the camera is in focus based on the ratio is disclosed. Patent Document 2 discloses a method of focusing the camera apparatus based on the maximum point calculated using the sum of absolute difference values of adjacent pixels in the autofocus camera apparatus.

JP 2002-318343 A Japanese Patent Laid-Open No. 06-113187

  Patent Document 1 calculates an evaluation value for the same inspection pattern, and Patent Document 2 calculates an evaluation value for the same subject to determine the degree of focusing. However, during normal flaw detection in a steel rolling line or the like, the surface of the object to be inspected is imaged. In this case, since the same subject is not photographed, the evaluation values as in Patent Documents 1 and 2 cannot be used. This is because when the image data to be evaluated changes, the evaluation value changes greatly, so that the evaluation values cannot be compared between different image data. Therefore, in these methods, even if the focus of the imaging device is deviated during flaw detection, this cannot be detected. For this reason, when the deviation of the imaging device increases with time due to vibration or the like, the detection accuracy of the surface defect inspection device deteriorates, and a defective product flows out.

  Therefore, the present invention has been made in view of the above problems, and an object of the present invention is to detect a focus shift of an imaging apparatus even during flaw detection, and to detect defective products due to a deterioration in determination accuracy. It is an object of the present invention to provide a new and improved focus shift detection device, focus shift detection method and program capable of preventing outflow and the like.

In order to solve the above-described problem, according to an aspect of the present invention, in a surface defect inspection apparatus that inspects a surface defect based on image data obtained by imaging the surface of an object to be inspected by an imaging apparatus, the imaging apparatus There is provided a focus shift detection device for detecting a focus shift. Such a defocus detection apparatus includes a calculation unit that performs a conversion operation for converting an image signal of image data into two-dimensionally arranged frequency components, and a two-dimensional array of frequency components of image data based on a predetermined rule. An array conversion unit that converts to a dimensional array, and a focus shift determination unit that determines the presence or absence of a focus shift of the imaging device based on a slope of a waveform formed by a part of the frequency component converted to the one-dimensional array. And an arithmetic unit that divides the image data into blocks of a predetermined size and converts each block generated by dividing the image data into two-dimensionally arranged frequency components using discrete cosine transform. For each two-dimensionally arranged frequency component, the absolute value of the two-dimensionally arranged frequency component of each block is added and the average for all blocks is calculated to calculate the average frequency component absolute value. A CT operation unit, and the array conversion unit reads the image data converted into the two-dimensionally arranged frequency components by performing zigzag scanning in order from the low frequency component to the high frequency component, and converts it into a one-dimensional array. Then, logarithm values of the average frequency component absolute values of the image data are arranged in the reading order of the frequency components when converting from the two-dimensional array of image data to the one-dimensional array, and a waveform for calculating the slope is formed. The focus shift determination unit calculates the slope of the waveform by linearly approximating the frequency component converted into the one-dimensional array .

  The focus shift detection device of the present invention converts an image signal of image data captured by an imaging device into two-dimensionally arranged frequency components, and further converts the frequency components into data that is converted into a one-dimensional array based on a predetermined rule. Based on this, it is determined whether or not there is a focus shift of the imaging apparatus. Thereby, even when the image data for evaluating the degree of focus shift changes, the focus shift of the imaging apparatus can be detected.

  The focus shift determination unit may calculate the slope of a waveform formed using a lower frequency component among the frequency components converted into a one-dimensional array.

  Further, the focus shift determination unit may calculate the slope of the waveform excluding the DC frequency component when converting the two-dimensional array of image data to the one-dimensional array.

  The focus shift determination unit determines that there is no focus shift of the imaging device when the absolute value of the tilt is smaller than a predetermined threshold, and determines that there is focus shift of the image capture device when the absolute value of the tilt is equal to or greater than the predetermined threshold. Can do.

In order to solve the above problems, according to another aspect of the present invention, there is provided a surface defect inspection apparatus for inspecting surface defects based on image data obtained by imaging the surface of an object to be inspected by an imaging apparatus. A focus shift detection method for detecting a focus shift of an imaging apparatus is provided. In such a defocus detection method, a calculation step for performing a conversion calculation for converting an image signal of image data into a two-dimensionally arranged frequency component and a two-dimensional arrangement of the frequency component of the image data are based on a predetermined rule. An array conversion step for converting into a dimensional array, and a focus shift determination step for determining the presence or absence of a focus shift of the imaging device based on a slope of a waveform formed by a part of the frequency component converted into a one-dimensional array. seen including, calculation step, converts an image dividing step of dividing the image data into blocks of a predetermined size, each block generated by dividing the image data, a two-dimensional array of frequency components using a discrete cosine transform For each two-dimensionally arranged frequency component, the absolute value of the two-dimensionally arranged frequency component of each block is added to obtain an average for all blocks. A DCT calculation step for calculating an absolute value of the average frequency component, and in the array conversion step, the image data converted into the two-dimensionally arranged frequency component is zigzag scanned in order from the low frequency component to the high frequency component. Read, convert to one-dimensional array, arrange logarithm of absolute value of average frequency component of image data in order of reading frequency component when converting from two-dimensional array of image data to one-dimensional array, A waveform for calculation is formed, and in the focus shift determination step, the slope of the waveform is calculated by linearly approximating the frequency component converted into a one-dimensional array .

