JPH01311377A - Defect inspecting device for long sheet - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of the Invention) The present invention relates to a long sheet defect inspection apparatus for inspecting a sheet-like object to be inspected that is printed with a repetitive pattern and is continuously sent out.

(Summary of the Invention) The present invention is a defect inspection system that optically continuously detects defective parts from image data obtained by imaging a long sheet-like object on which a repetitive image pattern is formed using an image sensor camera. The device automatically learns the mask pattern based on the image data that collects the repeated image pattern on the inspected object obtained by imaging, and at the same time learns the pattern manipulation pitch in the image data each time. By providing a mask processing function, the accumulation of pitch errors associated with the initial setting of mask patterns is eliminated, improving the accuracy of defect inspection.

(Prior art and its problems) In an inspection line for long sheets printed with repeated patterns and fed out, a CCD camera is used to layer and detect defects in the printed pattern, adhesion of dirt and dust, defects in the sheet material itself, etc. The image data obtained from the inspection object is superimposed with a repetitive pattern that has been learned in advance to remove the repetitive pattern from the image data obtained from the inspected object. A method was taken to detect.

However, with these inspection devices, the mask pattern used for mask processing is captured in advance from image data and created, and at the same time, the pitch at which the mask processing is performed using the mask pattern, that is, the pitch of the repeating pattern, is measured by a scraper using a scale, etc. It was set. During inspection, mask processing is performed for each pitch. Therefore, even if the setting pitch is set as accurately as possible, errors will accumulate. Furthermore, if the production conditions such as the temperature of the object to be inspected change from when the pitch was set, the object to be inspected will expand or contract, which will also cause cumulative errors.

In order to eliminate the negative effects of such accumulated errors, a method has been used in which the mask pattern is created by expanding it in the direction of movement.

As a result, defects occurring in the vicinity of the design, particularly in the vicinity of the movement direction of the design, were overlooked.

This was a major drawback of defect inspection equipment that adopted the mask processing method.

(Objective of the Invention) This invention was devised in view of the above points, and its purpose is to minimize the uninspected area.
To provide a defect inspection device for long sheets that reduces the number of defects overlooked and increases inspection accuracy.

(Structure and Effects of the Invention) In order to achieve the above object, the present invention allows a long sheet-like object to be inspected, which moves continuously in the longitudinal direction and has a repetitive image pattern formed on its surface, to be inspected in the moving direction. In a long sheet defect inspection device that optically detects defects in an inspected object based on image data obtained from an image sensor camera that scans and captures images in orthogonal directions,
(8) From the above image data, it is determined whether there is a scanning section in which all pixels are white, and if there is no scanning section, the above image data is ANDed in the flow direction to form a scan section in the moving direction in the image pattern. a detection means for detecting a stripe; and (b) a deletion means for deleting the stripe in the image data by mask processing using a stripe mask created based on the stripe image detected by the detection means; (e) From the above image data from which the stripe data has been deleted by this deletion means, count the scanning sections in which all pixels are white and the scanning sections including other black pixels, and count the scanning sections in which all pixels are white, and Detects the most frequent continuous part of the consecutive parts of the scanning section, and detects the first scanning section of the continuous scanning section that includes the black pixel located next to the first continuous part among the detected one or more continuous parts. a reference position specifying means for identifying the reference position of the repetitive image pattern; a reference pattern specifying means for performing OR addition in the moving direction and extracting a one-dimensional reference bit pattern consisting of white and black pixels corresponding to the pattern; and let the first position after the reference position specified by the reference position specifying means. However, the consecutive part of the scanning interval of black pixels that follows the most consecutive part of the white scanning interval detected by the same means is
After confirming that the continuous length is the same as the continuous length of the scanning section including the black pixel already specified by the same means, the bit pattern of the scanning section including the black pixel is extracted by the same procedure as the reference pattern specifying means, and the bit pattern of the scanning section including the black pixel is extracted. a next image pattern candidate selection means which compares the bit pattern with a reference bit pattern that has been selected and selects it as a candidate for the beginning of the next repeated image pattern if it is the same; A specific pattern is extracted as plane data from the center included in the standard bit pattern selected by the standard pattern specifying means in correspondence with each other, and if the pattern has a hollow part, it is filled in, and then the center of gravity of both is extracted. and a next image pattern determining means that compares the area and, if they match, determines the selected next image pattern candidate as the next repeated image pattern; Compare the respective pitches calculated from the scan section positions of the beginnings of at least three next image patterns sequentially determined by the next image pattern determining means, and the total length of one unit of image pattern in the moving direction divided by the number of scans. , a reference pitch determining means that determines the number of scans as the reference pitch if they all match, and determines the number of scans determined by majority vote as the reference pitch if they do not match;
hl For each reference pitch obtained by this reference pitch determining means, the above image data is added to a small number (3 pitches in total) of pixels representing black by superimposing them, and a predetermined mask offset value is set in each of the vertical and horizontal directions. A mask pattern creation means for creating a mask pattern by adding the above, is specified by the above means, and then mask processing is performed based on the created mask pattern to obtain defect image data.

