JP2007285754A - Flaw detection method and flaw detector - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a flaw detection method capable of accurately detecting display flaws, such as stain. <P>SOLUTION: The flaw detection method has a flaw emphasis processing method S4 for enhancing a flaw part by adapting a secondary difference filter, with respect to a photographed image, a luminance gradient detection processing process S5 for detecting luminance gradient, with respect to the photographed image and an error-detecting and deciding process S6. The error-detecting and deciding process S6 is equipped with a flaw region candidate extraction process for comparing the flaw enhanced image obtained in the flaw enhancement treatment method S4 with a predetermined threshold, to extract a flaw region candidate, the maximum luminance acquisition process for acquiring the maximum luminance in the extracted flaw region candidate; a maximum luminance gradient acquiring process for acquiring the maximum luminance gradient, in the same region as the flaw region candidate in the luminance gradient image obtained in the luminance gradient detection processing process S5; and a flaw region deciding process for deciding whether the flaw region candidate is a flaw region, on the basis of the maximum luminance and the maximum luminance gradient. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a defect detection method and apparatus for accurately detecting defects such as spots and unevenness in an inspection process of various products such as an inspection process in manufacturing a display device such as a liquid crystal panel and a projector which is an application product thereof.

In the case of detecting defects such as spots, unevenness, and streaks by image processing in an LCD panel inspection such as a TFT panel, a target image is acquired by a CCD camera or the like, and the defects are detected from the image.
Since the image acquired by the CCD camera includes various noises and shading caused by illumination and lenses, smoothing processing is performed to remove noise, and filtering processing is performed to emphasize defect components in the image. At the same time, the influence of shading is removed. And the area | region beyond a certain threshold value was extracted as a defect from the processing defect image.

  By the way, as a filter for emphasizing defects in the filter processing, a secondary differential filter (secondary differential filter) is usually used (see, for example, Patent Document 1).

JP 2004-53477 A

  Such a second-order differential filter emphasizes the defect component and removes shading, but the data opposite to the defect component is generated in a region that is not a defect around the defect. Ingredients). Since this portion has a low signal level, if a defect having a high contrast is to be detected, it is possible to remove this portion from the result by setting a sufficiently large value for the threshold for extracting the defect.

However, when there is a defect with high contrast and a defect with low contrast and both must be detected, it is difficult to deal with the threshold value, and the second-order differential filter generates the opposite component generated around the defect area. A false detection with a region as a defect occurs.
In addition, since the size of the defect that can be emphasized with respect to the size of the filter is determined in the second order differential filter, a defect having a size other than that cannot obtain the intended result. For example, a large defect is only partially emphasized, and a small defect has a detection level lower than the actual defect level.

  The present invention has been made in view of the above-described problems, and an object thereof is to provide a defect detection method and apparatus that can accurately detect display defects such as spots, unevenness, and streaks.

  The defect detection method of the present invention includes a defect emphasis processing step for emphasizing a defect portion by applying a secondary difference filter to a captured image, a luminance gradient detection processing step for detecting a luminance gradient for the captured image, A false detection judgment processing step, wherein the false detection judgment processing step compares the defect enhancement image obtained in the defect enhancement processing step with a predetermined threshold and extracts defect region candidates exceeding the threshold In the extraction step, the maximum luminance acquisition step for acquiring the maximum luminance in the defect region candidate extracted in the defect region candidate extraction step, and the luminance gradient image obtained in the luminance gradient detection processing step, the defect region candidate and A maximum luminance gradient acquisition step for acquiring a maximum luminance gradient in the same region, and whether or not the defect region candidate is a defect region is determined based on the maximum luminance and the maximum luminance gradient. Characterized in that it and a defect region determination step.

In the present invention, when a defect such as a spot or unevenness is detected, a defect region candidate is obtained by comparing the defect enhancement image obtained in the defect enhancement processing step using a conventional secondary difference filter with a predetermined threshold. In addition to extraction, the maximum luminance and the maximum luminance gradient in the defect area candidate are separately acquired, and it is determined whether these values are actual defect areas or false detections.
For this reason, a more correct defect area can be detected as compared with the case where the defect area is detected by comparing the defect enhancement result by the secondary difference filter with the threshold value. That is, when a defect is emphasized by applying a secondary differential filter, a signal having a component opposite to the defect component is generated outside the edge portion of the defect region. The signal strength of the opposite component is not necessarily high, but if a small threshold is used to detect a defect with a very small luminance value, the signal at the edge portion may be detected and erroneously detected as a defective area. There is.
On the other hand, according to the present invention, since the maximum luminance and the maximum luminance gradient are referred to, the erroneously detected portion can be easily determined. That is, the maximum luminance in the image (luminance value) enhanced by the secondary difference filter is the largest luminance value in the defect area candidate. On the other hand, since the maximum luminance gradient is the maximum value of the luminance difference between adjacent pixels, it is usually less than the maximum luminance. Therefore, for example, when the maximum brightness gradient is significantly larger than 1.5 times the maximum brightness, the portion is not an actual defect area, but an abnormal value, that is, a signal area of an opposite component generated by applying a second-order difference filter. Can be determined.
Further, for example, even when the maximum brightness gradient is abnormally small, such as less than 0.1 times the maximum brightness, since it is not actually present, it can be determined that it is a false detection part.
Therefore, by considering the maximum brightness and the maximum brightness gradient of the portion extracted as the defect area candidate, it can be easily and accurately determined whether the defect area candidate is an actual defect area or a false detection component.

