JP2012007952A - Visual inspection device - Google Patents

Abstract

An appearance inspection apparatus capable of determining the quality of an appearance with high accuracy based on a color image is provided.
A luminance average value calculation unit 12 obtains luminance average values Avr, Avg, Avb from filter image data Drf, Dgf, Dbf, respectively. The average value comparison unit 13 selects filter image data having the largest contrast between the foreign matter and the background color from the obtained luminance average values Avr, Avg, Avb using the lookup table LT. Then, the foreign matter determination processing unit 14 performs image processing based on the filter image data having the largest contrast between the selected foreign matter and the background color. Therefore, each filter image data Drf, Dgf. From Dbf, image processing is performed using the filter image data having the largest contrast between the foreign matter and the background color, and the foreign matter and the background color are determined. Therefore, the threshold value can be easily set and the foreign matter can be detected with high accuracy. It becomes possible.
[Selection] Figure 1

Description

  The present invention relates to an appearance inspection apparatus.

  Conventionally, for example, a semiconductor chip before dicing formed on a semiconductor wafer is picked up by a color image pickup device for each chip, and image data obtained by picking up the image is processed to determine the presence or absence of foreign matter on the chip. In addition, an appearance inspection is performed to determine whether the product is non-defective.

  Such determination of the presence or absence of non-defective products in the visual inspection is performed by, for example, aligning the non-defective image data group obtained from a plurality of non-defective products and calculating the average and standard deviation of luminance for each pixel having the same coordinates. Calculate it. Then, after aligning the image data acquired from the detected product, pass / fail judgment is performed by evaluating each pixel having the same coordinates by using an evaluation formula defined using the brightness and standard deviation. (For example, Patent Document 1).

JP 2005-265661 A

  By the way, in general, in this type of appearance inspection, the background color is assumed to be constant, and it is assumed that the background color does not change. However, even in a non-defective chip, for example, the color of the background color of the chip is different due to a slight difference in film thickness, resulting in variations in color within the range of non-defective chips. In such a case, in the color image picked up by the color image pickup device, the standard deviation becomes large when the average luminance and the standard deviation are calculated for each pixel based on the color of the background color that is different for each chip.

  Therefore, in the case of Patent Document 1, when the variation in the color of the background color is large within the range of these non-defective chips, the standard deviation becomes large and the inspection accuracy becomes low. There was a risk of overlooking.

  In addition, when a color imaging device is used and image processing is performed based on the color image, the red, green, and blue components of the light have different wavelengths. As for blue and blue, the focal lengths on which the filter images are formed are slightly shifted. As a result, since the resolving power is lowered, an advanced technique is required to interpolate this.

  In addition, when an image sensor in which a red, green, and blue filter is arranged in a Bayer array is used in the color imaging device, interpolation processing is performed on each filter image data. If this interpolation process is not performed with high accuracy, the resolving power will be reduced. Therefore, in order to interpolate this, a complicated interpolation process is required.

  The present invention provides an appearance inspection apparatus capable of accurately determining the quality of an appearance based on a color image.

  The appearance inspection apparatus according to the present invention includes an illumination unit that irradiates light to a workpiece, an imaging unit that images a predetermined area of the workpiece, and image processing performed by inputting captured image data from the imaging unit and performing foreign object processing. In the appearance inspection apparatus having an image processing unit that detects a defect, the image processing unit generates RGB filter image data of red, green, and blue from the color image data obtained by the imaging unit. An image data generation unit; a feature amount extraction unit that calculates a feature amount of an inspection region from each of the red, green, and blue filter image data; and the red, green, and blue extracted by the feature amount calculation unit A filter image comparison unit that inputs feature values of each of the filter image data, compares the feature amounts, and selects optimum filter image data that is easy to detect foreign matter, and the filter image comparison unit selects And performing foreign matter defect detection by image processing using the filtered image data.

  In the appearance inspection apparatus according to the invention, the feature amount extraction unit is a luminance average value calculation unit that calculates a luminance average value of an inspection region as a feature amount from the filter image data of red, green, and blue, and The filter image comparison unit is configured to obtain optimum filter image data based on the luminance average value of each of the red, green, and blue filter image data calculated by the luminance average value calculation unit and a lookup table created in advance. It is an average value comparison part which selects.

  In the appearance inspection apparatus according to the invention, the filter image comparison unit includes a luminance average value of the filter image data of red, green, and blue calculated by the luminance average value calculation unit, a lookup table created in advance, On the basis of the above, a defect is determined as the color abnormality of the workpiece.

  In the appearance inspection apparatus according to the invention, the feature amount extraction unit creates a density histogram of an inspection area from each of the red, green, and blue filter image data, and the slope on the low luminance value side from each density histogram As a feature amount, and the filter image comparison unit calculates a low intensity histogram of each of the red, green, and blue filter image data calculated by the low luminance side inclination calculation unit. A low-intensity side inclination comparison unit that compares the inclinations on the luminance value side and selects filter image data having the largest absolute value of the inclination as optimum filter image data for detecting low-luminance foreign matter. And

  In the appearance inspection apparatus according to the invention, the feature amount extraction unit creates a density histogram of an inspection area from each of the red, green, and blue filter image data, and the inclination on the high luminance value side from each density histogram Is calculated as a feature amount, and the filter image comparison unit calculates the height of the density histogram of each of the red, green, and blue filter image data calculated by the high luminance side inclination calculation unit. A high-intensity side inclination comparison unit that compares inclinations on the luminance value side and selects filter image data having the largest absolute value of the inclination as optimum filter image data for detecting high-intensity foreign matter. And

  In the appearance inspection apparatus according to the invention, the feature amount extraction unit creates a density histogram of an inspection area from each of the red, green, and blue filter image data, and the slope on the low luminance value side from each density histogram A low luminance side inclination calculation unit that calculates the characteristic amount as a feature amount, and creates a density histogram of the inspection region from each of the red, green, and blue filter image data, and the inclination on the high luminance value side is calculated from the respective density histograms. A high-intensity side inclination calculation unit that calculates as a feature quantity, and the filter image comparison unit includes a low-contrast histogram of the red, green, and blue filter image data calculated by the low-intensity side inclination calculation unit. Compare the slopes on the brightness side and select the filter image data with the largest absolute value of the slope as the optimal filter image data for detecting low-intensity foreign objects The slope of the high brightness value side of the density histogram of each of the red, green, and blue filter image data calculated by the low brightness side slope calculation unit and the high brightness side slope calculation unit is compared. And a high-luminance side inclination comparison unit that selects filter image data having the largest absolute value as filter image data optimal for detecting a high-luminance foreign matter.

  In the appearance inspection apparatus according to the invention, the feature amount extraction unit creates a density histogram of an inspection area from each of the red, green, and blue filter image data, and calculates a minimum luminance value from the density histogram as a feature amount. And the filter image comparison unit compares the minimum luminance values of the density histograms of the red, green, and blue filter image data calculated by the minimum luminance value calculation unit. The minimum luminance value comparison unit selects filter image data having the largest minimum luminance value as optimum filter image data for detecting a low-luminance foreign matter.

  In the appearance inspection apparatus according to the invention, the feature amount extraction unit creates a density histogram of an inspection area from each of the red, green, and blue filter image data, and calculates a maximum luminance value from each of the density histograms as a feature amount. The filter image comparison unit compares the maximum luminance values of the density histograms of the red, green, and blue filter image data calculated by the maximum luminance value calculation unit. The filter image data having the smallest maximum luminance value is a maximum luminance value comparison unit that selects filter image data that is optimal for detecting high-luminance foreign matter.

  In the appearance inspection apparatus according to the invention, the feature amount extraction unit creates a density histogram of an inspection area from each of the red, green, and blue filter image data, and calculates a minimum luminance value from the density histogram as a feature amount. A minimum luminance value calculation unit that calculates the maximum luminance value as a feature amount by creating a density histogram of the inspection area from each of the red, green, and blue filter image data. A value calculation unit, and the filter image comparison unit compares the minimum luminance values of the density histograms of the red, green, and blue filter image data calculated by the minimum luminance value calculation unit, and calculates the minimum A minimum luminance value comparison unit that selects filter image data having the largest luminance value as optimum filter image data for detecting low-luminance foreign matter, and Compares the maximum brightness values of the density histograms of the red, green, and blue filter image data calculated by the maximum brightness value calculation unit, and detects the filter image data with the smallest maximum brightness value and detects high brightness foreign matter And a maximum luminance value comparison unit that is selected as filter image data that is optimal for this.