  Furthermore, in order to solve the above-described problems, according to another aspect of the present invention, a program for causing a computer to function as the above-described defocus detection apparatus is provided. Such a program is stored in a storage device included in the computer, and is read and executed by a CPU included in the computer, thereby causing the computer to function as the focus shift detection device. A computer-readable recording medium on which the program is recorded is also provided. The recording medium is, for example, a magnetic disk or an optical disk.

  As described above, according to the present invention, the focus shift of the imaging apparatus can be detected even during flaw detection, and the focus shift detection apparatus and focus that can prevent the outflow of defective products due to the deterioration of the determination accuracy. A deviation detection method and program can be provided.

It is explanatory drawing which shows the typical steel rolling line to which the defect inspection process which concerns on embodiment of this invention is applied. It is a flowchart which shows a defect inspection process. It is a block diagram which shows the function structure of the focus shift | offset | difference detection means of the image processing apparatus which concerns on the embodiment. It is a flowchart which shows the focus shift detection method by a focus shift detection means. It is explanatory drawing which shows the focus shift detection method by a focus shift detection means. It is explanatory drawing which shows the calculation process of the DCT coefficient using the two-dimensional discrete cosine transform by a DCT calculating part. It is a graph which shows the relationship between the inclination of the straight line calculated by the focus shift | offset | difference determination part and the focus shift | offset | difference when changing the data number used for the straight line approximation. It is explanatory drawing which shows the graph showing the relationship between the index with respect to original image data, and the logarithm of the average of the absolute value of a DCT coefficient, the straight line obtained by carrying out the linear approximation of the said waveform, and the image which is verification object. . A graph representing a waveform indicating the relationship between an index for image data having a degree of defocus of “1” and the average logarithm of the absolute value of the DCT coefficient, and a straight line obtained by linear approximation of the waveform, and a verification target It is explanatory drawing which shows the image which is. A graph representing a waveform indicating the relationship between an index for image data with a degree of defocus of "2" and the average logarithm of the absolute value of the DCT coefficient, and a straight line obtained by linear approximation of the waveform, and a verification target It is explanatory drawing which shows the image which is. 2 is a block diagram illustrating an example of a hardware configuration of an image processing apparatus according to the embodiment. FIG.

  Exemplary embodiments of the present invention will be described below in detail with reference to the accompanying drawings. In addition, in this specification and drawing, about the component which has the substantially same function structure, duplication description is abbreviate | omitted by attaching | subjecting the same code | symbol.

<1. Outline of defect inspection in steel rolling line>
[Schematic configuration of steel rolling line]
First, a schematic configuration of a steel rolling line to which a defect inspection process according to an embodiment of the present invention is applied will be described with reference to FIG. In addition, FIG. 1 is explanatory drawing which shows the typical steel rolling line to which the defect inspection process which concerns on this embodiment is applied.

  The steel rolling line includes an unwinding reel 10 that is a starting point of the line, a winding reel 50 that is an end point, a transport roll 30 that advances a steel plate 20 that is an object to be inspected between these reels, and a rolling that performs a rolling process. And roll 40. Between the rolling roll 40 and the take-up reel 50, an illumination 60 for emitting light to the surface of the steel plate 20 and a camera 70 which is an imaging means for imaging the surface of the steel plate 20 are arranged. The camera 70 is provided so that light emitted from the illumination 60 and reflected by the surface of the steel plate 90 is input to the camera 70. Moreover, the camera 70 is provided so that the relative position with respect to the surface of the steel plate 20 which is a to-be-inspected object is always constant. The camera 70 is connected to the image memory 80 and outputs the captured image to the image memory 80. An image memory 80 and a display device 100 are connected to the image processing apparatus 90 that performs defect inspection processing.

  The strip-shaped steel sheet 20 sent out from the unwinding reel 10 is advanced in the direction of the arrow in FIG. 1 by the transport roll 30 and rolled by the rolling mill 40, and then taken up by the take-up reel 50. On the other hand, the defect inspection of the surface of the steel plate 20 is performed based on image data obtained by applying light to the rolled steel plate 20 using the illumination 60 and capturing the surface with the camera 70. The captured image data records the luminance of each pixel, and the image is displayed as an aggregate. The image data is stored in the image memory 80, and after the defect inspection process is performed in the image processing device 90, the inspection result is displayed on the display device 100.

