JP3333568B2 - Surface defect inspection equipment - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a surface defect inspection apparatus for extracting a unique portion in a grayscale image in the field of image processing in an image measurement apparatus and a defect inspection apparatus.

[0002]

2. Description of the Related Art Conventionally, in a photosensitive drum manufacturing process,
In order to inspect the appearance defects such as coating unevenness and agglomeration due to contamination caused when applying the photoreceptor to the surface of a substrate such as a drum or a sheet, a visual appearance inspection of the photoreceptor surface, for example, actually, Is visually observed as the color depth of the surface of the test sample. As a method of performing this defect inspection not by visual inspection but by automation, for example, there is a method of imaging the surface of an inspection sample using a one-dimensional CCD camera or the like and detecting a defect by an image processing method. In such a surface defect inspection, generally, image data obtained by imaging an inspection sample is binarized by a certain threshold value, and the rank of the defect is determined from a geometric feature of the binarized data. It is.

[0003]

The above-described discrimination method is suitable as a defect inspection method for discriminating, for example, "missing object" which should be normal. However, many of the defects that appear on the surface of the photoreceptor have a "light defect" in the light base, and it is extremely difficult to accurately capture the "feature amount" of such a light defect. There are many.

[0004] As a specific example, FIG.
(A) shows a profile of a faint extremely analog signal appearing on the surface of the photoconductor, that is, a profile of waveforms 1 and 2 of a luminance signal on the surface of the photoconductor at a typical defective portion appearing on the surface of the photoconductor. The luminance of the defective portion in the luminance distribution on the surface of the photoreceptor is large, and visually looks whiter than the underlayer on the surface.
Then, when such waveforms 1 and 2 of the luminance signal are binarized with a certain threshold value, pulse waveforms 3 and 4 having shapes as shown in FIGS. 11B and 12B are obtained. Therefore, FIG.
12A and 12A, the waveforms 1 and 2 are binarized even if they have different shapes, so that the waveforms shown in FIGS.
In this case, the pulse waveforms 3 and 4 have the same width d, so that the defect shape cannot be classified and determined.

[0005]

According to the first aspect of the present invention, there is provided a solid-state image pickup device for picking up an image of a surface of a test sample, and data for taking in a surface luminance distribution of the picked-up test image of the test sample as image data. a capturing means, and filter means that to extract frequency components of the image data by applying the matrix operator for this preparative incorporated image data, the component image signal components corresponding to the filter characteristic of the matrix first Binarizing means for binarizing with a single threshold, labeling means for labeling the binarized component image by geometric processing, and a defect based on the labeled image area A defect existence area determining means for determining an existence area; and a defect existence area in image data corresponding to a surface luminance distribution obtained by imaging the surface of the inspection sample by a solid-state imaging device. A defect region feature amount calculating means for calculating a variance value of the image brightness distribution in the region as the feature quantity, the feature quantity of the second separation identify defective area divided into the image data of the quality in the threshold value of the calculated defect areas It consists of identification means.

According to the second aspect of the present invention, a solid-state imaging device for imaging the surface of an inspection sample, data acquisition means for acquiring surface luminance distribution of the imaged inspection image of the inspection sample as image data, and this acquisition device. Filter means for applying a matrix operator to the embedded image data; binarization means for binarizing a component image of a signal component corresponding to the filter characteristics of the matrix with a predetermined threshold; Labeling means for performing labeling on the converted component image by geometric processing, defect existing area determining means for determining a defect existing area based on the labeled image area, and a surface of the inspection sample. Integrates image brightness in one direction with image data in image data corresponding to surface brightness distribution obtained by imaging with a solid-state image sensor
Equivalent to the defect existing area in the one-dimensional signal calculated
Using a one-dimensional signal from which only the signal distribution of the part
A defective area feature amount calculating means for calculating the number of pixels exceeding a first threshold value in the defect existing area as a feature amount, and converting the calculated feature amount of the defective area into a pass / fail image data with a second threshold value A defect area separation / identification means for separation / identification is provided.