  In the appearance inspection apparatus according to the invention, the feature amount extraction unit calculates a Fourier spectrum of an inspection region from each of the red, green, and blue filter image data, and calculates a maximum frequency from the Fourier spectrum of each filter image data. A maximum frequency component calculation unit that calculates a component as a feature amount, and the filter image comparison unit calculates a maximum Fourier spectrum of each of the red, green, and blue filter image data calculated by the maximum frequency component calculation unit. It is a maximum frequency component comparison unit that compares frequency components and selects filter image data having the largest maximum frequency component as optimum filter image data for detecting foreign matter.

  According to the present invention, it is possible to determine the quality of an appearance with high accuracy based on a color image.

It is a system block diagram for demonstrating the system configuration | structure of an external appearance inspection apparatus. It is explanatory drawing for demonstrating a semiconductor chip. It is explanatory drawing for demonstrating the structure of a look-up table. It is a system configuration figure showing the system configuration of the appearance inspection device of a 2nd embodiment. It is a figure which shows the shading histogram based on red filter image data. It is a figure which shows the shading histogram based on green filter image data. It is a figure which shows the shading histogram based on blue filter image data. It is a system configuration figure showing the system configuration of the appearance inspection device of a 3rd embodiment. It is a figure which shows the shading histogram based on red filter image data. It is a figure which shows the shading histogram based on green filter image data. It is a figure which shows the shading histogram based on blue filter image data. It is a system configuration figure showing the system configuration of the appearance inspection device of a 4th embodiment. It is a figure which shows the shading histogram based on red filter image data. It is a figure which shows the shading histogram based on green filter image data. It is a figure which shows the shading histogram based on blue filter image data. It is a system configuration figure showing the system configuration of the appearance inspection device of a 5th embodiment. It is a figure which shows the shading histogram based on red filter image data. It is a figure which shows the shading histogram based on green filter image data. It is a figure which shows the shading histogram based on blue filter image data. It is a system configuration figure showing the system configuration of the appearance inspection device of a 6th embodiment.

(First embodiment)
Hereinafter, a first embodiment in which an appearance inspection apparatus according to the present invention is embodied in an appearance inspection apparatus for a semiconductor chip formed on a semiconductor wafer will be described with reference to the drawings.

  In FIG. 1, an appearance inspection apparatus 1 includes an inspection table 2, and a semiconductor wafer W as a workpiece is placed on the inspection table 2. As shown in FIG. 2, a plurality of semiconductor chips CP before dicing are partitioned on the semiconductor wafer W. This semiconductor chip CP is a MEMS (Micro Electro Mechanical Systems) in which electronic circuits and mechanical elements are formed on the chip in a later step. In the present embodiment, each semiconductor chip CP has dicing marks AM formed at the four corners thereof, and a rectangular recess 3 is formed in the semiconductor chip CP. And the bottom part of the recessed part 3 is formed in square shape.

  An illumination device 5 as an illumination unit is disposed above the inspection table 2 on which the semiconductor wafer W is placed. The illumination device 5 includes a white LED as a light source, and irradiates the semiconductor wafer W on the inspection table 2 with white light.

  A color imaging device 8 as an imaging unit is disposed above the inspection table 2. The color imaging device 8 images each semiconductor chip CP of the semiconductor wafer W irradiated with white light from the illumination device 5. The color imaging device 8 includes a CCD (Charge Coupled Device) image sensor 8a, and R (red), G (green), and B (blue) color filters for each pixel (light receiving element) of the image sensor 8a. Any one of them is arranged to capture a color image. In the present embodiment, the arrangement of the color filters for each element is a so-called Bayer array.

Then, the color image data GD of each semiconductor chip CP of the semiconductor wafer W imaged by the color imaging device 8 is output to the image processing device 10 as an image processing unit.
The image processing device 10 performs image processing on the color image data GD imaged by the color imaging device 8 to detect the presence / absence of a defect in each semiconductor chip CP of the semiconductor wafer W, and performs pass / fail judgment. The image processing apparatus 10 includes an RGB image data generation unit 11, a luminance average value calculation unit 12, an average value comparison unit 13, a foreign matter determination processing unit 14, a storage unit 15, and a lookup table LT.

  The RGB image data generation unit 11 inputs color image data GD imaged by the color imaging device 8 (CCD image sensor 8a), decomposes the color image data GD, red filter image data Drf, green filter image data Dgf, Blue filter image data Dbf is generated.

  More specifically, pixel data from each pixel (light receiving element) in which the red filter of the image sensor 8a is arranged is red filter image data Drf. Further, pixel data from each pixel (light receiving element) in which the green filter is arranged becomes green filter image data Dgf. Further, pixel data from each pixel (light receiving element) in which the blue filter is arranged becomes blue filter image data Dbf.

  At this time, in the present embodiment, the CCD image sensor 8a employs a Bayer array for the color filter. Therefore, with respect to the total number of pixels (for example, N) of the image sensor 8a, the number of red and blue pixels is N / 4, and the number of green is N / 2.

  Therefore, the RGB image data generation unit 11 performs interpolation processing using the data of the pixels around each pixel, and performs red filter image data Drf, green filter including N pixels (that is, all pixels of the image sensor). Image data Dgf and blue filter image data Dbf are respectively generated.

  The red filter image data Drf, the green filter image data Dgf, and the blue filter image data Dbf each having N pixels, which are created by the RGB image data generation unit 11, are used as a luminance average value calculation unit 12 as a feature amount extraction unit. Is output.

  The luminance average value calculation unit 12 has a semiconductor chip CP with dicing marks AM formed at four corners of the red filter image data Drf, the green filter image data Dgf, and the blue filter image data Dbf. The average luminance values Avr, Avg, and Avb of the luminance values in the filter image data Drf, Dgf, and Dbf in the rectangular indentation 3 in the predetermined inspection area are respectively obtained. In the present embodiment, the luminance values of the filter image data Drf, Dgf, and Dbf are configured by 8-bit bit data, and are divided into gradations of 0 to 255.

  More specifically, for the red filter image data Drf in the predetermined inspection area, the luminance values of the pixels belonging to the inspection area are summed and divided by the number of pixels in the inspection area, and the luminance average based on the red filter image data Drf The value Avr is obtained. Similarly, average luminance values Avg and Avb based on the green filter image data Dgf and the blue filter image data Dbf are also obtained.

  The luminance average values Avr, Avg, Avb for the red, green, and blue filter image data Drf, Dgf, Dbf calculated by the luminance average value calculation unit 12 are output to an average value comparison unit 13 as a filter image comparison unit. Is done. The average value comparison unit 13 compares the average brightness values Avr, Avg, Avb with the selected filter image data Dsf stored in the look-up table LT provided in the image processing apparatus 10, and matches the selected filter image data. Search for Dsf.

  As shown in FIG. 3, the look-up table LT includes, for each color image data GD acquired from one semiconductor chip CP, red, green, and blue filter image data Drf, Dgf, Dbf in a predetermined inspection region. Is a table in which the average brightness values Avr, Avg, Avb are stored, and the selected filter image data Dsf for the average brightness values Avr, Avg, Avb are stored.

  More specifically, the color of the semiconductor chip CP imaged by the color imaging device 8 is within the rectangular frame surrounded by the recess 3 of the semiconductor chip CP if the film thickness is slightly different even if it is within the range of non-defective products. Changes. In this case, even if it is within the range of the non-defective product, based on the change in color, it is not a foreign material but is erroneously detected as a foreign material and is determined to be a defective product.

  Therefore, any of the red filter image data Drf, the green filter image data Dgf, and the blue filter image data Dbf obtained by decomposing the color image data GD acquired by the color imaging device 8 into red, green, and blue. It was verified whether the image data has the highest contrast between the foreign matter in the inspection area and the background color and can be inspected with high accuracy if inspection is performed using the filter image data. The lookup table LT is used to select filter image data to be used for inspection from the combination of the average luminance values Avr, Avg, and Avb.

The lookup table LT is created as follows.
That is, first, all the semiconductor chips CP that are non-defective products and different in color are prepared, and each of these semiconductor chips CP is imaged by the color imaging device 8. Then, for each of the non-defective semiconductor chips CP having different colors, luminance average values Avr, Avg, and Avb are obtained for the red, green, and blue filter image data in a predetermined inspection region. Then, the obtained combinations of the respective luminance average values Avr, Avg, Avb are stored in the lookup table LT for each good semiconductor chip CP having a different color.