With the above configuration, the present invention automatically learns a mask pattern based on a repeated image pattern on an object to be inspected from image data obtained by imaging,
At the same time, by having a function that performs mask processing while learning the repetition pitch of the design in the image data each time,
By eliminating the accumulation of pitch errors caused by the initial settings of the mask pattern, it becomes possible to set the length of the mask pattern in the moving direction to the minimum necessary, which reduces the uninspected area and prevents defects from being overlooked. This results in an effect of increasing inspection accuracy.

(Description of Embodiment) FIG. 1 is a block diagram showing the electrical configuration of an embodiment according to the present invention.

In the figure, a COD camera 1 is placed above a long sheet S that moves continuously.

This COD camera 1 continuously scans and images the sheet S in a direction orthogonal to the moving direction. A repetitive pattern image such as a printed mark formed on a sheet S is sent to a video amplifier 2 as a video signal consisting of a voltage depending on the degree of brightness.

Video Amp2 amplifies the input video signal and increases the S/N ratio with respect to the ground level, and then sends it to analog comparator 3.

Analog comparator 3 converts the input video signal into D
It compares it with the set value inputted from the /A converter 4, and converts a signal with a higher level than the set value to 1 representing black on a pixel-by-pixel basis. For low level signals, white is converted to hail O on a pixel by pixel basis. The binarized signal converted into 1 and O is sent to the image filter 51.

The image filter 51 samples each set bit in the width direction based on the input binary signal, and if the obtained data is black, the image is emphasized by extending the black state until the next sampling. do. Particularly small black dots among the black dots obtained at the same time are removed to prevent erroneous symbol recognition, and then sent to the synchronous OR circuit 52.

The synchronous OR circuit 52 performs OR addition of the input video signals for each pixel in the scanning section, and takes in the result in synchronization with the line pulse input from the frequency dividing circuit 9.

As a result, scanning data synchronized with the movement of the sheet is obtained, and the scanning data is sent to the stripe mask circuit 53.

The stripe mask circuit 53 removes black figures, ie, stripes, included in the input scanning data and formed in the movement direction by mask processing, and then sends the obtained data to the next image data creation circuit 61. send.

The image filter 51, the synchronous OR circuit 52, and the stripe mask circuit 53 constitute the symbol filter circuit 5.

The image data creation circuit 61 sends the input data to the DMA write circuit 62 in order to write it as image data.

The DM to B input circuit 62 sends the input image data to the next bus acquisition 73.

The image data writing circuit 6 is composed of the image data creation circuit 61 and the DMA writing circuit 62.

The bus readout 73 writes the input image data to the designated image memory 74 to be inspected and the mask image memory 75, or reads the designated data, and sends the read image data to the hardware arithmetic circuit 76. Note that the bus arbiter 73 is set to give priority to data transfer from the DMA write circuit 62 when mediating these data.

The bus 73, image memories 74 and 75, and hardware arithmetic circuit 76 are included in the CPU circuit 7.