[0007]

According to the third aspect of the present invention, the inspection sample
A solid-state imaging device for imaging the surface;
Captures the surface brightness distribution of the inspection sample as image data.
Data capturing means, and the captured image data
Inspection and one-dimensional signal calculated by integrating image brightness in the direction
The difference between the sample surface luminance distribution and the image data
Divide into blocks consisting of data from which light unevenness has been removed
Blocking means and the blocked block data
The amount of variance of image brightness within a block for each data
Block data feature value calculating means for calculating the
Pass / fail of the feature calculated for each lock data at a predetermined threshold
Image data is separated and identified, and the spatial
Defect area separation and identification means for detecting a defect having a wavenumber band;
Consisting of

[0009] In the present invention of claim 4 wherein, the solid-state imaging device for imaging a surface of the test sample, and data capture means for taking to the surface brightness distribution of the imaged the test sample and image data, are incorporated the preparative one member of the image data
Lock data and other blocks adjacent to this block data
Data and half of their block size or
A block dividing unit that divides the block data so as to partially overlap each other, a block data characteristic amount calculating unit that calculates a variance of image luminance in a block as a characteristic amount for each of the block data, A defect area separation / identification unit configured to separate and identify the calculated feature amount into pass / fail image data at a predetermined threshold and detect a defect having a spatial frequency band of the block size.

[0010]

[0011]

[0012]

[0013]

According to the first aspect of the present invention, a frequency component whose signal is to be detected is extracted by the operation of a matrix operator, and a defect position is accurately determined by binarization, geometric processing, and clustering. It is possible to separate and identify defect ranks by calculating a feature amount that returns to the original image, which is a multi-valued image, from the information on the defective position.
In particular, by setting the amount of dispersion as a feature,
Can be quantified and is the same as the point of interest visually
Automatic defect inspection becomes possible at the standard level .

[0014]

According to the second aspect of the present invention, the size of the defect can be quantified by using the number of pixels exceeding the threshold value as the feature amount, and the automatic defect inspection can be performed at the same level as the visual attention point. possible and Do Ri, also, integral de in a specific direction
By setting the data row as a threshold, the effect of shading
Realizes binarization using a dynamic threshold that is not affected
And improve the accuracy of defect detection.
That Do not.

[0016]

According to the third aspect of the present invention, it is possible to reduce the calculation and eliminate the redundancy of the detection result by dividing the data into blocks by the blocking means and calculating the characteristic amount for each block by the block data characteristic amount calculating means. possible and will, in particular, by the feature amount dispersion amount of the image brightness in the block, it is possible to quantitatively calculate the low-frequency defect pale contrast with Do Ri, also, the original image
Of the integral data sequence in a specific direction at
To reduce the effects of uneven lighting and shading.
Possible to remove the defect detection and that Do.

[0018]

According to the fourth aspect of the present invention, the block
Block data by block means and calculate block data feature
Means to calculate the feature value for each block.
To reduce computations and eliminate redundancy in detection results
In particular, the amount of variance of image brightness within a block is
By using feature values, low-frequency
Defects can be calculated quantitatively, and by performing one divided block so as to overlap each adjacent block, calculation of a defect feature amount between blocks is accurately performed. It becomes possible.

[0020]

[0021]