  Then, each filter image data Drf, Dgf. In the combination of the luminance average values Avr, Avg, and Avb of Dbf, any of the filter image data of the red filter image data Drf, the green filter image data Dgf, and the blue filter image data Dbf has a contrast between the foreign matter in the inspection region and the background color. Verify whether it appears the most. Then, by verifying, in the combination of the luminance average values Avr, Avg, and Avb, any of the red filter image data Drf, the green filter image data Dgf, and the blue filter image data Dbf is the foreign matter and the background color in the inspection region. The filter image data having the highest contrast is obtained, and the filter image data having the highest contrast is stored as the selected filter image data Dsf.

  Here, the reason why the filter image data has the largest contrast between the foreign matter in the inspection region and the background color is that the contrast of the background color (color) with respect to the foreign matter is the largest when detecting the foreign matter thereafter, and the foreign matter detection. This is because the threshold value in this case can be easily set and inspection can be performed with high accuracy.

  In addition, the look-up table LT is provided with color abnormality determination data NG for the color (color abnormality) of the semiconductor chip CP that is a defective product. The abnormal color is a color that cannot be visually inspected by the foreign matter, background color, and appearance in the inspection area. The semiconductor chip CP having such a color cannot be inspected, and the selected filter image data Dsf is not selected. It is set as taste abnormality determination data NG.

  As the color abnormality determination data NG, all semiconductor chips CP having different color abnormalities are prepared, and each of these semiconductor chips CP is similarly imaged by the color imaging device 8. Then, for each of the semiconductor chips CP having different color abnormalities, average brightness values Avr, Avg, Avb are obtained for the respective filter image data of red, green, and blue in a predetermined inspection region. Then, the obtained luminance average values Avr, Avg, and Avb are stored as color abnormality determination data NG in the lookup table LT.

  Incidentally, in the lookup table LT shown in FIG. 3, the average luminance value Avr of the red filter image data Drf is “55”, the average luminance value Avg of the green filter image data Dgf is “170”, and the average luminance of the blue filter image data Dbf is When the value Avb is “100”, the selected filter image data Dsf is green filter image data Dgf.

  Further, when the average luminance value Avr of the red filter image data Drf is “205”, the average luminance value Avg of the green filter image data Dgf is “110”, and the average luminance value Avb of the blue filter image data Dbf is “145”, The selected filter image data Dsf is red filter image data Drf.

  Therefore, the average value comparison unit 13 compares the luminance average values Avr, Avg, Avb calculated by the luminance average value calculation unit 12 with the selected filter image data Dsf stored in the lookup table LT, and matches the luminance average values. If there are Avr, Avg, and Avb, the selected filter image data Dsf for the luminance average values Avr, Avg, and Avb is read. That is, if the selected filter image data Dsf is the red filter image data Drf, the red filter image data Drf is selected as the selected filter image data Dsf.

  When the average value comparison unit 13 selects the red filter image data Drf as the selected filter image data Dsf, the foreign matter determination processing unit 14 selects the red filter image data Drf having the largest contrast between the foreign matter and the background color. Use it to perform image processing and detect foreign matter.

In the foreign object detection, the luminance value of all the pixels of the red filter image data Drf is compared with a predetermined threshold value, the size and shape of the foreign object are obtained, and the foreign object is detected and the non-defective product is determined.
At this time, since the foreign matter determination processing unit 14 performs image processing using the red filter image data Drf having the largest contrast between the foreign matter and the background color and determines the foreign matter and the background color, the threshold value can be easily set. In addition, highly accurate foreign object detection can be performed.

  Similarly, when green or blue filter image data Dgf or Dbf is selected as the selected filter image data Dsf, image processing is performed using the green or blue filter image data Dgf or Dbf having the largest contrast between the foreign matter and the background color. Since the determination of the foreign matter and the background color is performed, the threshold value can be easily set and high-precision foreign matter detection can be performed.

Then, the foreign matter determination processing unit 14 detects the foreign matter of each semiconductor chip CP, determines the presence / absence of the non-defective product, and stores the determination result in the storage unit 15.
When the combination of the luminance average values Avr, Avg, and Avb calculated by the luminance average value calculation unit 12 is the color abnormality determination data NG in the lookup table LT, the semiconductor chip CP is abnormal in color. Therefore, the semiconductor chip CP is determined as a defective product on the assumption that the foreign matter cannot be detected with high accuracy.

By configuring as described above, the present embodiment has the following effects.
(1) According to the present embodiment, the luminance average value calculation unit 12 calculates the luminance average values Avr, Avg, Avb from the respective filter image data Drf, Dgf, Dbf. The average value comparison unit 13 selects filter image data having the largest contrast between the foreign matter and the background color from the obtained luminance average values Avr, Avg, Avb using the lookup table LT. Then, the foreign matter determination processing unit 14 performs image processing based on the filter image data having the largest contrast between the selected foreign matter and the background color.

  Therefore, each filter image data Drf, Dgf. From Dbf, image processing is performed using the filter image data with the largest contrast between the foreign matter and the background color, and the foreign matter and the background color are judged. Therefore, it is easy to set the threshold and detect the foreign matter with high accuracy. It can be carried out.

(2) According to the present embodiment, the lookup table LT also stores the average luminance values Avr, Avg, Avb for the filter image data Drf, Dgf, Dbf for detecting the color abnormality of the semiconductor chip CP. Therefore, when it is determined that the color is abnormal, the subsequent processing operation by the foreign matter determination processing unit 14 can be omitted.
(Second Embodiment)
Next, a second embodiment will be described with reference to FIG.

  The appearance inspection apparatus 1 according to the first embodiment obtains luminance average values Avr, Avg, Avb in the respective filter image data Drf, Dgf, Dbf of red, green, and blue as feature quantities, and filters of red, green, or blue are based on the obtained values. The appearance inspection of the semiconductor chip CP was performed using any one of the filter image data of the image data Drf, Dgf, and Dbf.

  The appearance inspection apparatus 1 of the present embodiment creates a density histogram for each of the red, green, and blue filter image data Drf, Dgf, and Dbf, and each of the red, green, and blue filter image data Drf based on the density histogram. , Dgf, and Dbf, and a visual inspection is performed.

For convenience of explanation, portions common to the first embodiment are denoted by the same reference numerals, and detailed description thereof is omitted.
In FIG. 4, each of the filter image data Drf, Dgf, Dbf of red, green, and blue, each consisting of N pixels created by the RGB image data generation unit 11 based on the color image data GD, is calculated on the low luminance side inclination. Is output to the unit 12a.

  The low luminance side inclination calculating unit 12a as the feature amount extracting unit obtains luminance values of all the pixels for the red, green and blue filter image data Drf, Dgf and Dbf. In this embodiment, the luminance value is composed of 8-bit bit data, and is divided into gradations of 0 to 255. Then, when the luminance values of all the pixels are obtained, the low luminance side inclination calculating unit 12a calculates a density histogram based on the luminance values of all the pixels for each of the red, green, and blue filter image data Drf, Dgf, and Dbf. create.

  5, 6, and 7 show the density histograms created by the low luminance side inclination calculation unit 12 a based on the luminance values of all the pixels for each of the red, green, and blue filter image data Drf, Dgf, and Dbf. An example is shown.

  FIG. 5 is a red density histogram 21a based on the red filter image data Drf, where the horizontal axis represents the luminance value and the vertical axis represents the number of pixels. A small small luminance distribution 21b existing on the low luminance side away from the red / dark histogram 21a is a histogram of low luminance foreign matter (black pixel foreign matter).

  FIG. 6 is a green light / dark histogram 22a based on the green filter image data Dgf. Similarly, the horizontal axis represents the luminance value and the vertical axis represents the number of pixels. A small small luminance distribution 22b existing on the low luminance side away from the green / dark histogram 22a is a histogram of low luminance foreign matter (black pixel foreign matter).

  FIG. 7 is a blue shade histogram 23a based on the blue filter image data Dbf. Similarly, the horizontal axis represents the luminance value and the vertical axis represents the number of pixels. And the small small luminance distribution 23b which exists in the low-luminance side spaced apart from the blue shade histogram 23a is a histogram by a low-luminance foreign material (black pixel foreign material).