  In the present invention, the defect emphasis processing step includes an image reduction processing step of reducing the captured image in a plurality of stages to create each reduced image, and applying a secondary difference filter to each reduced image to reduce the plurality of stages. A defect-enhanced image creating step for creating a defect-enhanced image, wherein the luminance gradient detecting step applies a median filter to the captured image to remove a noise component, and removes the noise component A luminance gradient image creating step of creating a luminance gradient image by applying a luminance gradient detection filter to the image obtained, and the erroneous detection determination processing step includes extracting the defect region candidate from the luminance gradient image. It has a brightness gradient image reduction process that creates at least an image reduced to the same size as each defect-enhanced image, and includes a defect area candidate extraction process, a maximum brightness acquisition process, and a maximum brightness gradient. The resulting process, the defect region determining step is preferably carried out using the defect emphasizing image and the brightness gradient image of the same size.

According to such a configuration, since the captured image is reduced in a plurality of stages, defects of a plurality of sizes can be detected without changing the size of the secondary difference filter.
In the noise removal processing step, for example, a 7 × 7 median filter applied twice can be used. By using such a median filter, noises other than defects such as spots can be removed, and defects can be detected with higher accuracy.

  In the present invention, the luminance gradient image creation step calculates and calculates a luminance gradient between the target pixel selected in the imaging pixel and each comparison pixel arranged around the target pixel by a predetermined distance. It is preferable to include a luminance gradient calculation step in which the luminance gradient of the target pixel is the luminance gradient having the maximum absolute value.

In addition, what is necessary is just to set the said predetermined distance to about 1-3 pixels normally. When the distance is set to one pixel, the comparison pixel is arranged adjacent to the target pixel, the luminance gradient in the maximum eight directions can be detected, and the luminance gradient of the edge portion where the luminance changes rapidly can be detected with high accuracy.
In addition, disposing the distance at two pixels means that the comparison pixel is disposed adjacent to the outside of the pixel disposed adjacent to the target pixel, can detect a luminance gradient in a maximum of 12 directions or 16 directions, and has a luminance. It is possible to accurately detect the luminance gradient of the edge portion where the angle changes gently to some extent.
With such a configuration, the luminance gradient can be easily calculated, and the luminance gradient detection process can be executed in a short time.

  In the present invention, the defect area determination step includes Imax as the maximum luminance acquired in the maximum luminance acquisition step, Dmax as the maximum luminance gradient acquired in the maximum luminance gradient acquisition step, α, β (α <β ), It is determined whether or not the condition of the following expression (1) is satisfied. If the condition of expression (1) is satisfied, the defect area candidate is determined as a defect area, and the condition of expression (1) is satisfied. It is preferable to determine that the defective area candidate is a false detection component when the above condition is not satisfied.

  According to such a configuration, it is possible to determine whether or not it is a false detection component simply by substituting the maximum luminance Imax and the maximum luminance gradient Dmax into the conditional expression (1) and comparing them. Can be judged. For this reason, defect detection processing can be performed more efficiently than when the inspector visually inspects.

  The defect detection apparatus of the present invention includes a defect enhancement processing unit that applies a secondary difference filter to a captured image to enhance a defect portion, a luminance gradient detection processing unit that detects a luminance gradient with respect to the captured image, A defect detection candidate processing unit that compares the defect enhancement image obtained by the defect enhancement processing unit with a predetermined threshold value and extracts a defect region candidate exceeding the threshold value. The same as the defect area candidate in the extraction means, the maximum brightness acquisition means for acquiring the maximum brightness in the defect area candidate extracted by the defect area candidate extraction means, and the brightness gradient image obtained by the brightness gradient detection means A maximum luminance gradient acquisition means for acquiring a maximum luminance gradient in the area; and a defect that determines whether the defect area candidate is a defect area based on the maximum luminance and the maximum luminance gradient. Characterized in that it and a region determination unit.

  In the present invention as described above, since the maximum luminance and the maximum gradient are referred to, the same effects as the defect detection method can be obtained, for example, an erroneously detected portion can be easily determined.