An embodiment of the invention of EXAMPLES claim 1 and 2, wherein FIG. 1
This will be described with reference to FIG. FIG. 2 shows a configuration of an imaging system of the surface of the photoconductor of the surface defect inspection apparatus. The photoreceptor 5 as an inspection sample is rotated by a motor 6 about the longitudinal direction X as an axis. The surface of the photoreceptor 5 is illuminated by an illumination light source 7 (connected to a halogen lamp 7a), and the light reflected thereby is sequentially read by a one-dimensional CCD camera 8 as a solid-state image sensor. The one-dimensional CCD camera 8 is provided with a CCD image pickup device in which the pixel arrangement direction is arranged in parallel with the rotation axis of the photoconductor 5, and uses about 2048 pixels.
The image signal read by the one-dimensional CCD camera 8 is sent to an A / D converter 9 and digitized. The digitized component image is sent to the frame memory 10 as data capturing means, and the surface luminance distribution of the component image is captured as image data. The image data captured in this way is sent to the host computer 11, where image processing is performed according to an algorithm shown in FIG. The control signal from the host computer 11 is a pulse motor driver 1
2 drive the motor 6 to drive the photosensitive member 5
Is controlled.

In this case, the host computer 11 is provided with the following various means for performing image processing according to the algorithm shown in FIG. That is, as shown in FIG. 1, this configuration comprises a filter means a for applying a matrix operator to the image data fetched into the frame memory 10, and a signal component corresponding to the filter characteristic of the matrix. A binarizing unit b for binarizing the component image with a predetermined threshold value, a labeling unit c for labeling the binarized component image by geometric processing, and a labeled image area A defect existing area determining means d for determining a defect existing area based on the image data corresponding to a surface luminance distribution obtained by imaging the surface of a test sample corresponding to the defect existing area by the solid-state imaging device. A defect area characteristic amount calculating means e for calculating the characteristic amount of the defect area based on the defect area; a defect area for separating and identifying the calculated characteristic amount of the defect area into pass / fail image data at a predetermined threshold value It is characterized in that a separation identification means f.

As the feature quantity of the defect area calculated by the defect area feature quantity calculating means e, the following one is used. That is, the variance of the image luminance distribution in the defect existing area in the image data corresponding to the surface luminance distribution of the photoconductor 5 obtained by imaging the surface of the photoconductor 5 with the one-dimensional CCD camera 8 is used as the “first feature amount”. The amount is used (the invention of claim 1). In addition, as a “second feature amount”, a value exceeding a predetermined threshold value in a defect existing area in image data corresponding to a surface luminance distribution obtained by imaging the surface of the photoconductor 5 with the one-dimensional CCD camera 8 is used. Ru using the number of pixels. In this case, as the predetermined threshold value in the defect existing area, a one-dimensional signal is calculated by integrating the image luminance in one direction of the image data, and corresponds to the defect existing area in the calculated one-dimensional signal. The signal distribution of the part to be cut out is cut out, and only the cut out one-dimensional signal is used as a threshold (the invention according to claim 2 ).

In such a configuration, as a specific example of the various means described above, a method of detecting a defect composed of components having relatively high spatial frequencies will be described with reference to the flow of FIG. First, a two-dimensional image I (x, y) captured from the one-dimensional CCD camera 8 is stored in the frame memory 10 as an original image. Masking (a ₀ ) of the inspection symmetric region is performed on the stored image as preprocessing. Here, the processing symmetric region of the image I (x, y) is newly added to Ip
(X, y). Next, a Laplacian operator 13 as shown in FIG. 3 is applied to the original image of Ip (x, y) in order to detect a defect composed of high frequency components [a]. The Laplacian operator 13 acts to perform a second derivative of the image, and acts as a high-pass filter in terms of frequency. In this example, a 3 × 3 operator is used. Then, such a Laplacian operator 13 is expressed as Lap (x,
y), the image data operated by this operator is Iop
(X, y). FIG. 4A shows a Laplacian image signal 14 to which the Laplacian operator 13 is applied. In the Laplacian image signal 14, the low frequency component is removed and the high frequency component is filtered as a signal.

Next, when the Laplacian image signal 14 composed of the high-frequency components thus filtered is binarized by a predetermined threshold value L, a binarized image signal 15 as shown in FIG. Can be obtained [b]. The area A of the binarized image signal 15 is a defective area. In the related art (see the conventional example described above), as a geometric feature of the A region serving as such a defective region, for example, the “size” of the A region is set as the “feature amount” of the defective region. Since it is difficult to accurately classify the defect ranks using only simple feature amounts, in the present embodiment, the feature amount of the defect area is obtained by the following method.