[Overview of defect inspection process]
Next, a procedure of defect inspection processing performed by the image processing apparatus 90 will be described based on FIG. FIG. 2 is a flowchart showing the defect inspection process.

  First, when the image processing apparatus 90 takes in the image data of the surface of the object to be inspected stored in the image memory 80 (step S10), the image processing apparatus 90 performs normalization processing of the luminance of the image data by shading correction or the like (step S20). The normalization process is performed by subtraction with image data without defects or division with image data without defects as a denominator for the purpose of correcting illumination unevenness. Alternatively, the captured image data may be smoothed by a low-pass filter to extract only the luminance unevenness component, and normalization processing may be performed by performing subtraction or division. Subtraction or division in the normalization process is performed for each pixel. When the normalization process is subtraction, the normalization reference value is 0, and when it is division, the normalization reference value is 1. By performing such normalization processing, luminance normalized image data can be acquired.

  Next, the image processing device 90 binarizes the luminance normalized image data based on a predetermined threshold value, and acquires binarized image data (step S30). The predetermined threshold value is a value for specifying a portion to be regarded as a wrinkle that is experimentally set with a wrinkle sample or the like from the difference in luminance between the wrinkle portion on the surface of the steel plate and a healthy portion. When the normalized luminance exceeds a predetermined threshold range for each pixel of the luminance normalized image data, the image processing apparatus 90 determines that there is a defect candidate, and the normalized luminance falls within the predetermined threshold range It is determined that there is no defect candidate. As a result, binary image data having a value indicating whether each pixel has a wrinkle candidate or no wrinkle candidate is generated from the luminance normalized image data.

  Furthermore, the image processing apparatus 90 treats the adjacent wrinkle candidate portion as one group and the separated wrinkle candidate portions as different groups for the binarized image data. A labeling process is performed (step S40). The labeling process is a process of assigning a label number to each group in order. The image processing apparatus 90 treats the pixels of the binarized image with the wrinkle candidate adjacent to each other from the binarized image data as one defect group, and assigns consecutive label numbers to the separated defect groups. It will be attached in order. For example, the label numbers may be assigned in order of priority from left to right and from top to bottom of the image. Thereby, label image data to which a label number is assigned for each defect group is generated from the binarized image data.

  Thereafter, the image processing apparatus 90 extracts the feature of the defect from the label image data (Step S50). The defect features include, for example, the size and shape of the wrinkles. In this case, the area and existence area of the defect group may be extracted from the label image data. For example, a method of counting the number of pixels with the same label number using a method based on a labeling image may be used.

  Next, the defect is determined based on the feature amount obtained by the feature extraction of the defect (step S60). For example, a dendritic theory or a neural network can be used for determining the defect. As the defect types to be determined, for example, in the case of a defect in a steel plate, Scab, Blowhole, Sliver, Skin Inclusion, Scale, Scale, Gouge Etc.

  When determining the defect, the image processing apparatus 90 displays a determination result indicating the defect type or the like on the display device 100 (step S70). The output destination of the determination result may be an output device to some medium such as a printing device or a disk recording device, or a device for performing another process such as an analysis device or a transmission device. May be.

  In this manner, defects such as wrinkles generated on the surface of the steel plate 20 can be detected using the image data obtained by capturing the steel plate 20 conveyed by the steel rolling line with the camera 70.

<2. Surface inspection process>
Since the defect inspection process for the surface of the steel plate 20 being conveyed by the steel rolling line is performed based on the image data captured by the camera 70, it is possible to detect wrinkles and the like when the image data is out of focus. As a result, the accuracy of defect determination is reduced. In this case, there is a possibility that the defect cannot be detected although there is a defect, leading to the outflow of defective products. Therefore, the image processing apparatus 90 according to the present embodiment includes a focus shift detection unit that detects that the lens position of the camera 70 has shifted from the in-focus position. In the following, the focus shift detection means of the image processing apparatus 90 and the focus shift detection method using this will be described in detail with reference to FIGS.

[Configuration of defocus detection means]
First, the defocus detection unit of the image processing apparatus 90 according to the present embodiment will be described with reference to FIG. FIG. 3 is a block diagram illustrating a functional configuration of the focus shift detection unit of the image processing apparatus 90 according to the present embodiment. In FIG. 3, only elements constituting the focus shift detection unit in the image processing apparatus 90 are described, and description of elements that perform the above-described defect inspection processing and the like is omitted.

  As shown in FIG. 3, the defocus detection unit includes an image capturing unit 91, an image dividing unit 92, a DCT calculation unit 93, an array conversion unit 94, a defocus determination unit 95, and an output unit 96. Composed. In the present embodiment, the image dividing unit 92, the DCT calculating unit 93, and the array converting unit 94 array the two-dimensional spatial frequency distribution of the image data in one dimension using a discrete cosine transform (DCT; Discrete Cosine Transform). Processing is performed.