The CPU circuit 7 also includes a CPU 77 and a program memory 7.
8, a work memory 79, an internal bus connecting these, etc., forming a microcomputer. Furthermore, this microcomputer has
An output circuit 71 and an output circuit 72 are connected.

The microcomputer recognizes the pattern from the image data written in the image memory 74 to be inspected, and detects the mark and pitch length that serve as a reference for the pattern pitch.

A learning operation is also performed, and a mask pattern is created from the learned pattern using the hardware calculation circuit 76 and written into the mask image memory 75.

To detect defects, the image data to be inspected is read out from the image memory 74 to be inspected, and the pattern reference pitch mark is detected therein. At the same time, the mask pattern is read from the mask image memory 75. This mask pattern data and the image data to be inspected are compared by a hardware calculation circuit 76, and a process is performed to remove the mask pattern from the image data to be inspected. The presence or absence of defects is determined from the remaining image data.

The output circuit 71 sends the set parameters to the pattern filter circuit 5 and the D/A converter 4.

The output circuit 72 sends detected defect information and the like to the outside.

The line pulse generator 8 has a built-in oscillator and a frequency dividing circuit 9.
The reference clock is sent to

In addition, in the figure, the hardware arithmetic circuits in the pattern filter circuit 5, image data writing circuit 6, and CPU circuit 7 are dedicated circuits with a hardware configuration, so that they share the processing content with the microcomputer and are continuously input. This enables real-time processing of image data.

Further, L in the figure is a light source that illuminates the sheet S from the back side.

FIG. 2 is a perspective view showing the arrangement of the CCD camera 1 placed above the sheet S and the sheet S. As shown in FIG. The image is scanned and imaged in the direction of

The operation of the embodiment will be described below with reference to the drawings.

FIG. 3 is a general flowchart showing the operation of the apparatus of the embodiment.

As shown in the figure, when the supply of objects to be inspected such as sheets starts with the start of operation, the video signal obtained from the CCD camera 1 is first binarized and then imported into the inspection apparatus main body (step 301). .

Next, the pattern is repeatedly recognized from the captured image data,
The repeating pitch for the recognized symbol is calculated, and as a result, master data based on the repeating symbol is created (step 302).

Next, the obtained master data and the sequentially inputted image data are compared to detect defects (step 30).
3).

FIG. 4 is an explanatory diagram showing a process in which a video signal obtained by scanning the sheet S with the CCD camera 1 is converted into binary data.

In Figure a, when the sheet S is scanned in a direction perpendicular to the moving direction, the ring-shaped pattern existing on the scanning line is depicted as a 3 & Il striped pattern shown in the camera data in the figure. The captured video signal becomes an analog signal consisting of two levels depending on the degree of brightness and darkness, as shown in the video level in the figure.

When this analog signal is compared with the dark direction light level voltage shown in the figure in the analog comparator 3, the 2 (7!) converted data shown in the figure is obtained. is raining down each white pixel.

FIG. 5 is a signal waveform diagram illustrating the synthesis of binary data in the synchronous OR circuit 52.

The line pulses obtained from the line pulse generator 8 and the frequency dividing circuit 9 are generated in proportion to the moving speed of the sheet. That is, since the scanning/imaging cycle in the COD camera l is fixed, it is usually asynchronous with the sheet movement speed. On the other hand, the feeding speed of the object to be inspected is set variously. Therefore, the image data taken in from the COD camera 1 is taken in in synchronization with the movement of the sheet by the synchronous OR circuit 52, thereby responding to changes in the moving speed of the object to be inspected. .

In Figure a, the number of scans of the CCD camera 1 is ORed six times during one line pulse period, which shows a case where the moving speed of the object to be inspected is relatively slow. When the addition of black pixels by the prescribed OR addition is completed, processing data is taken in at a timing synchronized with the line pulse.

In the figure, the number of scans of the CCD camera 1 is ORed twice during one line pulse period, which indicates a case where the moving speed of the object to be inspected is relatively fast. When the addition of black pixels by the prescribed OR addition is completed, processing data is taken in at a timing synchronized with the line pulse.