Hereinafter, a method of obtaining a characteristic amount based on the region A will be sequentially described. First, as preprocessing, the binarized image signal 15 is expanded b ₁ and reduced ₂ to generate an expanded / reduced image signal 16 as shown in FIG. 4C. Removal is performed [b ₃ ]. The reason for performing such preprocessing is that an image immediately after binarization generally includes isolated points and holes due to the noise,
This is because they cause erroneous detection when labeling a defect position. Therefore, in the dilation algorithm, if the pixel of interest is an object, all surrounding pixels are replaced with the object. In the degeneration algorithm, if the pixel of interest is the background, all surrounding pixels are replaced with the background. By performing such expansion and contraction algorithm operations on all pixels in the processed image area, isolated points and holes can be removed.

Next, after such preprocessing is completed, a labeling process as shown in FIG. 5 is performed [c]. Here, the labeling process refers to an operation of assigning one number to one connected component. That is, first, when met the image data to the pixels 17a in the image region 17 scanned label is not attached, to impart numbered L ₁ to the pixel 17a. Next, it is checked whether or not there is a pixel connected to the vicinity of the pixel 17b.
Number _one . Further, pixels connected in the same manner are checked centering on the numbered pixel 17b and numbered. By repeating such an operation, L
_{When the} pixel numbered ₁ disappears, the image is scanned again, and the pixel 1 in the next image areas 18 and 19 is scanned.
8a, 18b, 19a~19d, ... the number L _2, L _3, L
Add ₄ sequentially. In this way, labeled image areas 17, 18, 19 can be defined at the defect positions.

Next, a description will be given of a method of obtaining the characteristic amount of the defective area from the image areas 17, 18, and 19 labeled in this way. First, the features here are:
The amount of dispersion of the luminance distribution in an appropriate region around the defect is used (corresponding to the first feature amount). In defining this variance, it is necessary to define a calculation area. It is necessary to select this calculation area as an area where a defect is presumed, and this area is defined as a defect existing area. The defect existing area is determined from the labeled image areas 17, 18, and 19.

Therefore, a method of determining a defect existing area will be described [d ₀ , d]. The defect existing area is determined in consideration of an appropriate margin amount for the fillet diameters of the labeled image areas 17, 18, and 19. In FIG. 5, a margin amount Wx is defined for a fillet diameter Df in the X direction, a region Dx is determined, and a region Dy is similarly determined in the Y direction. Thereby, the rectangular area defined by Dx and Dy is set as the defect existing area D. Therefore, in the image processing up to this point, it is possible to specify the defect existing area D in which it is estimated that a defect exists, based on the contrast and the size of the image area with respect to the captured image data.

Next, the feature amount of the defect existing area D is calculated [e, f]. As the feature amount, in addition to the variance in the defect existing area D, the number of pixels having a luminance exceeding a predetermined threshold is used (corresponding to the second feature amount).

Now, assuming that the defect existing area D is D: I (xi, yj) (i = m... N, j = k...) (1), the variance .delta. ,

[0032]

(Equation 1)

Can be expressed as In this case, δ
Indicates a large value in an image region with high contrast. Here, I (x, y) is a captured original image.

As described above, the position where a defect exists is determined by sequentially performing binarization, filtering and geometrical processing on the original image, and the signal characteristic amount is calculated from the original image corresponding to the defect position. As a result, the defect rank can be separated and identified, whereby the specific portion of the signal that changes in an analog manner and its characteristic amount can be accurately obtained. Further, in this case, the contrast of the defect area can be quantified by using the dispersion amount as the feature amount, and the automatic defect inspection can be performed at the same level as the visual attention point.
Furthermore, by using the number of pixels exceeding the threshold value as a feature amount, the size of a defect can be quantified, an automatic defect inspection can be performed at the same level as a visual point of interest, and an integrated data string in a specific direction can be obtained. Is a dynamic threshold which is not affected by shading.
The binarization can be realized, and the accuracy of defect detection can be improved.