FIG. 1 is a block diagram showing the configuration of a defect detection apparatus for detecting defects according to an embodiment of the present invention. In the present embodiment, a case where a spot defect is detected will be described as an example.
In FIG. 1, reference numeral 1 denotes a liquid crystal panel (liquid crystal light valve) to be inspected, and reference numeral 2 denotes a projector which is an image projection device, which can set the liquid crystal panel 1 from the outside. Reference numeral 3 denotes a pattern generator that is a pattern generation device that outputs various patterns to the liquid crystal panel 1. Reference numeral 4 denotes a screen. Reference numeral 5 denotes a CCD camera that is an image pickup unit that captures an image projected on the screen 4. It has a CCD with a resolution of. Reference numeral 6 denotes a computer device which is an image processing means for controlling the pattern generator 3 and the CCD camera 5 to detect a spot defect in the liquid crystal panel 1, and 7 is a display device connected to the computer device 6.

  The computer device 6 includes an image input unit 60, a background image difference processing unit 61, a display area extraction unit 62, a defect enhancement processing unit 63, a luminance gradient detection processing unit 64, and an erroneous detection determination processing unit 65. ing.

As shown in FIG. 2, the defect enhancement processing unit 63 includes an image reduction unit 631, a defect enhancement image creation unit 632, and a noise removal unit 633.
Further, as shown in FIG. 3, the luminance gradient detection processing unit 64 includes a noise removing unit 641 and a luminance gradient image creating unit 642.
As shown in FIG. 4, the erroneous detection determination processing means 65 includes a defect area candidate extraction means 651, a brightness gradient image reduction means 652, a maximum brightness acquisition means 653, a maximum brightness gradient acquisition means 654, and a defect area. Determination means 655.

Next, the operation of the defect detection apparatus configured as described above will be described. The general procedure for defect detection is as follows.
That is, the liquid crystal panel 1 to be inspected is set on the projector 2, the pattern generator 3 is controlled by the computer device 6 to display a specific brightness pattern on the liquid crystal panel 1, and this is projected onto the screen 4 by the projector 2. To do. Then, the image projected on the screen 4 is photographed by the CCD camera 5, the image of the photographed data is output to the computer device 6, the defect detection processing is performed by the computer device 6, and the spot defect detection result of the liquid crystal panel 1 is obtained. Is displayed on the display device 7.

Here, the detailed operation of the defect detection by the computer device 6 will be described with reference to the flowchart of FIG.
First, an image input process (imaging process) is performed by the image input means 60 of the computer device 6 (step S1).
In the image input step S 1, first, an image projected on the screen 4 is photographed by the CCD camera 5, and an image of the photographed data is taken into the image input means 60 of the computer device 6. At this time, the photographing data is taken into the computer device 6 as digital data of, for example, 4096 gradations (12 bits) by an A / D converter (not shown). Note that the imaging data is not limited to 4096 gradations, and may be 1024 gradations (10 bits) or 256 gradations (8 bits). The higher the gradation, the higher the accuracy of processing becomes possible. However, since an expensive CCD camera 5 capable of acquiring high gradation data is required, the gradation of the imaged data depends on the defect inspection target or the like. What is necessary is just to set to a required gradation. The image input unit 60 stores the captured image in a storage unit (not shown).

Next, the background image difference processing means 61 takes the difference between the input image and the background image created in advance, and from the captured image data, a defect-like condition caused by something other than the liquid crystal panel such as an illumination or a lens. A background image difference processing step for removing the luminance change is performed (step S2).
In this background image difference processing step S2, the background image difference processing means 61 subtracts the background image shown in FIG. 6B from the input image (projected image) shown in FIG. A background difference image shown in FIG. The background image is an image of luminance change of the optical system excluding the liquid crystal panel 1. Since the luminance change on the defect due to the projection lamp and the projection lens occurs in both the input image and the background image, if the background image is subtracted from the input image, the background difference image is not a liquid crystal panel such as a projection lamp or a projection lens. The luminance change component on the defect caused by the object is removed.

Subsequently, the display area extracting unit 62 performs a display area extracting process for extracting only the screen portion of the inspected part from the background difference image (step S3). When an image of a TFT panel to be inspected is picked up by the CCD camera 5, in order to pick up an image of the entire TFT panel, the surroundings of the panel must be somewhat copied. For this reason, there is data unrelated to the panel around the panel projection image in the captured image, and the projection image is not rectangular but distorted. Therefore, the display area extraction unit 62 performs a process of cutting out only the portion of the TFT panel projection image, and creates an inspection target image.
That is, in the display area extraction processing step S3, the display area extraction means 62 uses the pattern matching processing (several tens of points near the four corners of the image data) to convert the coordinates of the four corners of the inspected part image (background difference image) shown in FIG. The pattern matching process is performed with four corner reference images prepared in advance for each of four small areas of pixels × several tens of pixels, and the coordinates of the four corners are specified). The display area is extracted by affine transformation so that As a result, as shown in FIG. 7B, the peripheral edge on the screen 4 is removed, and only the screen portion having an accurate rectangle is extracted.
Thus, the preparation process before the defect detection process is completed.