Note that as the moving speed of the object to be inspected becomes particularly high, the CCD
If the line pulse period is shorter than the scanning period of camera 1, data will be captured overlappingly in one scanning period.
Set it longer than the scanning period of camera 1.

FIG. 6 is an explanatory diagram of detection of a stripe pattern in the stripe mask circuit 53 and its erasure.

In this invention, as will be described later, a pattern is repeatedly recognized based on whether the result of ORing black pixels included in each scanning section is white or black. Therefore, if a stripe is included in the image data, all scanning sections will be black and cannot be recognized. If a stripe is found there, that stripe must be deleted.0For example, in the figure, two continuous stripes are formed in the moving direction in the data obtained from a moving non-inspection object. There is. Therefore, when black pixels are OR-added for each scanning section, the result is that black is included in all sections of the scanning data synchronized with sheet movement, as shown on the right side of the figure.

Next, in order to extract only this stripe, all the scan data is ANDed in the movement direction.

Then, only the portion of all the scan data that is a black pixel can be extracted as a stripe as shown in the lower part of the figure.

Next, a stripe mask is created from the obtained stripes. Using this stripe mask, mask processing is performed on the input data for each scanning section to remove stripes. This masking process is performed using a mask register.

FIG. 7 is an explanatory diagram for specifying a reference position of a repeated symbol from input image data.

For all the scan data that is synchronized with the movement of the sheet, OR addition of black pixels is performed for each scan section.As shown on the right side of the figure, the scan portion where the image data includes a pattern is black; Other scanned areas appear white. Furthermore, since the result of this OR addition corresponds to the symbol, scanning sections with the same addition result appear consecutively. Then, counting the continuous scanning sections of the white part or the black part results in the numbers shown in the figure. Among these consecutive numbers, the consecutive number of black that follows the largest number that rains white is determined as the reference pattern.

In addition, if multiple numbers with the largest number falling in white are counted, the first one of them is used.

In the figure, since there are four consecutive white sections with a number of 40, the first section is adopted, and the next black section with a consecutive number of 10 is specified as the top position of the reference symbol. Furthermore, a certain tolerance value is added to each of the consecutive numbers of white and black toward the front from the specified reference position, and the result is used as pitch candidate data.

FIG. 8 is an explanatory diagram for measuring the symbol ratio of the specified reference symbol in the scanning direction.

The scanning section including the reference symbol specified in FIG. 7 is taken out, and the 10 consecutive scanning sections are OR-added in the moving direction. As a result, the number of black pixels in the pattern part is 10 via), and the number of white pixels is 60.50 from the left.
50.50 bits.

FIG. 9 is an explanatory diagram for similarly measuring the pattern ratio in the width direction for each black section that is set as pitch candidate data in FIG. 7.

In the figure, as a result of sequentially measuring the symbol ratio, it is shown that the section of Candidate 2 matches the reference symbol.

This candidate 2 section is selected as the next pitch candidate.

Figure 10 shows the next pitch candidates selected in Figure 9.
Furthermore, it is an explanatory diagram for comparing individual symbols.

A pattern located in the center of the field of view of the reference pattern in FIG. 9,
Each of the corresponding central diagrams is extracted from the next pitch candidates. Figure a is the pattern taken out. In the figure, after filling in the hollow part inside, the filled black pixels are counted and the pattern area is found to be 80.

Figure d shows another extracted pattern, and since there is no hollow part, if you count the black pixels without filling it, Figure e
The pattern area is 60 as shown in . Therefore, when comparing the pattern area 80 of the figure and the pattern area 60 of the figure e, they do not match. If they do not match, the next pitch candidate will be disqualified. If the pattern areas of both of them match,
Furthermore, as shown in Figure C, the centroids of both are determined, and the obtained centroids are matched and the designs are superimposed. If the symbols match here, the next pitch candidate is recognized as the symbol to be repeated next, and at the same time the pitch from the reference symbol is calculated.

Further, if the patterns do not match or the pattern areas do not match, the process starts to produce another candidate.