  The image capturing unit 91 captures image data used to determine whether or not the lens position of the camera 70 is shifted from the in-focus position from the image memory 80 (see FIG. 1). The image memory 80 stores image data captured by the camera 70. The image capturing unit 91 captures image data stored in the image memory 80 at a predetermined timing, and outputs the captured image data to the image dividing unit 92. The image data capture timing can be, for example, the timing for starting the focus shift determination process.

  The image dividing unit 92 divides the image data into a plurality of blocks composed of n × n pixels (n is a positive integer). At this time, the image dividing unit 92 determines the division position of the image data so that the blocks do not overlap. When the image division unit 92 determines the division position of the image data, the image division unit 92 instructs the DCT calculation unit 93 to execute a discrete cosine transform process.

  The DCT calculation unit 93 calculates DCT coefficients that are two-dimensionally arranged frequency components for each block generated by dividing the image data using discrete cosine transform. The discrete cosine transform is a technique for converting an image signal into a spatial frequency component, and is widely used for signal compression processing. The DCT calculation unit 93 calculates n × n DCT coefficients for each block of image data from an image signal that is information of each pixel using a basis vector of two-dimensional discrete cosine transform. Then, the DCT calculation unit 93 takes the absolute value of the DCT coefficient of each block and calculates the average of all blocks from the absolute value of the DCT coefficient for the n × n frequency components, thereby obtaining the average frequency component absolute value. Obtain the average DCT coefficient absolute value. After calculating the average DCT coefficient absolute value of the image data, the DCT calculation unit 93 instructs the array conversion unit 94 to execute an array conversion process for converting the two-dimensional spatial frequency distribution into a one-dimensional array.

  The array conversion unit 94 converts the two-dimensional spatial frequency distribution of the image data into a one-dimensional array. The array conversion unit 94 rearranges the two-dimensionally arrayed average DCT coefficient absolute values into a one-dimensional array according to a predetermined rule. As the predetermined rule, for example, a zigzag scan in which zigzag scanning is performed from a low frequency component to a high frequency component can be used. By converting a two-dimensional array to a one-dimensional array by zigzag scanning, the amount of change in frequency components between adjacent data can be reduced, so the waveform shape of the data obtained by converting to a one-dimensional array can be made smooth. Can do.

  Further, the array conversion unit 94 converts the two-dimensional spatial frequency distribution of the image data into a one-dimensional array, and then takes a logarithm of the index representing the scan order of each frequency component and the average DCT coefficient absolute value (average DCT). A graph representing the relationship with the coefficient logarithm value) is created. When creating the graph, the array conversion unit 94 instructs the focus shift determination unit 95 to execute the focus shift determination process.

  The defocus determination unit 95 determines whether the lens position of the camera 70 is deviated from the in-focus position based on the graph created by the array conversion unit 94. The defocus determination unit 95 linearly approximates the graph using data of frequency components within a predetermined range from the graph created by the array conversion unit 94. For the linear approximation, for example, a general method such as a least square method can be used. The defocus determination unit 95 compares the absolute value of the slope of the straight line obtained by linearly approximating the graph with a predetermined threshold value, and determines whether the lens position of the camera 70 is shifted based on the magnitude relationship. The determination is made, and the determination result is output to the output unit 96.

  The output unit 96 outputs the determination result input from the focus shift determination unit 95 to the display device 100 in order to notify the user. The output unit 96 can also output image data acquired from the image memory 80 and information such as a graph created by the array conversion unit 94 to the display device 100 for display.

  The configuration of the focus shift detection unit of the image processing apparatus 100 according to the present embodiment has been described above.

[Defocus detection method]
Next, a focus shift detection method by the focus shift detection means of the image processing apparatus 90 will be described with reference to FIGS. FIG. 4 is a flowchart showing a focus shift detection method by the focus shift detection means. FIG. 5 is an explanatory diagram showing a focus shift detection method by the focus shift detection means. FIG. 6 is an explanatory diagram showing a DCT coefficient calculation process using the two-dimensional discrete cosine transform by the DCT calculation unit 93. FIG. 7 is a graph showing the relationship between the slope of the straight line calculated by the focus shift determination unit 95 and the degree of focus shift when the number of data used for the linear approximation is changed.

  In the focus shift detection method by the focus shift detection means, as shown in FIG. 4, first, the image capturing section 91 captures image data from the image memory 80 into the image processing apparatus 90 (step S100). In the image memory 80, a frame image or a cut-out image of an image captured by the camera 70 is stored as image data. For example, the image data may be taken in at the timing when the defocus determination process is started, or may be performed when the image data is recorded in the image memory 80.

  When an image is captured from the image memory 80, the image dividing unit 92 divides the image data (step S102). In such a step, the image dividing unit 92 divides the image data into blocks composed of n × n pixels so that the blocks do not overlap (wrap). For example, as shown in FIG. 5, when the image data is divided into blocks of 8 × 8 pixels, the image data of 640 × 480 pixels is divided into 80 × 60, that is, 4800 blocks.