Next, the computer apparatus 6 operates the defect emphasis processing means 63 to sequentially execute the defect emphasis processing step S4, the luminance gradient detection processing step S5, and the erroneous detection determination processing step S6.
When the defect enhancement processing step S4 is executed, as shown in FIG. 8, first, an image reduction processing step of creating a reduced image from the image extracted by the display area extraction means 62 is performed by the image reduction means 631 ( Step S41).
In the image reduction processing step S41, the image reduction unit 631 creates an image reduced from the inspection target image to a predetermined size, for example, a 1/4 size. Specifically, as shown in FIG. 9, when creating a 1/4 size reduced image, a 1/2 size reduced image is obtained by setting the average value of 4 pixels of the inspection target image to 1 pixel. A 1 / 4-size reduced image is created by setting the average value of 4 pixels of the 1/2 reduced image to 1 pixel. Then, by performing the same processing based on this reduced image, processing for creating a 1/8 reduced image, a 1/16 reduced image, and the like is performed.

Therefore, if the flattened original image is 1.3 million pixels of 1300 × 1000, the reduced image of ½ is 325,000 pixels of 650 × 500, and the reduced image of ¼ is 80,000 of 325 × 250. It will have 1250 pixels.
In this way, a reduced image of a predetermined size is created because a defect enhancement filter such as a top hat filter used in the defect enhancement image creation means 632 described later is designed to emphasize a spot defect of a predetermined size. Therefore, by reducing the image size, a relatively large spot that cannot be emphasized with the image size before the reduction can be emphasized by reducing the image size. Accordingly, the size of the reduced image may be set according to the defect size to be detected. Note that when a small spot defect is detected, the reduced image creating step by the image reducing unit 631 may not be performed.

Next, as shown in FIG. 8, the defect-enhanced image creating unit 632 of the defect enhancement processing unit 63 performs a defect-enhanced image creating process for creating an image in which a spot defect is emphasized with respect to the created reduced image (step). S42). The defect-enhanced image creating means 632 emphasizes the defect using a defect enhancement filter (secondary difference filter) corresponding to the defect to be detected. In the present embodiment, the spot defect is emphasized using the top hat filter 635 shown in FIG.
Since the top hat filter 635 has a coefficient of 28 (4 + 3 × 8) applied to the defect enhancement target pixel (the pixel at the center of the filter 635), the defect enhancement image creation means 632 is applied when the filter is applied or The applied enhancement value (luminance enhancement value) of all pixels to which the filter has been applied is divided by “28” to match the luminance level of the original image.

Next, the noise removal means 633 of the defect enhancement processing means 63 applies a median filter to the image enhanced in the defect enhancement image creation step S42 to connect and smooth the defect components separated by noise. Then, a noise removal process for removing noise other than the spot defect is performed (step S43).
As a median filter, although not shown, for example, a general median filter in which the median value (median value) of each luminance value of 9 × 3 pixels is set to the luminance value of the target pixel located at the center of 3 × 3, for example. Can be used.
Although the median filter may be applied once, in this embodiment, the median filter is applied twice to remove noise components with high accuracy.
As described above, the defect enhancement processing step S4 is completed, and a defect-enhanced image reduced to each size in which a spot defect is enhanced and noise is removed is obtained.

Next, the computer device 6 operates the luminance gradient detection processing means 64 to execute the luminance gradient detection processing step S5. In the luminance gradient detection processing step S5, as shown in FIG. 11, first, the noise removal processing step for the image extracted by the display area extraction means 62 is performed by the noise removal means 641 of the luminance gradient detection processing means 64 (step S51).
Specifically, the noise removing unit 641 applies the special 7 × 7 median filter 645 shown in FIG. 12 twice to the captured image (unreduced image).

In the normal median filter, the median value is obtained from all the luminance values in the 7 × 7 pixels, but the median filter 645 of this time does not use 3 pixels at each corner, that is, a total of 12 pixels at the four corners. In the filter 645, the luminance value of only the portion where the number “1” is entered (a total of 37 pixels) is acquired, and the median value is obtained.
That is, the median value is obtained by acquiring the luminance value of a pixel area that is almost circular. This is because if the pixel at the corner is included, the luminance gradient component may change, and appropriate noise removal processing cannot be performed.

  Note that the median filter 645 is used for noise removal of an original image (a non-reduced 1/1 image) for obtaining a luminance gradient because, for example, the luminance gradient component changes when a smoothing filter is used. On the other hand, if the median filter is used, the gradient component does not change so much, and noise can be removed while maintaining the luminance gradient component.