  When the low luminance side inclination calculation unit 12a creates the red, green, and blue gray histograms 21a, 22a, and 23a, the low luminance side slope S1 (= tan θ1: provided that the gray histograms 21a, 22a, and 23a are absolute Value), S2 (= tan θ2: where absolute value), S3 (= tan θ3: where absolute value) are calculated, and these gradients S1, S2, and S3 are output to the low luminance side gradient comparison unit 13a.

  The low luminance side inclination comparison unit 13a as the filter image comparison unit compares the inclinations S1, S2, and S3 (absolute values), and calculates filter image data of a grayscale histogram having the largest inclination. This is because the contrast between the low-intensity distributions 21b, 22b, and 23b due to the adjacent low-intensity foreign matter is large and the threshold Bk1 can be easily set as the grayscale histogram having the largest inclination.

  In each of the density histograms 21a, 22a, and 23a of the red, green, and blue filter image data Drf, Dgf, and Dbf shown in FIGS. 5, 6, and 7, the slope S1 (absolute value) of the red density histogram 21a is shown. The red filter image data Drf is selected because it is the largest.

  At this time, the low luminance side inclination comparison unit 13a uses a luminance value intermediate between the minimum luminance value Bmin1 of the red / dark histogram 21a and the maximum luminance value Bmax1 of the small luminance distribution 21b due to the adjacent low luminance foreign matter for the appearance inspection. Calculated as the threshold value Bk1.

  Then, when the red filter image data Drf is selected in the low luminance side inclination comparison unit 13a, the foreign matter determination processing unit 14 performs image processing using the red filter image data Drf and the threshold value Bk1 to obtain a low luminance foreign matter. Detection of (black pixel foreign matter) is performed.

  The low-intensity foreign object detection is performed by comparing the threshold value Bk1 with the luminance values of all the pixels of the red filter image data Drf, and determining the size and shape of the low-intensity foreign substance from the pixels having the luminance value less than the threshold value Bk1 Is detected and a non-defective product is determined.

  At this time, the foreign matter determination processing unit 14 performs image processing using the red filter image data Drf and the threshold value Bk1 with which the low brightness foreign matter can be extracted with high accuracy in the red filter image data Drf, and thereby the low brightness foreign matter and the background color. Therefore, highly accurate low-intensity foreign object detection can be performed.

  Similarly, when green or blue filter image data Dgf or Dbf is selected, a threshold value for detecting a low-intensity foreign matter is obtained, and an image is obtained using the threshold value and the green or blue filter image data Dgf or Dbf. Since the low brightness foreign object and the background color are determined by performing the processing, it is possible to detect the low brightness foreign object with high accuracy.

Then, the foreign matter determination processing unit 14 performs low-luminance foreign matter detection of each semiconductor chip CP, determines the presence or absence of the non-defective product, and stores the determination result in the storage unit 15.
By configuring as described above, the present embodiment has the following effects.

  (1) According to the present embodiment, the low luminance side inclination calculating unit 12a creates the respective red, green, and blue gray histograms 21a, 22a, and 23a from the respective filter image data Drf, Dgf, and Dbf. Then, the low luminance side inclination calculating unit 12a obtains the low luminance side inclinations S1, S2, and S3 of the respective grayscale histograms 21a, 22a, and 23a.

  The low luminance side inclination comparison unit 13a compares the magnitudes of the inclinations S1, S2, and S3, and the contrast between the small luminance distributions 21b, 22b, and 23b due to the low luminance foreign matter adjacent to the grayscale histogram having the largest inclination is compared. The filter image data of the lightness and darkness histogram having the largest inclination because the threshold Bk1 is the largest and is most easily set was selected.

The foreign matter determination processing unit 14 performs image processing based on filter image data in which the contrast between the selected foreign matter and the background color is large and the threshold value Bk1 is most easily set.
Therefore, each filter image data Drf, Dgf. Since image processing is performed using filter image data in which the contrast between the foreign matter and the background color is large and the threshold value Bk1 is most easily set from Dbf, high-precision, low-luminance foreign matter detection can be performed.

  (2) According to the present embodiment, the magnitudes of the slopes S1, S2, and S3 are compared, and the threshold value Bk1 is compared with the small brightness distributions 21b, 22b, and 23b due to the adjacent low brightness foreign matter in the density histogram having the largest slope. Since the threshold value is set, it is very easy to set the threshold value.

In the present embodiment, the threshold value Bk1 is set to the middle between the minimum luminance value Bmin1 and the maximum luminance value Bmax1 of the adjacent small luminance distribution 21b.
(Third embodiment)
Next, a third embodiment will be described with reference to FIG.

  The appearance inspection apparatus 1 according to the second embodiment uses the low luminance side slopes S1, S2, and S3 of the density histograms 21a, 22a, and 23a of the red, green, and blue filter image data Drf, Dgf, and Dbf as feature amounts. Based on these, the appearance inspection of the low-intensity foreign matter (black pixel foreign matter) on the semiconductor chip CP was performed from any one of the filter image data of red, green, or blue filter image data Drf, Dgf, Dbf.

The appearance inspection apparatus 1 according to the present embodiment is characterized in that an appearance inspection of high-intensity foreign matter (white pixel foreign matter) on the semiconductor chip CP is performed.
For convenience of explanation, portions common to the second embodiment are denoted by the same reference numerals, and detailed description thereof is omitted.

  In FIG. 8, each of the filter image data Drf, Dgf, and Dbf of red, green, and blue, each consisting of N pixels, created by the RGB image data generation unit 11 based on the color image data GD, is calculated on the high luminance side inclination. Is output to the unit 12b.

  The high luminance side inclination calculating unit 12b as the feature amount extracting unit obtains luminance values of all pixels for the red, green, and blue filter image data. In this embodiment, the luminance value is composed of 8-bit bit data, and is divided into gradations of 0 to 255. Then, when the luminance values of all the pixels are obtained, the high luminance side inclination calculating unit 12b obtains a density histogram based on the luminance values of all the pixels for each of the red, green, and blue filter image data Drf, Dgf, and Dbf. create.

  9, 10, and 11 show the density histograms created by the high luminance side inclination calculation unit 12 b based on the luminance values of all the pixels for each of the red, green, and blue filter image data Drf, Dgf, and Dbf. An example is shown.

  FIG. 9 is a red density histogram 31a based on the red filter image data Drf, where the horizontal axis represents the luminance value and the vertical axis represents the number of pixels. A small small luminance distribution 31b existing on the high luminance side away from the red / dark histogram 31a is a histogram of high luminance foreign matter (white pixel foreign matter).

  FIG. 10 is a green shade histogram 32a based on the green filter image data Dgf. Similarly, the horizontal axis represents the luminance value and the vertical axis represents the number of pixels. And the small small luminance distribution 32b which exists in the high-intensity side spaced apart from the green shade histogram 32a is a histogram by a high-intensity foreign material (white pixel foreign material).

  FIG. 11 is a blue shade histogram 33a based on the blue filter image data Dbf. Similarly, the horizontal axis represents the luminance value and the vertical axis represents the number of pixels. And the small small luminance distribution 33b which exists in the high-intensity side spaced apart from the blue shade histogram 33a is a histogram by a high-intensity foreign material (white pixel foreign material).

  When the high-intensity side inclination calculation unit 12b creates the red, green, and blue density histograms 31a, 32a, and 33a, the high-intensity side inclination S4 (= tan θ4: absolute) of each of the density histograms 31a, 32a, and 33a. Value), S5 (= tan θ5: where absolute value), S6 (= tan θ6: where absolute value) are calculated, and these gradients S4, S5, and S6 are output to the high-luminance-side gradient comparison unit 13b.

  The high-intensity side inclination comparison unit 13b as a filter image comparison unit compares the inclinations S4, S5, and S6 (absolute values), and calculates filter image data of a grayscale histogram having the largest inclination.

  This is because the contrast between the small luminance distributions 31b, 32b, and 33b due to the high-intensity foreign matter adjacent to the high-luminance side is large and the threshold value can be easily set as the grayscale histogram having the largest inclination.

  In each of the density histograms 31a, 32a, and 33a of the red, green, and blue filter image data Drf, Dgf, and Dbf shown in FIGS. 9, 10, and 11, the slope S4 (absolute value) of the red density histogram 31a is obtained. The red filter image data Drf is selected because it is the largest.

  At this time, the high luminance side inclination comparison unit 13b uses a luminance value intermediate between the maximum luminance value Bmax2 of the red / dark histogram 31a and the minimum luminance value Bmin2 of the small luminance distribution 31b due to the adjacent high luminance foreign matter for appearance inspection. Calculated as the threshold Bk2.