  When the division position of the image data is determined, the DCT calculation unit 93 performs discrete cosine transform for each block of the image data (step S104). The DCT calculation unit 93 performs a conversion process using the basis vector of the two-dimensional discrete cosine transform shown in FIG. 6, and calculates a DCT coefficient. In the two-dimensional discrete cosine transform, an inner product with a base vector of the two-dimensional discrete cosine transform is calculated for one block composed of n × n pixels. The value of the inner product is a DCT coefficient. At this time, each basis of the basis vector has the same size as the block size (ie, n × n). Since the discrete cosine transform is an orthogonal transform, there are n × n coefficients after conversion, and therefore n × n DCT coefficients are calculated for one block. For example, as shown in FIG. 5, 64 DCT coefficients are calculated for a block composed of 8 × 8 pixels. Here, the (0, 0) component is a direct current component, and the other components are alternating current components. As described above, the DCT calculation unit 93 calculates a DCT coefficient for each block of the image data.

  Next, the DCT calculator 93 calculates the absolute value of the DCT coefficient of each block (step S106), and further calculates the average of the absolute values of the DCT coefficients of all blocks (step S108). The DCT calculation unit 93 takes absolute values for the n × n DCT coefficients calculated for each block. Then, the DCT calculation unit 93 adds the absolute values of the DCT coefficients in the same frequency component for all the blocks of the image data, takes an average, and calculates the average DCT coefficient. For example, the absolute value of the average DCT coefficient in the (0, 0) component is obtained by taking and adding the absolute value of the (0, 0) component of the DCT coefficients of each block and dividing the added value by the total number of blocks. Can be calculated. Then, the DCT calculation unit 93 calculates the absolute value of the DCT coefficient of each block, and finally calculates the average DCT coefficient absolute value of all n × n blocks.

  Thereafter, the array conversion unit 94 arranges the average DCT coefficient absolute values of all the n × n blocks in a line to form a one-dimensional array (step S110). The rearrangement from the two-dimensional array to the one-dimensional array is performed by zigzag scanning, for example, as shown in FIG. In the zigzag scan, average DCT coefficient absolute values arranged in n × n are arranged in a line in order from a low frequency component (upper left) to a high frequency component (lower right) to form a one-dimensional array. At this time, the index in which the reading order of frequency components by zigzag scanning is assigned in order from 1 is used.

  Then, the array conversion unit 94 creates a graph in which the horizontal axis represents the index and the logarithm of the average DCT coefficient absolute value (average DCT coefficient logarithmic value) represents the vertical axis (step S112). From this graph, it is possible to recognize the change in the absolute value of the average DCT coefficient based on the change in the frequency component. As long as it is not an artificially generated lattice pattern or the like, the waveform of this graph has an average DCT coefficient logarithmic value as the index value increases, that is, from a low frequency component to a high frequency component, as shown in FIG. It tends to become gradually smaller.

  Next, the focus shift determination unit 95 calculates the inclination of the graph created by the array conversion unit 94 in order to determine whether the lens position of the camera 70 is shifted from the in-focus position (step S114). The inclination of the graph can be obtained by a linear approximation method such as a least square method. Here, the slope of the graph is calculated using data of low frequency components among n × n pieces of data constituting the graph.

  The number of data used for the linear approximation is preferably the number that can acquire the slope of the straight line that can recognize the deviation of the camera 70, and the maximum number of data is preferably about 1/2 to 1/3 of the total number of data. In general, typical characteristics of an image are included in a low frequency component (that is, data having a small index value), and a component having a high frequency component (that is, a large index value) is originally small in value. There are large fluctuations. For this reason, if data with a large index value is included, the variation in the slope of the straight line increases, and the estimation accuracy of the deviation amount deteriorates.

  For example, in a graph composed of 8 × 8 (64) data, it is preferable to use less than 30 data from low frequency component data. FIG. 7 shows the relationship between the inclination of the graph and the degree of deviation of the lens position from the in-focus position (degree of focus deviation) when the number of data used for linear approximation is changed. The total number of data is 64. Here, the degree of defocus is a quantitative expression of the degree of defocus of image data captured by the camera 70, and the larger the value, the larger the defocus. The degree of defocus “0” is a state where the lens position is not deviated from the in-focus position.

  As shown in FIG. 7, the relationship between the degree of defocus and the inclination of the graph changes depending on the number of data used for linear approximation. For example, when the number of data used for linear approximation is 10, the absolute value of the slope of the graph tends to increase as the degree of focus shift increases. On the other hand, when the number of data used for linear approximation is 20 or more, the absolute value of the slope of the graph increases as the degree of defocus increases, but as the degree of defocus further increases, the absolute value of the slope of the graph increases. The value decreases again. Therefore, it is preferable to determine the number of data used for the linear approximation in accordance with the degree of focus shift to be detected by the focus shift detection means.