Next, the luminance gradient image creating unit 642 of the luminance gradient detection processing unit 64 sequentially selects target pixels one by one from all the pixels of the image processed in the noise removal processing step S51 (step S52).
When the target pixel is selected in the target pixel selection step S52, the luminance gradient image creating unit 642 calculates the luminance gradient using the luminance gradient detection filter 646 shown in FIG. 13 (step S53).
The luminance gradient detection filter 646 uses the center pixel of the set detection area as a target pixel, and uses a pixel that is arranged in a substantially circular shape around the target pixel and that is separated from the target pixel by about two pixels as a comparison pixel. The luminance gradient between the target pixel and each comparison pixel is calculated, and the one with the maximum absolute value is set as the luminance gradient in the target pixel. The luminance gradient means a luminance change rate (gradient component) for one pixel. Specifically, the luminance gradient is obtained by dividing the luminance difference between two pixels of the image by the distance between the pixels.

Here, a specific procedure for calculating the luminance gradient in the luminance gradient calculating step S53 will be described.
The luminance gradient detection filter 646 subtracts the luminance value of each comparison pixel from the luminance value of the target pixel, divides the luminance difference by the length (distance) between the target pixel and the comparison pixel, and determines between the target pixel and each comparison pixel. A luminance gradient is calculated.
For example, in the luminance gradient detection filter 646 of FIG. 13, the luminance value of the target pixel is “O”, the luminance values of the comparison pixels are “S1 to S12”, and the distance between the target pixel and each comparison pixel is D (O, S1). When ˜D (O, S12), luminance gradients g1 to g12 are calculated using the following equations (2) to (13).

g1 = (O−S1) / D (O, S1) (2)
g2 = (O−S2) / D (O, S2) (3)
g3 = (O−S3) / D (O, S3) (4)
g4 = (O−S4) / D (O, S4) (5)
g5 = (O−S5) / D (O, S5) (6)
g6 = (O−S6) / D (O, S6) (7)
g7 = (O−S7) / D (O, S7) (8)
g8 = (O−S8) / D (O, S8) (9)
g9 = (O−S9) / D (O, S9) (10)
g10 = (O−S10) / D (O, S10) (11)
g11 = (O−S11) / D (O, S11) (12)
g12 = (O−S12) / D (O, S12) (13)

Then, the calculated luminance gradients g1 to g12 having the maximum absolute value are selected and set as the luminance gradient of the target pixel detected by the luminance gradient detection filter 646.
Here, the distance between the target pixel and the luminance comparison pixel can be calculated using an expression for obtaining a distance between two points on the XY coordinates. For example, when obtaining the distance D (O, S1), assuming that the coordinates of the target pixel are (X _O , Y _O ) and the coordinates of the comparison pixel are (X ₁ , Y ₁ ), the distance D (O, S1) is an expression. Required by (14).

  When the luminance gradient calculating step S53 is completed, the luminance gradient image creating unit 642 determines whether the selection of the inspection target pixel is completed (step S54). Here, when the inspection has not been completed for the pixels of the entire image, the target pixel selection step S52 and the luminance gradient calculation step S53 are repeatedly executed.

  On the other hand, when the inspection has been completed for the pixels of the entire image, the luminance gradient detection processing step S5 is terminated. Then, a luminance gradient image is created by calculating the luminance gradient for all pixels. Therefore, in this embodiment, the target pixel selection step S52, the luminance gradient calculation step S53, and the inspection target pixel selection end determination step S54 constitute a luminance gradient image creation step.

When the luminance gradient detection processing step S5 ends, the computer device 6 operates the erroneous detection determination processing means 65 to execute the erroneous detection determination processing (step S6).
In the erroneous detection determination processing step S6, as shown in FIG. 14, first, the defective region candidate extraction step is executed by the defective region candidate extraction means 651 of the erroneous detection determination processing means 65 (step S61).
In the defect area candidate extraction step S61, the defect area candidate extraction unit 651 cuts out a white spot defect and a black spot defect from each defect enhanced image (reduced image) obtained in the defect enhancement processing step S4. A threshold is set, and each spot defect area candidate is cut out. That is, the defect such as a spot includes a white spot defect having a higher luminance value than other pixel portions and a black spot defect having a lower luminance value. For this reason, the white spot defect threshold value and the black spot defect threshold value are set as threshold values, and the white spot defect area candidate is extracted by comparing with the white spot defect threshold value, and the black spot defect threshold value is compared with the black spot defect threshold value. Spot defect region candidates are extracted.