Here, the image in the center of the field of view was compared with CC.
This is because when a pattern is photographed with a D camera, the image at the center is the clearest when lens aberrations and the like are taken into consideration.

FIG. 11 is a diagram for explaining the repetition pitch up to the next repeated symbol recognized in the first θ diagram.

Usually, after identifying three or more identical symbols next to the reference symbol, the repetition pitch of each symbol is calculated based on the number of scanning sections.

In the figure, the repetition pitch from the standard symbol to the next certified symbol is O, and the repetition pitch is L from the standard symbol to the next certified symbol.
, and then repeats until the next authorized symbol as pitch 2. Compare these repeating pitches,
If all are the same, that value is determined as the repeat pitch,
If they do not match, the values are determined by majority vote.

FIG. 12 is an explanatory diagram of creating a mask pattern based on the determined repetition pitch.

The figure shows the case of a character pattern. Figure a shows a case where four pitches worth of patterns are superimposed, and some a[IM can be seen at the pixel level constituting the character pattern.

The figure shows the case where a mask pattern is formed by further adding an offset value in the width direction to the pattern formed in figure a.

Figure C shows a case where a mask pattern is formed by further adding an offset value in the movement direction to the pattern formed in Figure A.

Note that when creating these mask patterns, minute black dots present in the image data in addition to character designs are ignored and removed.

Figure d shows the mask pattern formed by the above process.
, 0 showing an enlarged view of the portion 0 The portion of the created mask pattern that protrudes beyond the pitch start position at the upper end is canted. However, these cuts will be added to the rear end of the pitch. The warping does not cause any trouble in actual processing operations.

FIG. 13 is an explanatory diagram of mask processing using the mask pattern formed by the above processing.

When =i is activated, after a mask pattern is formed by the above processing, image data manually input from a CCD camera is sequentially stored in a buffer.

From the binarized data that collects the image data in the buffer,
Similar to the mask pattern learning described above, search for and find the maximum value of the number of consecutive white sections.

Next, the symbols in the corresponding sections are similarly checked to confirm that they are the same.

Based on the confirmed position, it is checked whether the next corresponding symbol exists at the position of the repeated pitch obtained by the above learning.

As shown in the figure, the image data in the buffer is sequentially superimposed on the pattern to be inspected and the mask pattern. All pixels included in the mask pattern of the mask pattern are removed to obtain the test result pattern shown on the right side. The remaining black pixels in this inspection result pattern are defects detected from the object to be inspected.

It is determined whether the size of each defect exceeds the allowable value, and if it exceeds the allowable value and is determined to be defective,
It is output to the outside together with the location information.

In the above embodiment, regardless of the moving speed of the sheet S, the mask pattern can be created by learning at the start of the inspection.

Furthermore, during inspection, the reference position corresponding to the mask pattern can be detected in real time from continuously inputted image data, and mask processing using the mask pattern is performed based on the obtained reference position. It will be done.

In this way, since the reference position is detected each time from continuous image data, there is no cumulative error between the mask pattern and the image data as in the conventional method. The extra bulge that was conventionally provided in the direction of movement of the mask pattern is no longer necessary.As a result, the uninspected area that occurs in the direction of movement of the mask pattern can be set to a minimum, making it possible to prevent defects from being overlooked. can be reduced.

Also, when creating the mask pattern and after starting the manga,
Even if the inspection conditions such as the moving speed of the object to be inspected are slightly changed, the inspection apparatus can sufficiently cope with the change.

In addition, there is no need for phase management of the pattern of a moving object, which was conventionally strict, and the usability of the apparatus is improved.