Next, an explanation based on an embodiment of the invention of claim 3 and 4 described in Figures 7-10. The description of the same parts as those of the first and second aspects of the present invention will be omitted, and the same reference numerals will be used for the same parts.

As shown in FIG. 7, the present configuration employs a blocking unit g for dividing the captured image data into blocks of a predetermined size, and a feature amount for each of the blocked block data. A block data feature quantity calculating means h for calculating, and a defect area separation for separating and identifying the feature quantity calculated for each block data into good or bad image data with a predetermined threshold value and detecting a defect having a spatial frequency band of the block size. It is characterized by having identification means i.
3 ). In this embodiment, the surface inspection defect device includes a solid-state imaging device that captures an image of the surface of the inspection sample and a data capturing unit that captures the surface luminance distribution of the captured inspection sample as image data. Is done.

In this case, the image data to be blocked by the blocking means g is a one-dimensional signal calculated by integrating the image luminance in one direction of the captured image data and the surface luminance distribution of the inspection sample. from the difference value between the image data that consist removing data of illumination unevenness. Further, the blocking means g divides one block data and another block data adjacent to the block data so that half or a part of the block size overlaps each other (the invention according to claim 4 ). Further, the feature amount of the block data calculated by the block data feature amount calculating means h is made up of the variance of the image luminance in the block.

In such a configuration, in this embodiment,
A method for detecting a defect composed of components having relatively low spatial frequencies will be described. The defect 20 having such a low frequency component may be, for example, as shown in FIG. 8, the coating unevenness that occurs when the photosensitive member 5 is applied to the substrate. Such a defect 20 is generated in the longitudinal direction (X
Direction) has a large spread to about 1/4 to 1/3 of the size, and a small change in luminance. And this kind of defect 20
When the detection is performed by an image processing technique, even if the defect 20 to be detected and the illumination unevenness change in luminance, if they are close in spatial frequency, it is difficult to separate and detect the luminance change. In general, in the imaging system having the configuration shown in FIG. 2 described above, shading of illumination occurs in the longitudinal direction (X direction) of the photoconductor 5, causing unevenness in the surface luminance distribution of the photoconductor 5.

Therefore, in this embodiment, in order to remove the above-mentioned influence of illumination unevenness and the like, the illumination distribution of the image data is determined by a signal obtained by integrating the luminance distribution image in the rotation direction of the photoconductor 5. Substitute. Then, a value obtained by subtracting the integrated signal from the original image is regarded as shading-corrected data, and the defect 20 is detected in the image after the shading correction. Hereinafter, a specific example thereof will be described based on the flow of FIG.

First, the low frequency component defect here has a low spatial frequency and does not require much image resolution.
256 × 2 if the photoconductor 5 has a length of about 400 mm
It is sufficient to capture an image at a pixel density of about 56. For this reason, the original image read at a pixel density of 2048 × 2048 pixels in the apparatus of FIG. 2 is reduced to a pixel density of 256 × 256 [a ₁ ] in order to reduce the processing time. Here, the reduced image is represented by Is (x, y).
Then, the shading data Ishd (x) becomes

[0041]

(Equation 2)

The operation is performed using this equation and stored in the memory [a ₂ , a ₃ ].

Then, based on Isd (x), Is (x, y)
When the image data after the correction is Icmp (x, y), it can be expressed as: Icmp (x, y) = Is (x, y) -Ishd (x) (4) . Thus, the image data Icm corrected for shading due to illumination unevenness
p (x, y) can be obtained [a ₄ ].