  Then, when the red filter image data Drf is selected in the high luminance side inclination comparison unit 13b, the foreign matter determination processing unit 14 performs image processing using the red filter image data Drf and the threshold value Bk2 to obtain a high luminance foreign matter. Detection is performed.

  In the high-intensity foreign matter detection, the threshold value Bk2 is compared with the luminance values of all the pixels of the red filter image data Drf, and the size and shape of the high-intensity foreign matter is obtained from the pixels having the luminance value equal to or higher than the threshold value Bk2. Is detected and non-defective product is judged.

  At this time, the foreign matter determination processing unit 14 performs image processing using the red filter image data Drf and the threshold value Bk2 from which the high brightness foreign matter can be extracted with high accuracy in the red filter image data Drf. Therefore, high-accuracy high-intensity foreign object detection can be performed.

  Similarly, when green or blue filter image data Dgf or Dbf is selected, a threshold value for detecting a high-intensity foreign object is obtained, and an image is obtained using the threshold value and the green or blue filter image data Dgf or Dbf. Since the process determines the high-intensity foreign object and the background color, highly accurate high-intensity foreign object detection can be performed.

Then, the foreign matter determination processing unit 14 performs high-intensity foreign matter detection of each semiconductor chip CP, determines whether there is a non-defective product, and stores the determination result in the storage unit 15.
By configuring as described above, the present embodiment has the following effects.

  (1) According to the present embodiment, the high luminance side inclination calculating unit 12b creates the respective red, green, and blue gray histograms 31a, 32a, and 33a from the respective filter image data Drf, Dgf, and Dbf. Then, the high luminance side inclination calculating unit 12b calculates the high luminance side inclinations S4, S5, and S6 of the respective grayscale histograms 31a, 32a, and 33a.

  The high luminance side inclination comparison unit 13b compares the magnitudes of the inclinations S4, S5, and S6, and the threshold value Bk2 is compared with the small luminance distributions 31b, 32b, and 33b due to the adjacent high luminance foreign matter with the grayscale histogram having the largest inclination. The filter image data of the light and shade histogram having the largest inclination is selected because it is easy to set.

The foreign matter determination processing unit 14 performs image processing based on the filter image data in which the threshold Bk2 between the selected foreign matter and the background color is most easily set.
Therefore, each filter image data Drf, Dgf. Since image processing is performed using filter image data in which the threshold Bk2 between the foreign matter and the background color is most easily set from Dbf, high-accuracy high-intensity foreign matter detection can be performed.

  (2) According to the present embodiment, the gradients S4, S5, and S6 are compared, and the threshold Bk2 is compared with the small luminance distributions 31b, 32b, and 33b of the high-intensity foreign matter adjacent to the grayscale histogram having the largest gradient. Since the threshold value is set, it is very easy to set the threshold value.

In the present embodiment, the threshold value Bk2 is set to the middle between the maximum luminance value Bmax2 and the minimum luminance value Bmin2 of the adjacent small luminance distribution 31b.
(Fourth embodiment)
Next, a fourth embodiment will be described with reference to FIG.

  The appearance inspection apparatus 1 of the second embodiment uses the gradients S1, S2, and S3 of the density histogram as the feature values of the filter image data Drf, Dgf, and Dbf for red, green, and blue. On the other hand, the appearance inspection apparatus 1 of the present embodiment is characterized in that the minimum luminance value is set as the feature amount of each of the filter image data Drf, Dgf, and Dbf of red, green, and blue.

For convenience of explanation, portions common to the second embodiment are denoted by the same reference numerals, and detailed description thereof is omitted.
In FIG. 12, the RGB image data generation unit 11 generates red, green, and blue filter image data Drf, Dgf, and Dbf each consisting of N pixels based on the color image data GD. The minimum luminance value calculation unit 12c Is output.

  The minimum luminance value calculation unit 12c as a feature amount extraction unit obtains the luminance values of all the pixels for the red, green, and blue filter image data Drf, Dgf, and Dbf. In this embodiment, the luminance value is composed of 8-bit bit data, and is divided into gradations of 0 to 255. Then, when the luminance values of all the pixels are obtained, the minimum luminance value calculating unit 12c creates a density histogram for each of the filter image data Drf, Dgf, and Dbf for red, green, and blue based on the luminance values of all the pixels. To do.

  FIG. 13, FIG. 14 and FIG. 15 show an example of the density histogram created by the minimum luminance value calculation unit 12c based on the luminance values of all the pixels for each of the red, green and blue filter image data Drf, Dgf and Dbf. Indicates.

  FIG. 13 is a red density histogram 41a based on the red filter image data Drf, where the horizontal axis represents the luminance value and the vertical axis represents the number of pixels. A small small luminance distribution 41b existing on the low luminance side away from the red / dark histogram 41a is a histogram of low luminance foreign matter (black pixel foreign matter).

  FIG. 14 is a green light / dark histogram 42a based on the green filter image data Dgf. Similarly, the horizontal axis represents the luminance value and the vertical axis represents the number of pixels. And the small small luminance distribution 42b which exists in the low-luminance side spaced apart from the green shade histogram 42a is a histogram by a low-luminance foreign material (black pixel foreign material).

  FIG. 15 is a blue shade histogram 43a based on the blue filter image data Dbf. Similarly, the horizontal axis represents the luminance value and the vertical axis represents the number of pixels. And the small small luminance distribution 43b which exists in the low-luminance side spaced apart from the blue shade histogram 43a is a histogram by a low-luminance foreign material (black pixel foreign material).

  When the minimum luminance value calculation unit 12c creates the red, green, and blue density histograms 41a, 42a, and 43a, the minimum brightness value Bmin3r, Bmin3g, and Bmin3b of each of the density histograms 41a, 42a, and 43a are obtained. The values Bmin3r, Bmin3g, and Bmin3b are output to the minimum luminance value comparison unit 13c.

  The minimum luminance value comparison unit 13c as the filter image comparison unit compares the minimum luminance values Bmin3r, Bmin3g, and Bmin3b, and calculates filter image data of a grayscale histogram having the largest minimum luminance value b.

  This is because the contrast between the small luminance distributions 41b, 42b, and 43b due to the adjacent low-intensity foreign matter is large and the threshold value Bk3 can be easily set as the grayscale histogram having the largest minimum luminance value.

  In each of the density histograms 41a, 42a, and 43a of the red, green, and blue filter image data Drf, Dgf, and Dbf in FIGS. 13, 14, and 15, the minimum brightness value Bmin3r of the red density histogram 41a is the largest. Therefore, the red filter image data Drf is selected.

  At this time, the minimum luminance value comparison unit 13c uses a threshold value for using an intermediate luminance value for the appearance inspection between the minimum luminance value Bmin3r of the red / dark histogram 41a and the maximum luminance value Bmax3 of the small luminance distribution 41b due to the adjacent low luminance foreign matter. Calculate as Bk3.

  Then, when the red filter image data Drf is selected in the minimum luminance value comparison unit 13c, the foreign matter determination processing unit 14 performs image processing using the red filter image data Drf and the threshold value Bk3, so that the low brightness foreign matter is detected. Perform detection.

  The low-intensity foreign object detection is performed by comparing the threshold value Bk3 with the luminance values of all the pixels of the red filter image data Drf, and obtaining the size and shape of the low-intensity foreign substance from the pixels having the luminance value equal to or lower than the threshold value Bk3. Is detected and non-defective product is judged.

  At this time, the foreign matter determination processing unit 14 performs image processing using the red filter image data Drf and the threshold value Bk3 that can extract the low brightness foreign matter in the red filter image data Drf with high accuracy, and thereby the low brightness foreign matter and the background color. Therefore, highly accurate low-intensity foreign object detection can be performed.

  Similarly, when green or blue filter image data Dgf or Dbf is selected, a threshold value for detecting a low-intensity foreign matter is obtained, and an image is obtained using the threshold value and the green or blue filter image data Dgf or Dbf. Since the low brightness foreign object and the background color are determined by performing the processing, it is possible to detect the low brightness foreign object with high accuracy.

Then, the foreign matter determination processing unit 14 performs low-luminance foreign matter detection of each semiconductor chip CP, determines the presence or absence of the non-defective product, and stores the determination result in the storage unit 15.
By configuring as described above, the present embodiment has the following effects.