  For example, when it is desired to detect a case where the degree of defocus is “2”, the absolute value of the inclination of the graph is not shifted from the in-focus position as the degree of defocus increases. The number of data is greater than the absolute value of the slope of the graph when the degree of defocus is “0”. In the example of FIG. 7, the number of data used for linear approximation may be set to a value smaller than 30.

  Further, data having an index of 1, that is, data of a direct current component (0, 0) is excluded from data that is linearly approximated. This is because the data of the DC component (0, 0) does not include information on the lens position shift. By excluding such data from the data used for the linear approximation, the lens position shift of the camera 70 is eliminated. Can be detected accurately.

  When the absolute value of the inclination of the graph is calculated in step S114, the focus shift determination unit 95 determines whether or not the absolute value of the inclination is smaller than a predetermined threshold Th (step S116). The focus shift detection method in this embodiment uses the tendency that the absolute value of the slope of the straight line increases as the degree of focus shift increases, as shown in the graph of FIG. It is determined whether or not a deviation has occurred. The predetermined threshold Th is, for example, an intermediate value between the absolute value of the slope of the graph when no focus deviation occurs and the absolute value of the slope of the graph calculated from the image data when the focus deviation occurs. can do. Or it is good also as an absolute value of the inclination of the graph computed based on image data in permissible judgment accuracy as image data used for focus shift judgment processing.

  In the present embodiment, since the slope of the graph calculated in step S114 is a negative value, focus shift determination processing is performed using the absolute value of this slope. Therefore, the absolute value of the slope of the graph increases as the degree of defocus increases. From this, when the absolute value of the slope of the graph is smaller than the threshold value Th, it is determined that the influence on the accuracy of wrinkle detection is within an allowable range, and normal flaw detection processing is continued (step S118).

  On the other hand, when the absolute value of the inclination is equal to or greater than the threshold value Th, the focus shift determination unit 95 has a shift in the lens position of the camera 70 that affects the accuracy of wrinkle detection on the surface of the steel plate 20. The determination is made and an abnormality notification is sent to the display device 100 (step S120). For example, a message indicating that the lens position of the camera 70 may be shifted from the in-focus position is displayed on the display device 100. When the user confirms the message on the display device 100, the user adjusts the lens position of the camera 70 to eliminate the focus shift of the image. As described above, it is possible to detect the case where the lens position of the camera 70 is shifted during the normal flaw detection, and to prevent the outflow of defective products due to the decrease in the determination accuracy of the wrinkle detection.

<3. Simulation results>
A simulation for verifying the effect of detecting the lens position shift of the camera 70 by the image processing apparatus 90 including the focus shift detection unit according to the present embodiment described above was performed. In such a simulation, low-pass filter processing was performed on image data (original image data) captured by the camera 70 in which no defocusing occurred, and image data in which a focus shift occurred in a pseudo manner was generated. Then, for the original image data and the generated image data, a graph was created by arranging frequency components from a two-dimensional array to a one-dimensional array using a focus shift detection method, and the inclination of the graph was calculated.

  The simulation results are shown in FIGS. 8 to 10, a graph showing a relationship between an index for each image data and an average DCT coefficient logarithm, a graph showing a straight line obtained by linearly approximating the waveform, an image to be verified, Is shown. FIG. 8 shows original image data (that is, a state where there is no focus deviation), FIG. 9 shows image data whose degree of focus deviation is “1”, and FIG. 10 shows image data whose degree of focus deviation is “2”. The verification result is shown. In this simulation, linear approximation was performed using data having a zigzag scan index of 2 to 20 out of a total of 64 data.

  In a state where there is no focus deviation, as shown in FIG. 8, the slope of the graph obtained by linearly approximating the waveform indicating the relationship between the index and the average DCT coefficient logarithm was about −0.01. Similarly, when calculation is performed for image data with a degree of defocus of “1”, the slope of the graph is about −0.0256 as shown in FIG. When calculating for image data with a degree of defocus of “2”, the slope of the graph was about −0.0538 as shown in FIG.

  In such a simulation, when the image data used for determining whether or not a focus shift has occurred, the inclination of the graph calculated by the image processing apparatus 90 is larger than the tilt when no focus shift has occurred. As it turns out, it gradually decreases (increases as an absolute value) and changes about twice. Accordingly, the threshold value Th of the graph inclination for detecting the focus shift can be set to, for example, -0.02. In general, the threshold value Th is an intermediate between the inclination of the graph calculated based on the original image data when no focus deviation occurs and the inclination of the graph calculated based on image data when the focus deviation occurs. A value may be used.