A low-frequency defect is detected based on the image data Icmp (x, y) thus corrected for shading. In consideration of the fact that the defect size is largely spread over the inspection area, the low-frequency defect is left as it is. It is not appropriate to detect in the form of Therefore, in the inspection processing of the low-frequency defect as in the present embodiment, the inspection image is divided into several blocks using the blocking means g without obtaining the “defect shape itself”, and the block data characteristic is obtained. The characteristic amount is calculated for each block by the amount calculating means h. In this case, the feature amount in each block is set as the variance of the image luminance distribution.

FIG. 9 shows an example of dividing an image into blocks. Original image Icmp after shading correction
The block data 21 in (x, y) is defined as Ib (x, y). As an example, Icmp (x, y) (x = 1 to 256, y
= 1 to 256), and Ib (x, y) (x = 1 to 256,
y = 1 to 256), the variance δb in each block data 21 is

[0046]

(Equation 3)

Can be expressed as

However, with the simple grid-like division as shown in FIG. 9, it is impossible to efficiently detect a defect at the boundary of the block data 21. Therefore, FIG.
, The divided block data 21a is arranged so as to overlap with other block data 21b, 21c... Adjacent thereto. However, the block size in this case is 32 × 32. By arranging them in such an overlapping manner, it is possible to evaluate a defective portion without dividing it at a block boundary [i].

As described above, by dividing the data into blocks by the blocking means g and calculating the characteristic amount for each block data 21 by the block data characteristic amount calculating means h, it is possible to reduce the calculation and eliminate the redundancy of the detection result. it can. In addition, by using an integral data sequence in a specific direction in the original image as a threshold value in the block data 21, it is possible to perform defect detection in which effects of illumination unevenness and shading are removed. Further, by using the variance of the image luminance in the block data 21 as the feature amount, it is possible to quantitatively calculate a low-frequency defect having a light contrast. Furthermore, by performing one divided block data a so as to overlap each of the adjacent block data 21b, 21c,..., It is possible to accurately calculate a feature amount of a defect between blocks. .

[0050]

According to the first aspect of the present invention, there is provided a solid-state image pickup device for picking up an image of a surface of a test sample, data taking means for taking in a surface luminance distribution of the picked-up test image of the test sample as image data, By operating the matrix operator on the captured image data,
Filter means that to extract frequency components of the data, the component image signal components corresponding to the filter characteristic of the matrix
Binarizing means for binarizing with the first threshold value, labeling means for performing labeling on the binarized component image by geometric processing, and based on the labeled image area Defect existing area determining means for determining a defect existing area;
Defect area feature amount calculation means for calculating, as a feature amount, a variance of an image brightness distribution in a defect existing area in image data corresponding to a surface brightness distribution obtained by imaging the surface of the inspection sample by a solid-state imaging device, Since the calculated feature amount of the defective area is composed of a defective area separation and identification means for separating and identifying the image data of pass / fail with the second threshold value, the frequency component for which the signal is to be detected is extracted by the operation of the matrix operator. Determining defect positions accurately by binarization, geometric processing, and clustering, and separating and identifying defect ranks by calculating the return feature amount to the original image, which is a multi-valued image, from information on the determined defect positions. In particular, by using the variance as the feature value, the contrast of the defect area can be quantified, and the automatic defect detection can be performed at the same level as the visual attention point. It is those that can.

According to a second aspect of the present invention, there is provided a solid-state image pickup device for picking up an image of the surface of a test sample, data taking means for taking in the surface luminance distribution of the picked-up test image of the test sample as image data, Filter means for applying a matrix operator to the embedded image data; binarization means for binarizing a component image of a signal component corresponding to the filter characteristics of the matrix with a predetermined threshold; Labeling means for performing labeling on the converted component image by geometric processing, defect existing area determining means for determining a defect existing area based on the labeled image area, and a surface of the inspection sample. Integrates image brightness in one direction with image data in image data corresponding to surface brightness distribution obtained by imaging with a solid-state image sensor
Equivalent to the defect existing area in the one-dimensional signal calculated
Using a one-dimensional signal from which only the signal distribution of the part
A defective area feature amount calculating means for calculating the number of pixels exceeding a first threshold value in the defect existing area as a feature amount, and converting the calculated feature amount of the defective area into a pass / fail image data with a second threshold value since more the isolation identifying defective region separation identification means, can quantify the size of the defect, can be performed automatically defect inspection at the same level as the target point by visual observation and Do Ri, also,
By setting the integral data sequence in a specific direction as the threshold,
Dynamic thresholds that are not affected by
Binarization to improve defect detection accuracy
A shall Do can be.