In addition, what is necessary is just to set an optimal value for each threshold value according to the condition of an image. For example, the average value of the spot emphasis value (luminance value) of the spot defect emphasis image (composite image) and its standard deviation are obtained, and the threshold value is set by the following equation.
White spot defect threshold wslevel = avr + α1 ・ σ + β1
Black spot defect threshold bslevel = avr + α2 ・ σ + β2
Here, avr is an average value of the composite image, σ is a standard deviation of the composite image, α1, α2, β1, and β2 are arbitrarily determined depending on the situation of the image to be inspected.

Next, a luminance gradient image reduction processing step for reducing the luminance gradient image obtained in the luminance gradient detection processing step S5 to a predetermined size is performed by the luminance gradient image reduction means 652 (step S62).
That is, the luminance gradient image obtained by the luminance gradient detection processing unit 64 is the same size (1/1) as the original captured image, whereas the defect-enhanced image from which the erroneous detection factor is removed is the defect-enhanced image. Since the reduction processing is performed by the processing means 63, the sizes do not match and cannot be compared.
For this reason, the luminance gradient image reducing unit 652 reduces the luminance gradient image to the same size as the defect-enhanced image from which the defect area candidate is extracted. For example, when a defect area candidate is found in the 1/4 reduced image among the defect emphasized images and it is determined whether the defect area candidate is actually a defective portion, the luminance gradient image reducing unit 652 includes the luminance gradient image. Is reduced to 1/4 size.

  In the reduction processing by the image reduction unit 631 of the defect enhancement processing unit 63, the luminance average value of four pixels of the image before reduction is replaced with one pixel. However, in the reduction processing by the luminance gradient image reduction unit 652, the reduction before processing is performed. The value that maximizes the absolute value of the luminance of the four pixels in the image is replaced with one pixel. As a result, the image can be reduced while maintaining the inclination component (gradient component).

Next, the maximum luminance acquisition unit 653 and the maximum luminance gradient acquisition unit 654 of the erroneous detection determination processing unit 65 execute a maximum luminance acquisition step S63 and a maximum luminance gradient acquisition step S64 as shown in FIG.
That is, the maximum luminance acquisition unit 653 performs particle analysis processing based on the reduced image obtained by extracting the defect area candidate of the defect emphasized image, and determines the maximum in each defect area candidate (white spot defect area candidate and black spot defect area candidate). The luminance (maximum luminance value) Imax is acquired for each reduced image and each defect area candidate. That is, when a plurality of white spot defect area candidates and black spot defect area candidates exist in one reduced image, the maximum luminance Imax is acquired for each defect area candidate.
The maximum luminance gradient acquisition unit 654 acquires the maximum luminance gradient Dmax in each defect area candidate from the luminance gradient image reduced to the same size as the defect emphasized image.

  Next, the defect area determination means 655 of the error detection determination processing means 65 determines whether the defect area candidate is an actual defect area or an error detection component based on the following conditional expression (15) (step) S65).

  Here, the maximum luminance gradient Dmax is the maximum value of the luminance difference between two pixels that are separated by a distance of 1 (one pixel) within the defect area candidate, that is, two adjacent pixels. The brightness should be less than Imax. However, the brightness enhancement filter such as the top hat filter 635 can enhance the image with the highest accuracy when the size of the central portion where “3” or “4” is set is substantially equal to the defect size. On the other hand, if it is too large or too small, the emphasized luminance value may be smaller than the maximum luminance of the actual defective portion. For this reason, in the equation (15), for example, β is set to “1.4”, multiplied by the maximum luminance Imax, and when the maximum luminance gradient Dmax becomes larger than that value, it is determined that it is a false detection region. Yes.

  Further, even when the maximum brightness gradient Dmax is very small, such as less than 1/10 of the maximum brightness Imax, it is impossible in practice. Therefore, in the equation (15), for example, α is set to 0.1 and the maximum brightness Imax is set. When the maximum luminance gradient Dmax is smaller than that value, it is determined that the region is a false detection region.

  Note that specific numerical values of the coefficients α and β are set according to the detected image, the detection target defect, and the filter to be applied. The relationship between α and β may be α <β, α may be set to a value near 0, and β may be set to a value near 1. For example, α is set to about 0 to 0.2, and β is set to about 0.9 to 1.5.

By the above determination process, the erroneous detection component can be removed from the defect enhanced image using the secondary difference filter (top hat filter 635).
In the above-described embodiment, a filter for emphasizing and detecting a spot defect is used. However, even when a filter for emphasizing and detecting a streak or uneven defect is used, a luminance gradient component is separately obtained, and the maximum luminance Imax and the maximum luminance are obtained. By comparing the gradient Dmax, it is possible to perform a process of removing a false detection component.

  After removing the false detection component, the rank of the defect defect in the image is conventionally classified based on the calculated defect area candidate area, average brightness, and edge gradient, for example, for the remaining defect area candidates. What is necessary is just to perform the defect determination process currently performed.