  In this way, the two-dimensional spatial frequency distribution of the image data that is out of focus is converted into a one-dimensional array by zigzag scanning, and linear approximation is performed using data of the low frequency portion of the one-dimensionally arranged data. It can be seen that the degree of defocus can be determined from the calculated slope of the graph. Therefore, by using the image processing apparatus 90 according to the present embodiment, it is possible to detect whether or not a lens position shift of the camera 70 has occurred.

<4. Hardware configuration example>
The image processing apparatus 90 according to the present embodiment described above can be realized by an information processing apparatus 200 such as a computer, for example. As shown in FIG. 11, for example, the information processing apparatus 200 includes a CPU (Central Processing Unit) 201, a ROM (Read Only Memory) 202, a RAM (Random Access Memory) 203, a bus 204, an interface 205, An input device 206, an output device 207, a storage device 208, and a communication device 209 are provided.

  The CPU 201 functions as an arithmetic processing device and a control device, and controls the overall operation in the image processing device 90 according to various programs. Further, the CPU 201 may be a microprocessor. The ROM 202 stores programs used by the CPU 201, calculation parameters, and the like. The RAM 203 temporarily stores programs used in the execution of the CPU 201, parameters that change as appropriate during the execution, and the like. These are connected to each other by a bus 204. The bus 204 is connected to an interface 205 that connects the input device 206, the output device 207, the storage device 208, and the communication device 209 to each other.

  The input device 206 is an input means for a user to input information, such as a mouse, keyboard, touch panel, button, microphone, switch, and lever, and an input control circuit that generates an input signal based on the input by the user and outputs it to the CPU 201 Etc.

  The output device 207 includes a display device such as a CRT (Cathode Ray Tube) display device, a liquid crystal display (LCD) device, an OLED (Organic Light Emitting Diode) device, and a lamp. Furthermore, the output device 207 includes an audio output device such as a speaker and headphones.

  The storage device 208 is a data storage device configured as an example of a storage unit of the image processing device 90, and includes, for example, an HDD (Hard Disk Drive). The storage device 208 drives a hard disk, for example, and stores programs executed by the CPU 201 and various data. The communication device 209 is a communication interface configured by a communication device or the like for connecting to the communication network 210, for example. As the communication device 209, for example, a wireless LAN (Local Area Network) compatible communication device, a wireless USB compatible communication device, a communication device that performs wired communication, or the like can be used.

  Note that the information processing apparatus 200 may include a storage medium reader / writer that is built in or externally attached to the information processing apparatus 200, for example. The storage medium reader / writer reads information recorded on a removable recording medium such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, and outputs the information to the RAM 203. The information processing apparatus 200 can also include an interface connected to an external device, and can be used as a connection port with an external device capable of transmitting data by, for example, a USB (Universal Serial Bus).

  Further, the above-described defocus detection method of the image processing apparatus 90 according to the present embodiment may be executed by dedicated hardware, but may be executed by software. When a series of processing is performed by software, for example, the above-described series of processing can be realized by causing a general-purpose or dedicated computer to execute the program.

  The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to such examples. It is obvious that a person having ordinary knowledge in the technical field to which the present invention pertains can come up with various changes or modifications within the scope of the technical idea described in the claims. Of course, it is understood that these also belong to the technical scope of the present invention.

  For example, in the above embodiment, the two-dimensional frequency distribution of the image data is expressed using discrete cosine transform, but the present invention is not limited to such an example. For example, a two-dimensional frequency distribution of image data may be expressed using discrete Fourier transform.

  In the above embodiment, an image captured by the camera 70 is once recorded in the image memory 80, and the image processing apparatus 90 accesses the image memory 80 to acquire image data. However, the present invention is limited to this example. Not. For example, the image memory 80 may be provided in the image processing apparatus 90, and the image data is input from the camera 70 to the image processing apparatus 90 and temporarily stored in the image memory 80. The image data may be read from the image memory 80. Further, the image data captured by the camera 70 may be directly output to the image processing apparatus 90 without using the image memory 80.

  Furthermore, in the said embodiment, although the notification of the abnormality detected by the defect inspection process was performed using the display apparatus 100, this invention is not limited to this example. For example, the user may be notified of a message in which an abnormality has been detected by an audio output device such as a speaker.

DESCRIPTION OF SYMBOLS 10 Unwinding reel 20 Steel plate 30 Conveying roll 40 Rolling roll 50 Take-up reel 60 Illumination 70 Camera 80 Image memory 90 Image processing apparatus 91 Image taking-in part 92 Image division part 93 DCT calculating part 94 Array conversion part 95 Focus deviation determination part 96 Output unit 100 Display device

Claims (6)