  (1) According to the present embodiment, the minimum luminance value calculation unit 12c creates the red, green, and blue gray histograms 41a, 42a, and 43a from the filter image data Drf, Dgf, and Dbf, respectively. Then, the minimum luminance value calculation unit 12c calculates the minimum luminance values Bmin3r, Bmin3g, and Bmin3b of the respective grayscale histograms 41a, 42a, and 43a.

  The minimum luminance value comparison unit 13c compares the minimum luminance values Bmin3r, Bmin3g, and Bmin3b, and the density histogram having the largest minimum luminance value is compared with the small luminance distributions 41b, 42b, and 43b due to the adjacent low luminance foreign matter. Since the threshold value Bk3 is most easily set, the filter image data of the light and shade histogram having the largest minimum luminance value is selected.

The foreign matter determination processing unit 14 performs image processing based on the filter image data in which the threshold value Bk3 between the selected foreign matter and the background color is most easily set.
Therefore, each filter image data Drf, Dgf. Since image processing is performed using the filter image data in which the threshold Bk3 between the foreign matter and the background color is most easily set from Dbf, highly accurate low-luminance foreign matter detection can be performed.

  (2) According to the present embodiment, the magnitudes of the minimum luminance values Bmin3r, Bmin3g, and Bmin3b are compared, and the density histogram having the largest minimum luminance value is compared with the adjacent small luminance distributions 41b, 42b, and 43b of the low luminance foreign matter. Since the threshold value Bk3 is set, the threshold setting operation becomes very easy.

In the present embodiment, the threshold Bk3 is set to the middle between the minimum luminance value Bmin3r and the maximum luminance value Bmax3 of the small luminance distribution 41b adjacent to the threshold Bk3.
(Fifth embodiment)
Next, a fifth embodiment will be described with reference to FIG.

  The appearance inspection apparatus 1 according to the fourth embodiment inspects low-luminance foreign matter (black pixel foreign matter) using the feature amount as the minimum luminance value of the density histogram, whereas the appearance inspection device 1 according to the present embodiment uses the feature amount as light and shade. It is characterized in that a high-intensity foreign substance (white pixel foreign substance) is inspected as the maximum luminance value of the histogram.

For convenience of explanation, portions common to the fourth embodiment are denoted by the same reference numerals, and detailed description thereof is omitted.
In FIG. 16, each of the filter image data Drf, Dgf, and Dbf of red, green, and blue, each consisting of N pixels, created by the RGB image data generation unit 11 based on the color image data GD, is the maximum luminance value calculation unit 12d. Is output.

  The maximum luminance value calculation unit 12d as the feature amount extraction unit obtains the luminance values of all the pixels for the red, green, and blue filter image data Drf, Dgf, and Dbf. In this embodiment, the luminance value is composed of 8-bit bit data, and is divided into gradations of 0 to 255. When the luminance values of all the pixels are obtained, the maximum luminance value calculating unit 12d creates a density histogram based on the luminance values of all the pixels for each of the red, green, and blue filter image data Drf, Dgf, and Dbf. To do.

  FIG. 17, FIG. 18 and FIG. 19 show an example of the density histogram created by the maximum luminance value calculation unit 12d based on the luminance values of all the pixels for each of the red, green and blue filter image data Drf, Dgf and Dbf. Indicates.

  FIG. 17 is a red density histogram 51a based on the red filter image data Drf, where the horizontal axis represents the luminance value and the vertical axis represents the number of pixels. A small small luminance distribution 51b existing on the high luminance side away from the red / dark histogram 51a is a histogram of high luminance foreign matter (white pixel foreign matter).

  FIG. 18 is a green light / dark histogram 52a based on the green filter image data Dgf. Similarly, the horizontal axis represents the luminance value and the vertical axis represents the number of pixels. A small small luminance distribution 52b existing on the high luminance side away from the green / dark histogram 52a is a histogram of high luminance foreign matter (white pixel foreign matter).

  FIG. 19 shows a blue shade histogram 53a based on the blue filter image data Dbf. Similarly, the horizontal axis represents the luminance value and the vertical axis represents the number of pixels. And the small small luminance distribution 53b which exists in the high-intensity side spaced apart from the blue shade histogram 53a is a histogram by a high-intensity foreign material (white pixel foreign material).

  When the maximum brightness value calculation unit 12d creates the respective density histograms 51a, 52a, and 53a for red, green, and blue, the maximum brightness values Bmax4r, Bmax4g, and Bmax4b of the respective density histograms 51a, 52a, and 53a are obtained, and these maximum brightness values are obtained. The values Bmax4r, Bmax4g, and Bmax4b are output to the maximum luminance value comparison unit 13d.

  The maximum brightness value comparison unit 13d as the filter image comparison unit compares the maximum brightness values Bmax4r, Bmax4g, and Bmax4b, and calculates filter image data of a grayscale histogram having the smallest maximum brightness value.

  This is because the threshold value Bk4 between the small luminance distributions 51b, 52b, and 53b due to the adjacent high-intensity foreign matter is easier to set for the grayscale histogram having the smallest maximum luminance values Bmax4r, Bmax4g, and Bmax4b.

  In each of the density histograms 51a, 52a, and 53a of the red, green, and blue filter image data Drf, Dgf, and Dbf in FIGS. 17, 18, and 19, the maximum luminance value Bmax4r of the red density histogram 51a is the smallest. Therefore, the red filter image data Drf is selected.

  At this time, the maximum luminance value comparison unit 13d uses a threshold value for using an intermediate luminance value for the appearance inspection between the maximum luminance value Bmax4r of the red / dark histogram 51a and the minimum luminance value Bmin4 of the small luminance distribution 51b due to the adjacent high luminance foreign matter. Calculate as Bk4.

  Then, when the red filter image data Drf is selected in the maximum luminance value comparison unit 13d, the foreign matter determination processing unit 14 performs image processing using the red filter image data Drf and the threshold value Bk4, so that the high brightness foreign matter is detected. Perform detection.

  In the high-intensity foreign matter detection, the threshold value Bk4 is compared with the luminance values of all the pixels of the red filter image data Drf, and the size and shape of the high-intensity foreign matter is obtained from the pixels having the luminance value equal to or higher than the threshold value Bk4. Is detected and a non-defective product is determined.

  At this time, the foreign matter determination processing unit 14 performs image processing using the red filter image data and the threshold value Bk4 with which the high brightness foreign matter can be extracted with high accuracy in the red filter image data Drf, so that the high brightness foreign matter and the background color are detected. Therefore, highly accurate high-intensity foreign matter detection can be performed.

  Similarly, when green or blue filter image data Dgf or Dbf is selected, a threshold value for detecting a high-intensity foreign object is obtained, and an image is obtained using the threshold value and the green or blue filter image data Dgf or Dbf. Since the process determines the high-intensity foreign object and the background color, highly accurate high-intensity foreign object detection can be performed.

Then, the foreign matter determination processing unit 14 performs high-intensity foreign matter detection of each semiconductor chip CP, determines whether there is a non-defective product, and stores the determination result in the storage unit 15.
By configuring as described above, the present embodiment has the following effects.

  (1) According to the present embodiment, the maximum luminance value calculation unit 12d creates the respective red, green, and blue gray histograms 51a, 52a, and 53a from the respective filter image data Drf, Dgf, and Dbf. Then, the maximum luminance value calculation unit 12d calculates the maximum luminance values Bmax4r, Bmax4g, and Bmax4b of the respective grayscale histograms 51a, 52a, and 53a.

  The maximum luminance value comparison unit 13d compares the maximum luminance values Bmax4r, Bmax4g, and Bmax4b, and the density histogram having the smallest maximum luminance value is compared with the small luminance distributions 51b, 52b, and 53b due to the adjacent high luminance foreign matter. Since the threshold value Bk4 is most easily set, the filter image data of the light and shade histogram having the smallest maximum luminance value is selected.

The foreign matter determination processing unit 14 performs image processing based on the filter image data in which the threshold value Bk4 between the selected foreign matter and the background color is most easily set.
Therefore, each filter image data Drf, Dgf. Since image processing is performed using the filter image data in which the threshold Bk4 between the foreign matter and the background color is most easily set from Dbf, high-accuracy high-intensity foreign matter detection can be performed.

  (2) According to the present embodiment, the maximum brightness values Bmax4r, Bmax4g, and Bmax4b are compared, and the density histogram having the smallest maximum brightness value is compared with the adjacent small brightness distributions 51b, 52b, and 53b of the high brightness foreign matter. Since the threshold value Bk4 is set, the threshold setting operation becomes very easy.