According to this embodiment, there are the following effects.
(1) When a defect area candidate is extracted by comparing the defect enhancement image obtained in the defect enhancement processing step S4 using the secondary difference filter with a predetermined threshold, the maximum luminance Imax and the maximum in the defect area candidate are extracted. Since the luminance gradient Dmax is separately acquired and it is determined whether it is an actual defect area or erroneous detection based on these values, it is possible to reliably remove erroneous detection components that occur during enhancement processing.
For this reason, since the threshold value that could not be reduced because an erroneously detected component is detected conventionally can be reduced, a low-contrast defect can be detected with high accuracy.
Furthermore, if the defect area is too large or too small with respect to the size of the secondary difference filter, the deformation and detection level of the defect detection shape becomes small and becomes a false detection component, but this can also be removed, Defects of various sizes can be detected with high accuracy.

(2) Further, since it can be determined whether or not it is a false detection component only by comparing the maximum luminance Imax and the maximum luminance gradient Dmax based on the equation (15), it can be automatically determined by the computer device 6. For this reason, defect detection processing can be performed more efficiently than in the case where an inspector visually confirms.
In addition, in Equation (15), the coefficients α and β are set according to the detected image, the defect to be detected, and the filter to be applied, so that detection with higher accuracy can be performed.

(3) Since the second-order difference filter is applied by reducing the captured image in a plurality of stages, it is possible to detect defect areas of various sizes.
The luminance gradient image is also reduced in accordance with the size of the defect-enhanced image. At the time of reduction, the value that maximizes the absolute value of the four pixels before reduction is replaced with one pixel. Compared to the case of replacing with a value, the image can be reduced while maintaining the inclination component, and the luminance gradient can be obtained with high accuracy even in the reduced image.

(4) Since the median filter 645 applied for noise removal before detecting the luminance gradient is applied except for the four corners, only pixels that are at substantially the same distance from the target pixel can be obtained. Since the influence of pixels having different distances can be reduced, appropriate noise removal processing can be performed.
Further, since noise removal is performed by applying the median filter 645, it is possible to remove noise while maintaining a luminance gradient component as compared with the case where a smoothing filter is used.

(5) Since the luminance gradient detection filter 646 that can detect the luminance gradient in 12 directions with respect to the target pixel is used, the luminance gradient can be detected with high accuracy.
In addition, since the brightness gradient image creating unit 642 obtains the distance between the pixels from the coordinates of the target pixel and the comparison pixel, the brightness gradient can be detected with higher accuracy. That is, since the luminance gradient detection filter 646 is arranged in a matrix, “variation” occurs in the distance between the center coordinates of each comparison pixel and the center coordinates of the target pixel. However, in this embodiment, since the distance between each coordinate is obtained and the brightness gradient is calculated using this distance, the brightness gradient can be made highly accurate.

The present invention is not limited to the above embodiment.
For example, in the luminance gradient image creating unit 642 of the above-described embodiment, the luminance gradient is obtained using one type of luminance gradient detection filter 646, but another luminance gradient detection filter having a different distance between the target pixel and the comparison pixel is used. You may detect together. For example, when three types of luminance gradient detection filters of distances 1 to 3 are used, edge gradients in various directions including high frequency components can be accurately detected by combining the results of the filters.

Further, the defect enhancement processing means 63 is not limited to the one using the top hat filter 635, and an appropriate filter may be used according to the defect type (stain, unevenness, stripe, etc.) to be emphasized.
Furthermore, the calculation method of the luminance gradient (luminance change rate) is not limited to the method of calculating the luminance difference of each pixel and dividing the luminance difference by the distance between the pixels as in the above-described embodiment. A method may be used.

The detection target of the spot defect is not limited to the liquid crystal light valve using the TFT element as described above, but a liquid crystal panel, plasma display, EL display, DMD (direct mirror device) using other diode elements. ), And display devices and products such as front projectors and rear projectors using them, and even when used for these, they are excluded from the scope of the present invention. It goes without saying that it is not.
Furthermore, the present invention is not limited to the inspection of various display devices. For example, if there are scratches such as spots or streaks on a printed matter, a case of a household electrical appliance, a car body, etc., these are imaged to detect defects such as spots. Since a defect can be detected if a certain image is obtained, it can also be applied to spot inspection of various products such as surface coating and printed matter.
In short, since the present invention can detect any defect having a luminance difference from the surroundings, it can be used to detect a luminance spot / unevenness defect or a color spot / unevenness defect.