[0052]

[0053]

[0054] According to a third aspect, a solid-state imaging device for imaging a surface of the test sample, and data capture means for taking to the surface brightness distribution of the imaged the test sample and image data, are incorporated the preparative while there the image data
The one-dimensional signal calculated by integrating the image brightness in the
Illumination from the difference value between the image data of surface brightness distribution of sample and
Blocking means for dividing into blocks each including data from which unevenness has been removed, and a variance of image luminance in each block as a feature amount for each of the block data blocks.
Area calculating means for calculating a block data feature amount, and a defect area separation for separating and identifying the feature amount calculated for each block data into good or bad image data with a predetermined threshold value and detecting a defect having a spatial frequency band of the block size. since more the identifying means, the image data into blocks by the blocking means, by calculating the feature quantity for each the block of image data by block data feature calculating unit redundancy operation reduces the detection result Yes is possible to eliminate the
And, in particular, the amount of variance in image brightness within a block.
Low-frequency with low contrast
Defects can be calculated quantitatively.
The sequence of integral data in a specific direction in the image
Value, the effect of uneven lighting and shading
That Do is possible to remove the defect detecting.

[0055]

According to a fourth aspect of the present invention, a table of inspection samples is provided.
Solid-state image sensor for imaging a surface, and the imaged inspection
Data that captures the surface brightness distribution of the sample as image data
Data capturing means and the captured image data
Block data and other blocks adjacent to this block data.
And half of the block size
Blocking means for dividing parts so that they overlap each other
And the block for each block data
To calculate the variance of image brightness in the
Lock data characteristic amount calculating means and each block data
The feature amount calculated in the above is converted into good or bad image data at a predetermined threshold.
Separately identified and has a spatial frequency band of its block size
It consists of defect area separation and identification means for detecting defects
In the block data,
Data for each block
This reduces computation and eliminates redundancy in detection results.
In particular, the image brightness within the block
By setting the variance of
Low frequency defects can be calculated quantitatively.
In addition, one divided block is divided into adjacent blocks
By overlapping each time, block
It is possible to calculate the feature amount of defects in between with high accuracy
Doo ing.

[0057]

[Brief description of the drawings]

1 is a flowchart showing a state of defect inspection algorithms of the surface defect inspection apparatus according to an embodiment of the invention of claim 1 and 2 wherein.

FIG. 2 is a configuration diagram illustrating a state of an imaging control system of the surface defect inspection apparatus.

FIG. 3 is a schematic diagram illustrating a configuration of a Laplacian operator.

FIG. 4A is a waveform diagram showing a Laplacian image,
(B) is a waveform diagram showing a binarized image, and (c) is a waveform diagram showing an expansion / contraction image.

FIG. 5 is a schematic diagram showing a labeling process.

FIG. 6 is a schematic diagram showing the width of a labeled defect existing area.

7 is a flowchart showing a state of defect inspection algorithms of the surface defect inspection apparatus according to an embodiment of the invention of claim 3 and 4 wherein.

FIG. 8A is a perspective view showing a shape of a low-frequency defect portion,
(B) is a development view thereof.

FIG. 9 is a schematic diagram illustrating image data that is divided into blocks.

FIG. 10 is a schematic diagram illustrating image data that is overlapped and blocked.

FIG. 11 is a waveform diagram showing an image signal detected from a light defect portion.

12 is a waveform diagram showing a binarized obtained pulse shape to Figure 1 1 of the waveform at the threshold.