In the present embodiment, the threshold value Bk4 is set to the middle between the maximum luminance value Bmax4r and the minimum luminance value Bmin4 of the adjacent small luminance distribution 51b.
(Sixth embodiment)
Next, a sixth embodiment will be described with reference to FIG.

The appearance inspection apparatus 1 of the present embodiment is characterized in that the maximum frequency component of the Fourier spectrum is used as the feature amount of each of the filter image data Drf, Dgf, and Dbf of red, green, and blue.
For convenience of explanation, portions common to the above-described embodiments are denoted by the same reference numerals, and detailed description thereof is omitted.

  In FIG. 20, each of the filter image data Drf, Dgf, and Dbf for red, green, and blue, each consisting of N pixels created by the RGB image data generation unit 11 based on the color image data GD, is represented by a maximum frequency component calculation unit 12e. Is output.

  The maximum frequency component calculation unit 12e as the feature amount extraction unit includes a Fourier transform circuit, and the filter image data Drf, Dgf, Dbf for red, green, and blue filter image data Drf, Dgf, Dbf, respectively. The frequency spectrum is obtained by Fourier transform.

  When the maximum frequency component calculation unit 12e obtains a Fourier spectrum for each of the filter image data Drf, Dgf, and Dbf for red, green, and blue, the highest frequency component (the maximum frequency component Fr, Fg, Fb) is obtained for each filter image data Drf, Dgf, Dbf.

  When the maximum frequency component calculation unit 12e obtains the maximum frequency components Fr, Fg, and Fb for the filter image data Drf, Dgf, and Dbf for red, green, and blue, the maximum frequency components Fr, Fg, and Fb are determined as the maximum frequency components. It outputs to the comparison part 13e.

  The maximum frequency component comparison unit 13e as the filter image comparison unit compares the maximum frequency components Fr, Fg, and Fb, and determines filter image data having the highest maximum frequency component.

This is because the filter image data having the highest frequency component has a larger density difference between adjacent pixels and a higher resolving power for detecting a foreign object and better edge detection.
That is, in the image processing, when the image is dull, it means that the resolving power is low, and it is difficult to detect a foreign object (edge). This is because the difference in shading between adjacent pixels is small, that is, the differential value is small, the high frequency component in the Fourier spectrum of the image is small, and the maximum frequency component is small.

  Incidentally, this is because the wavelengths of the red, green, and blue components of the light are different, and the focal length at which each filter image forms is slightly shifted due to the chromatic aberration of the imaging lens of the color imaging device 8. As a result, a difference in resolving power occurs between the red, green, and blue filter image data Drf, Dgf, and Dbf.

  The red, green and blue filter image data Drf, Dgf and Dbf are also interpolated by the filter image data Drf, Dgf and Dbf based on the image sensor 8a in which the red, green and blue filters are arranged in a Bayer array. Differences in resolution occur for images.

Then, for example, when the highest frequency component is the maximum frequency component Fr of the red filter image data Drf, the maximum frequency component comparison unit 13e selects the red filter image data Drf.
When the maximum frequency component comparison unit 13e selects the red filter image data Drf, the foreign matter determination processing unit 14 performs image processing using the red filter image data Drf and a predetermined threshold value to detect foreign matters.

In the foreign object detection, the luminance value of all the pixels of the red filter image data Drf is compared with a threshold value, the size and shape of the foreign object is obtained, the foreign object is detected, and the non-defective product is determined.
At this time, the foreign matter determination processing unit 14 can perform high-precision foreign matter detection by performing image processing using red filter image data having high resolving power and excellent edge detection.

By configuring as described above, the present embodiment has the following effects.
(1) According to the present embodiment, the maximum frequency component calculation unit 12e obtains a Fourier spectrum for each filter image data Drf, Dgf, Dbf, and the highest high frequency component (maximum frequency component) among the obtained frequency components. Fr, Fg, Fb) were calculated.

The maximum frequency component comparison unit 13e compares the maximum frequency components Fr, Fg, and Fb, and selects filter image data having the highest maximum frequency component.
The foreign matter determination processing unit 14 performs image processing based on filter image data that has a large contrast between adjacent pixels and a high resolution for foreign matter detection and excellent edge detection.

  Therefore, each filter image data Drf, Dgf. Since image processing is performed using filtered image data having high resolving power and excellent edge detection from among Dbf, highly accurate foreign object detection can be performed.

In addition, this invention is not limited to the said embodiment, You may implement as follows.
(1) In each of the above embodiments, the color image pickup device 8 used is a single plate type color image pickup device in which the image sensor 8a is a single plate color CCD image pickup device in which color filters are arranged in a Bayer array. It is also possible to use a so-called three-plate type color image pickup device that acquires image data of each color separately for each color.

  (2) In the first embodiment, the average brightness values Avr, Avg, Avb are obtained from the filter image data Drf, Dgf, Dbf, respectively, and the obtained average brightness values Avr, Avg, Avg, Selection filter image data corresponding to the combination of Avb was selected.

  In the case of detecting low-intensity foreign matter, filter image data having the largest luminance average value among the average luminance values Avr, Avg, Avb obtained from the respective filter image data Drf, Dgf, Dbf, that is, low luminance. Low-brightness foreign matter detection may be performed using filter image data in which the contrast between the foreign matter and the background color (color) is the largest.

  On the contrary, when high-intensity foreign matter detection is performed, the filter image data having the smallest luminance average value among the average luminance values Avr, Avg, and Avb, that is, the contrast between the high-luminance foreign matter and the background color (color) is the highest. You may implement so that a high-intensity foreign material detection may be performed using the filter image data which becomes large. In this case, the look-up table LT is not required, and the cost can be reduced.

  (3) In the second embodiment, the slopes S1, S2, and S3 on the low luminance side are extracted as feature quantities, and any one of the filter image data Drf, Dgf, and Dbf is extracted based on these slopes S1, S2, and S3. Selected to perform low-intensity foreign object detection. In the third embodiment, gradients S4, S5, and S6 on the high luminance side are extracted as feature amounts, and one of the filter image data Drf, Dgf, and Dbf is selected based on these gradients S4, S5, and S6. High-intensity foreign matter detection.

  One appearance inspection apparatus 1 extracts inclinations S1, S2, S3 on the low luminance side and inclinations S4, S5, S6 on the high luminance side. Based on these slopes S1 to S6, filter image data suitable for low-luminance foreign matter detection and filter image data suitable for high-luminance foreign matter detection are selected to detect low-luminance foreign matter (black pixel foreign matter) and high-luminance foreign matter. You may implement with the one external appearance inspection apparatus 1 combining the detection of (white pixel foreign material).

  (4) In the fourth embodiment, the minimum luminance values Bmin3r, Bmin3g, and Bmin3b are extracted as feature quantities, and any one of the filter image data Drf, Dgf, and Dbf is extracted based on these minimum luminance values Bmin3r, Bmin3g, and Bmin3b. Selected to perform low-intensity foreign object detection. In the fifth embodiment, the maximum luminance values Bmax4r, Bmax4g, and Bmax4b are extracted as feature amounts, and any one of the filter image data Drf, Dgf, and Dbf is selected based on the maximum luminance values Bmax4r, Bmax4g, and Bmax4b. High-intensity foreign matter detection.

  From this, the minimum luminance values Bmin3r, Bmin3g, and Bmin3b and the maximum luminance values Bmax4r, Bmax4g, and Bmax4b are extracted by one visual inspection apparatus 1. Based on these minimum luminance values Bmin3r, Bmin3g, Bmin3b and maximum luminance values Bmax4r, Bmax4g, Bmax4b, filter image data suitable for low-luminance foreign matter detection and filter image data suitable for high-luminance foreign matter detection are selected and low The detection of the luminance foreign matter (black pixel foreign matter) and the detection of the high-intensity foreign matter (white pixel foreign matter) may be performed by a single appearance inspection apparatus 1.

  (5) When detecting low-intensity foreign matter (black pixel foreign matter), the second embodiment and the fourth embodiment are simultaneously performed, and the filter image data for which the threshold values Bk1 and Bk3 are easier to set is selected. Luminance foreign matter may be detected.

  (6) When high-intensity foreign matter (white pixel foreign matter) is detected, the third embodiment and the fifth embodiment are simultaneously performed, and the filter image data for which the threshold values Bk2 and Bk4 are easier to set is selected. Luminance foreign matter may be detected.