In the embodiment, image processing has been described for a case where a black pattern is formed on a white background, but it is also possible to process a case where a white pattern is formed on a black background.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the electrical configuration of the embodiment according to the present invention, FIG. 2 is a perspective view showing the arrangement of the CCD camera I and the seat S, and FIG. 3 is a general diagram showing the VtW operation of the embodiment. Flowchart, Figure 4 is an explanatory diagram showing the process of converting a video signal to binary data, Figure 5 is a synchronous OR
A signal waveform diagram illustrating the synthesis of binarized data in the circuit 52, FIG. 6 is a diagram illustrating the detection and erasure of a stripe pattern in the stripe mask circuit 53, and FIG. An explanatory diagram for specifying the reference position of a repeating pattern, Fig. 8 is an explanatory diagram for measuring the pattern ratio in the scanning direction of the identified standard pattern, and Fig. 9 is an explanatory diagram for measuring the pattern ratio in the width direction for the black section. , 1st
FIG. 0 is an explanatory diagram for comparing next pitch candidates with single symbols, FIG. 11 is an explanatory diagram for explaining repeat pitch, FIG. 12 is an explanatory diagram for creating a mask pattern, and FIG. 13 is an explanatory diagram for mask processing. 1... CCD camera 2... Video/kmp 3... Analog comparator 4... D/A converter 5... Pattern filter circuit 51 ... Image filter 52 ... Synchronous OR circuit 53 ... Strive mask circuit 6 ...
Image data writing circuit 61... Image data creation circuit 62... DMA writing circuit 7... CPU circuit 71... Output circuit 72... Output circuit 73...Bus arbiter 74...Test image memory 75...Mask image memory 76...Hard arithmetic circuit 77...CPU 78 ...Program memory 79 ...Work memory 8 ... Line pulse generator 9 ... Frequency division circuit

Claims (1)

[Claims] 1. An image sensor camera that scans and images a long sheet-like object to be inspected that moves continuously in the longitudinal direction and has a repeated image pattern formed on its surface in a direction perpendicular to the direction of movement. Based on the image data obtained from
In a long sheet defect inspection device that optically detects defects in an object to be inspected, (a) It is determined from the above image data whether there is a scanning section in which all pixels are white, and if there is no scanning section, the above image data is and (b) using a stripe mask created based on the stripe image detected by this detection means. , a deletion means for deleting the stripes in the image data by mask processing; (c) from the image data from which the stripe data has been deleted by the deletion means, a scanning section in which all pixels are white, and other black pixels; , count the scanning sections including each pixel, detect the largest number of consecutive white scanning sections among the sections, and among the detected 1 or more consecutive sections, the next continuous section after the first continuous section is counted. (d) including a black pixel starting from the reference position specified by the reference position specifying means; a reference pattern specifying means for ORing the black pixels included in each scanning section in the movement direction for the continuous part of the scanning section, and extracting a one-dimensional reference bit pattern consisting of white and black pixels corresponding to the pattern; e) A continuous scanning section of black pixels that is located first after the reference position specified by the reference position specifying means and that follows the most continuous section of white scanning sections detected by the same means,
After confirming that the continuous length is the same as the continuous length of the scanning section including the black pixel already specified by the same means, the bit pattern of the scanning section including the black pixel is extracted by the same procedure as the reference pattern specifying means, and the bit pattern of the scanning section including the black pixel is extracted. (f) a bit pattern selected by the next image pattern candidate selection means; A specific pattern is extracted as plane data from the center included in the standard bit pattern selected by the standard pattern specifying means in correspondence with each other, and if the pattern has a hollow part, it is filled in, and then the center of gravity of both is extracted. (g) a reference position specified by the reference position specifying means; and (g) a reference position specified by the reference position specifying means. ,
Compare the respective pitches, which are calculated from the scan section positions of the beginnings of at least three next image patterns sequentially determined by the next image pattern determining means, and represent the total length in the moving direction of one unit of image pattern by the number of scans. , a reference pitch determining means that determines the number of scans as the reference pitch if all match, and if they do not match, determines the number of scans determined by majority vote as the reference pitch;
(h) For each reference pitch obtained by this reference pitch determination means, at least three pitches of the above image data are superimposed and pixels representing black are added, and predetermined mask offset values are set in the vertical and horizontal directions respectively. A mask pattern creation means for creating a mask pattern by adding the above, 1. A defect inspection apparatus for a long sheet, characterized in that after specifying the defect by the above means, mask processing is performed based on the created mask pattern to obtain defect image data.