[Explanation of symbols]

 Reference Signs List 5 inspection sample 8 solid-state imaging device 10 data acquisition means 13 operator 21, 21a to 21c block data a filter means b binarization means c labeling means d defect existing area determination means e defect area feature quantity calculation means f defect area separation Identification means g Blocking means h Data feature amount calculation means i Defect area separation identification means

──────────────────────────────────────────────────続 き Continuation of front page (56) References JP-A-63-75507 (JP, A) JP-A-4-3271 (JP, A) JP-A-2-185192 (JP, A) JP-A-5-185 82618 (JP, A) JP-A-3-58289 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G01N 21/84-21/958 G06T 1/00-9/40

Claims (4)

(57) [Claims] 1. A solid-state imaging device for imaging the surface of an inspection sample, data acquisition means for acquiring surface luminance distribution of the inspection image of the inspection sample taken as image data, and On the other hand, a filter means for extracting a frequency component of image data by operating a matrix operator, and binarizing a component image of a signal component corresponding to a filter characteristic of the matrix with a first threshold value. Binarizing means, labeling means for labeling the binarized component image by geometric processing, and defect existing area for determining a defect existing area based on the labeled image area Determining means for determining the amount of dispersion of the image luminance distribution in the defect existing area in the image data corresponding to the surface luminance distribution obtained by imaging the surface of the inspection sample with the solid-state imaging device. Surface defects, wherein the defect region feature amount calculating means for calculating an amount, that the feature amount of the calculated defect area becomes more and the second image data in isolation identifying defective region separation identification means acceptability in threshold Inspection equipment. 2. A solid-state image sensor for imaging the surface of an inspection sample, data acquisition means for acquiring surface luminance distribution of the imaged inspection image of the inspection sample as image data, and Filter means for applying an operator of a matrix to the matrix, binarization means for binarizing a component image of a signal component corresponding to the filter characteristic of the matrix with a predetermined threshold value, Labeling means for performing labeling by geometric processing, a defect existing area determining means for determining a defect existing area based on the labeled image area, and imaging of the surface of the inspection sample by a solid-state image sensor. Data in the image data corresponding to the surface luminance distribution obtained by
One-dimensional signal calculated by integrating image brightness in one direction
Only the signal distribution of the part corresponding to the defect existence area of the signal
In the defect existing area using the extracted one-dimensional signal
A defect area feature amount calculating means for calculating the number of pixels exceeding the first threshold value as a feature amount; and a defect area separation identification for separating the calculated feature amount of the defective area into pass / fail image data by a second threshold value. A surface defect inspection device characterized by comprising: 3. A solid-state imaging device for imaging a surface of an inspection sample.
A device and a surface luminance component of the imaged test sample.
And data capture hand stage for taking the cloth as image data, this
Integrates the image brightness of the captured image data in one direction
Signal calculated by the above and the surface luminance distribution of the inspection sample
Data from which illumination unevenness has been removed from the difference value with the image data
Blocking means for dividing the data into blocks
In a block for each block data block
Block that calculates the variance of image brightness
Data characteristic amount calculation means and data calculated for each block data.
And separate the feature values into pass / fail image data with a predetermined threshold
Defect with a spatial frequency band of the block size
And a defect area separation / identification means for outputting
Surface defect inspection apparatus that. 4. A solid-state imaging device for imaging the surface of an inspection sample, data acquisition means for acquiring the imaged surface luminance distribution of the inspection sample as image data, and a block for acquiring the acquired image data. Data and this block
Block data adjacent to the block data
Half or part of the block size overlaps each other
Data dividing means, a block data characteristic amount calculating means for calculating a variance of image luminance in a block as a characteristic amount for each of the divided block data, and a characteristic amount calculated for each of the block data And a defect area separating and identifying means for detecting defects having a spatial frequency band of the block size by separately identifying image data of good or bad with a predetermined threshold value.