  DESCRIPTION OF SYMBOLS 1 ... Appearance inspection apparatus, 2 ... Inspection table, 3 ... Recessed part, 5 ... Illumination apparatus, 8 ... Color imaging apparatus, 8a ... Image sensor, 10 ... Image processing apparatus, 11 ... RGB image data generation part, 12 ... Brightness average value Calculation unit, 12a ... Low luminance side inclination calculation unit, 12b ... High luminance side inclination calculation unit, 12c ... Minimum luminance value calculation unit, 12d ... Maximum luminance value calculation unit, 12e ... Maximum frequency component calculation unit, 13 ... Average value comparison , 13a ... Low luminance side inclination comparison unit, 13b ... High luminance side inclination comparison unit, 13c ... Minimum luminance value comparison unit, 13d ... Maximum luminance value comparison unit, 13e ... Maximum frequency component comparison unit, 14 ... Foreign substance determination processing unit 15a, 31a, 41a, 51a ... red density histogram, 21b, 31b, 41b, 51b ... small luminance distribution, 22a, 32a, 42a, 52a ... green density histogram, 22b, 3 b, 42b, 52b ... small luminance distribution, 23a, 33a, 43a, 53a ... blue density histogram, 23b, 33b, 43b, 53b ... small luminance distribution, LT ... look-up table, Avr, Avg, Avb ... luminance average value, Bk1, Bk2, Bk3, Bk4 ... Threshold, Bmax4r, Bmax4g, Bmax4b ... Maximum brightness value, Bmin3r, Bmin3g, Bmin3b ... Minimum brightness value, CP ... Semiconductor chip, Drf ... Red filter image data, Dgf ... Green filter image data, Dbf ... Blue filter image data, Fr, Fg, Fb ... Maximum frequency components, GD ... Color image data (captured image data), S1, S2, S3 ... Inclination (low luminance side), S4, S5, S6 ... Inclination (high luminance) side).

Claims (10)

An illumination unit that emits light to the workpiece;
An imaging unit for imaging a predetermined area of the workpiece;
In an appearance inspection apparatus having an image processing unit that inputs imaged image data by the imaging unit and performs image processing to detect a foreign object defect of the workpiece,
The image processing unit
RGB image data generation unit for creating red, green, and blue filter image data from the color image data obtained by the imaging unit;
A feature amount extraction unit that calculates a feature amount of an inspection region from each of the red, green, and blue filter image data;
A filter image that inputs the feature values of the filter image data of red, green, and blue extracted by the feature amount calculation unit, compares the feature amounts, and selects the optimum filter image data that is easy to detect foreign matter. A visual inspection apparatus, comprising: a comparison unit, wherein foreign object defects are detected by image processing using filter image data selected by the filter image comparison unit. The appearance inspection apparatus according to claim 1,
The feature amount extraction unit is a luminance average value calculation unit that calculates a luminance average value of an inspection region as a feature amount from the filter image data of red, green, and blue,
The filter image comparison unit is configured to obtain an optimum filter image based on a luminance average value of each of the red, green, and blue filter image data calculated by the luminance average value calculation unit and a lookup table created in advance. An appearance inspection apparatus, which is an average value comparison unit for selecting data. The appearance inspection apparatus according to claim 2,
The filter image comparison unit, based on the average brightness value of each of the red, green, and blue filter image data calculated by the average brightness value calculation unit, and a lookup table created in advance, the color of the work An appearance inspection apparatus characterized by determining a defect as a taste abnormality. The appearance inspection apparatus according to claim 1,
The feature amount extraction unit creates a density histogram of an inspection region from each of the red, green, and blue filter image data, and calculates a low brightness value-side slope as a feature amount from each density histogram A side inclination calculator,
The filter image comparison unit compares the gradients on the low luminance value side of the density histograms of the red, green, and blue filter image data calculated by the low luminance side gradient calculation unit, and the absolute value of the gradient An appearance inspection apparatus, which is a low-luminance side inclination comparison unit that selects filter image data having the largest filter image data as optimal filter image data for detecting low-luminance foreign matter. The appearance inspection apparatus according to claim 1,
The feature amount extraction unit creates a density histogram of an inspection region from each of the red, green, and blue filter image data, and calculates a gradient on the high brightness value side as a feature amount from each density histogram A side inclination calculator,
The filter image comparison unit compares the gradients on the high luminance value side of the gray histograms of the red, green, and blue filter image data calculated by the high luminance side inclination calculating unit, and calculates the absolute value of the inclination. A high-intensity side inclination comparison unit that selects filter image data having the largest filter image data as optimum filter image data for detecting high-intensity foreign matter. The appearance inspection apparatus according to claim 1,
The feature amount extraction unit includes:
A low-intensity side inclination calculation unit that creates a density histogram of an inspection region from each of the red, green, and blue filter image data, and calculates the inclination on the low-luminance value side as a feature amount from each of the density histograms;
A high-intensity-side inclination calculating unit that creates a density histogram of an inspection region from each of the red, green, and blue filter image data, and calculates the inclination on the high-luminance value side as a feature amount from each of the density histograms. ,
The filter image comparison unit
Comparing the slopes on the low brightness value side of the density histograms of the red, green, and blue filter image data calculated by the low brightness side slope calculation unit, the filter image data having the largest absolute value of the slope is obtained. A low-luminance side slope comparison unit that selects filter image data that is optimal for detecting low-luminance foreign matter,
Comparing the slopes on the high brightness value side of the density histogram of the red, green, and blue filter image data calculated by the high brightness side slope calculation unit, the filter image data having the largest absolute value of the slope is obtained. An appearance inspection apparatus comprising: a high-luminance side inclination comparing unit that selects filter image data that is optimal for detecting high-luminance foreign matter. The appearance inspection apparatus according to claim 1,
The feature quantity extraction unit creates a density histogram of an inspection area from each of the red, green, and blue filter image data, and calculates a minimum brightness value as a feature quantity from each density histogram. And
The filter image comparison unit compares the minimum luminance values of the density histograms of the red, green, and blue filter image data calculated by the minimum luminance value calculation unit, and the filter image having the largest minimum luminance value. An appearance inspection apparatus, which is a minimum luminance value comparison unit that selects data as optimum filter image data for detecting low-luminance foreign matter. The appearance inspection apparatus according to claim 1,
The feature quantity extraction unit creates a density histogram of an inspection region from each of the red, green, and blue filter image data, and calculates a maximum brightness value as a feature quantity from each density histogram And
The filter image comparison unit compares the maximum luminance values of the density histograms of the red, green, and blue filter image data calculated by the maximum luminance value calculation unit, and the filter image having the smallest maximum luminance value. An appearance inspection apparatus, which is a maximum luminance value comparison unit that selects data as optimum filter image data for detecting high-intensity foreign matter. The appearance inspection apparatus according to claim 1,
The feature amount extraction unit includes:
A minimum brightness value calculation unit that creates a density histogram of an inspection area from each of the red, green, and blue filter image data, and calculates a minimum brightness value as a feature amount from each density histogram;
A maximum brightness value calculation unit that creates a density histogram of an inspection region from each of the filter image data of red, green, and blue, and calculates a maximum brightness value as a feature amount from each density histogram, and
The filter image comparison unit
The minimum luminance value calculated by the minimum luminance value calculating unit is compared with the minimum luminance value of the density histogram of each filter image data of red, green, and blue, and the filter image data having the largest minimum luminance value is compared with the low luminance foreign matter. A minimum luminance value comparison unit that is selected as filter image data that is optimal for detection;
The maximum luminance value of the red, green, and blue filter image data calculated by the maximum luminance value calculation unit is compared, and the filter image data having the smallest maximum luminance value is compared with the high luminance foreign matter. An appearance inspection apparatus comprising: a maximum luminance value comparison unit that is selected as filter image data that is optimal for detection. The appearance inspection apparatus according to claim 1,
The feature amount extraction unit calculates a Fourier spectrum of an inspection region from each filter image data of red, green, and blue, and calculates a maximum frequency component as a feature amount from the Fourier spectrum of each filter image data. A frequency component calculator,
The filter image comparison unit compares the maximum frequency components of the Fourier spectrum of the red, green, and blue filter image data calculated by the maximum frequency component calculation unit, and the filter having the largest maximum frequency component An appearance inspection apparatus, which is a maximum frequency component comparison unit that selects image data as filter image data that is optimal for detecting foreign matter.

Images