The block diagram which shows the structure of the defect detection apparatus of the screen by embodiment of this invention. The block diagram which shows the structure of the defect emphasis processing means of the defect detection apparatus. The block diagram which shows the structure of the brightness | luminance gradient detection process means of the same defect detection apparatus. The block diagram which shows the structure of the false detection judgment processing means of the same defect detection apparatus. The flowchart for demonstrating operation | movement of the defect detection apparatus. Explanatory drawing which shows the background image difference process of the same defect detection apparatus. Explanatory drawing which shows the display area extraction process of the same defect detection apparatus. The flowchart for demonstrating a defect emphasis processing process. Explanatory drawing which shows the reduction process of an image. The figure which shows the top hat filter which is an example of a secondary difference filter. The flowchart for demonstrating a brightness | luminance gradient detection process. The figure which shows the example of a median filter. The figure which shows the example of a brightness | luminance gradient detection filter. The flowchart explaining a misdetection judgment processing process.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Liquid crystal panel, 2 ... Projector, 3 ... Pattern generator, 4 ... Screen, 5 ... CCD camera, 6 ... Computer apparatus, 7 ... Display apparatus, 60 ... Image input means, 61 ... Background image difference processing means, 62 ... Display Area extracting means, 63 ... Defect enhancement processing means, 64 ... Luminance gradient detection processing means, 65 ... False detection judgment processing means, 631 ... Image reduction means, 632 ... Defect enhancement image creation means, 633 ... Noise removal means, 635 ... Top Hat filter, 641... Noise removal means, 642... Brightness gradient image creation means, 645... Median filter, 646... Brightness gradient detection filter, 651... Defect area candidate extraction means, 652. , 654... Luminance gradient acquisition means, 655... Defective area determination means.

Claims (5)

A defect emphasis processing step of emphasizing a defect portion by applying a secondary difference filter to the captured image;
A luminance gradient detection processing step for detecting a luminance gradient with respect to the captured image;
An erroneous detection determination processing step,
The false detection determination processing step includes
A defect region candidate extraction step of comparing the defect enhancement image obtained in the defect enhancement processing step with a predetermined threshold and extracting a defect region candidate exceeding the threshold;
A maximum luminance acquisition step of acquiring the maximum luminance in the defect region candidate extracted in the defect region candidate extraction step;
In the luminance gradient image obtained in the luminance gradient detection processing step, a maximum luminance gradient acquisition step of acquiring a maximum luminance gradient in the same region as the defect region candidate;
A defect region determination step of determining whether the defect region candidate is a defect region based on the maximum luminance and the maximum luminance gradient;
A defect detection method comprising: The defect detection method according to claim 1,
The defect enhancement processing step includes an image reduction processing step for reducing the captured image in a plurality of stages to create each reduced image, and a defect enhancement reduced in a plurality of stages by applying a secondary difference filter to each reduced image. A defect-enhanced image creation process for creating an image,
The luminance gradient detection step includes applying a median filter to the captured image to remove a noise component, and applying a luminance gradient detection filter to the image from which the noise component has been removed. A brightness gradient image creating step for creating
The erroneous detection determination processing step includes a luminance gradient image reduction processing step that creates at least an image obtained by reducing the luminance gradient image to the same size as each defect-emphasized image from which defect area candidates are extracted.
A defect detection method, wherein the defect area candidate extraction step, the maximum luminance acquisition step, the maximum luminance gradient acquisition step, and the defect region determination step are performed using a defect enhanced image and a luminance gradient image of the same size. In the defect detection method according to claim 1 or 2,
The luminance gradient image creating step obtains a luminance gradient between the target pixel selected in the imaging pixel and each comparison pixel arranged at a predetermined distance around the target pixel, and among the calculated luminance gradients A defect detection method comprising: a luminance gradient calculation step in which a luminance gradient of a target pixel is a pixel having a maximum absolute value. In the defect area determination step, the maximum luminance acquired in the maximum luminance acquisition step is Imax, the maximum luminance gradient acquired in the maximum luminance gradient acquisition step is Dmax, and the predetermined coefficients are α and β (α <β). To determine whether or not the following equation (1) is satisfied:

When the condition of the formula (1) is satisfied, the defect area candidate is determined as a defective area, and when the condition of the expression (1) is not satisfied, the defect area candidate is determined as a false detection component. Defect detection method. Defect enhancement processing means for applying a secondary difference filter to the captured image to enhance the defect portion;
Luminance gradient detection processing means for detecting the luminance gradient with respect to the captured image;
An erroneous detection determination processing means,
The erroneous detection determination processing means includes
A defect region candidate extracting unit that compares the defect enhanced image obtained by the defect enhancement processing unit with a predetermined threshold and extracts a defect region candidate that exceeds the threshold;
Maximum brightness acquisition means for acquiring the maximum brightness in the defect area candidate extracted by the defect area candidate extraction means;
In the luminance gradient image obtained by the luminance gradient detection means, maximum luminance gradient acquisition means for acquiring the maximum luminance gradient in the same area as the defect area candidate;
Defect area determination means for determining whether or not the defect area candidate is a defect area based on the maximum brightness and the maximum brightness gradient;
A defect detection apparatus comprising:

Images