In a surface defect inspection apparatus that inspects a surface defect based on image data obtained by imaging the surface of an object to be inspected by an imaging apparatus, the focus deviation detection apparatus detects a focus deviation of the imaging apparatus,
An arithmetic unit that performs a conversion operation to convert an image signal of the image data into two-dimensionally arranged frequency components;
An array conversion unit that converts a two-dimensional array of frequency components of the image data into a one-dimensional array based on a predetermined rule;
A focus shift determination unit that determines the presence or absence of a focus shift of the imaging device based on a slope of a waveform formed by a part of the frequency component converted into the one-dimensional array;
Equipped with a,
The computing unit is
An image dividing unit for dividing the image data into blocks of a predetermined size;
Each block generated by dividing the image data is converted into a two-dimensionally arranged frequency component using a discrete cosine transform, and the two-dimensional array of each block is obtained for each of the two-dimensionally arranged frequency components. A DCT calculation unit that calculates an average frequency component absolute value by adding the absolute values of the frequency components thus obtained and taking an average over all blocks;
Further comprising
The array conversion unit reads the image data converted into two-dimensionally arranged frequency components by performing zigzag scanning in order of low frequency components to high frequency components, converts the image data into a one-dimensional array, and converts the image data In a reading order of frequency components when converting from a two-dimensional array to a one-dimensional array, a logarithmic value of the average frequency component absolute value of the image data is arranged to form a waveform for calculating the slope,
The focus shift detection device, wherein the focus shift determination unit calculates the slope of the waveform by linearly approximating the frequency components converted into the one-dimensional array . The defocus determination unit, among the converted frequency components in the one-dimensional array, and calculates the gradient of the waveform to be formed by using a low frequency component, the focus deviation detection according to claim 1 apparatus. The defocus determination unit, when converting from two-dimensional array of said image data into one-dimensional array, with the exception of the DC frequency component, and calculating a tilt of the waveform, according to claim 1 or 2 Defocus detection device described in 1. The defocus determination unit
When the absolute value of the tilt is smaller than a predetermined threshold, it is determined that there is no focus shift of the imaging device,
When the absolute value of the slope is greater than a predetermined threshold value, and judging that there is defocus of the image pickup device, focus deviation detecting device according to any one of claims 1-3. A focus shift detection method for detecting a focus shift of the imaging device of a surface defect inspection device for inspecting a surface defect based on image data obtained by imaging the surface of an object to be inspected by an imaging device,
A calculation step of performing a conversion calculation for converting the image signal of the image data into two-dimensionally arranged frequency components;
An array conversion step of converting a two-dimensional array of frequency components of the image data into a one-dimensional array based on a predetermined rule;
A focus shift determination step of determining the presence or absence of a focus shift of the imaging device based on a slope of a waveform formed by a part of the frequency component converted into the one-dimensional array;
Including
The calculation step includes:
An image dividing step of dividing the image data into blocks of a predetermined size;
Each block generated by dividing the image data is converted into a two-dimensionally arranged frequency component using a discrete cosine transform, and the two-dimensional array of each block is obtained for each of the two-dimensionally arranged frequency components. A DCT operation step of calculating an average frequency component absolute value by adding the absolute values of the frequency components and calculating an average for all blocks;
Including
In the array conversion step, the image data converted into two-dimensionally arranged frequency components is read by zigzag scanning in the order of low frequency components to high frequency components, converted into a one-dimensional array, and the image data In a reading order of frequency components when converting from a two-dimensional array to a one-dimensional array, a logarithmic value of the average frequency component absolute value of the image data is arranged to form a waveform for calculating the slope,
In the focus shift determination step, the inclination of the waveform is calculated by linearly approximating the frequency component converted into the one-dimensional array, and the focus shift detection method. Computer
An image signal of image data picked up by the image pickup device of a surface defect inspection device for inspecting a surface defect based on image data obtained by picking up the surface of the object to be inspected by the image pickup device into two-dimensionally arranged frequency components. An arithmetic means for performing a conversion operation to convert;
Array conversion means for converting a two-dimensional array of frequency components of the image data into a one-dimensional array based on a predetermined rule;
Defocus determination means for determining the presence or absence of defocus of the imaging device based on a slope of a waveform formed by a part of the frequency component converted into the one-dimensional array;
Equipped with a,
The computing means is
Image dividing means for dividing the image data into blocks of a predetermined size;
Each block generated by dividing the image data is converted into a two-dimensionally arranged frequency component using a discrete cosine transform, and the two-dimensional array of each block is obtained for each of the two-dimensionally arranged frequency components. DCT calculation means for calculating the average frequency component absolute value by adding the absolute values of the frequency components thus obtained and taking the average over all blocks;
Further comprising
The array conversion means reads the image data converted into the two-dimensionally arranged frequency components by zigzag scanning in order from the low frequency component to the high frequency component, converts the image data into a one-dimensional array, and converts the image data In a reading order of frequency components when converting from a two-dimensional array to a one-dimensional array, a logarithmic value of the average frequency component absolute value of the image data is arranged to form a waveform for calculating the slope,
The program for functioning as a focus shift detection device, wherein the focus shift determination means calculates the slope of the waveform by linearly approximating the frequency components converted into the one-dimensional array .