JPH10325806A - Method and apparatus for inspecting external appearance - Google Patents

Abstract

PROBLEM TO BE SOLVED: To rapidly and accurately execute an external appearance inspection of a fine pattern by comparing inspection image feature data generated based on an intensity difference of pixels of an inspection pixel region with reference image feature data formed based on an inspection difference of pixels of a reference pixel region at each pixel. SOLUTION: A partial pattern of an object 101 to be inspected by a photosensor 103 is imaged, the other partial pattern is imaged by a photosensor 106, and image data to be inspected and the reference image data are respectively stored in image memories 105 and 108. The stored images are aligned by a matching unit 109, and the matched images are respectively stored in image memories 110 and 111. A feature extractor 112 extracts the respective feature data, and stores them in a feature data map format corresponding to each pixel in feature memories 113 and 114. A comparison deciding unit 115 compares a feature data map to be compared with a reference feature data map at each pixel and decides whether it allows within an allowable range or not.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a visual inspection of an object to be inspected, and more particularly, to a method and an apparatus for inspecting a fine pattern using image processing.

[0002]

2. Description of the Related Art As an appearance inspection method using image processing, a method of extracting a characteristic amount such as an area of a defective portion from an image of a test object and comparing the characteristic amount with a predetermined value, or registering a non-defective image as a template in advance. In addition, a method of performing comparison by a residual sequential test method, normal correlation matching, or the like is generally adopted. However, with such a general inspection method, it has been extremely difficult to inspect a fine pattern such as a wafer, a reticle, or a photomask for LSI manufacture at high speed and with high accuracy. This is because, in the appearance inspection of a fine pattern, very strict accuracy is required, so that it is necessary to increase the resolution of an image, which increases the amount of inspection information and requires a lot of time for data processing.

As a method corresponding to such miniaturization of an LSI pattern, there is an inspection method disclosed in Japanese Patent Publication No. Hei 8-10463. In this inspection method, a method is employed in which a difference between a reference image and a test image is squared, and a value obtained by integrating in a region near a target pixel is compared with a preset value to make a determination. ing.

[0004]

However, in the above-described inspection method, since integration is performed in a pixel vicinity area and comparison is made, a minute defect of one pixel or less may not be detected. Furthermore, since the difference between the reference image and the test image is squared, the load on the image processor increases, which hinders an improvement in processing speed. Furthermore, when the brightness of the images to be compared has a small difference uniformly in the whole area or in the area near the target pixel, a large difference occurs because the integration is performed in the area near the target pixel. It may be detected as a defect.

In the case of inspecting a fine pattern such as a wafer, a reticle, or a photomask for manufacturing an LSI, there is often a defect extending between inspection screens. Even if one defect extends over a plurality of screens, the above-described inspection method recognizes the defect as a plurality of defects.

SUMMARY OF THE INVENTION An object of the present invention is to provide a method and an apparatus capable of performing a fine pattern appearance inspection at high speed and with high accuracy.

[0007]

SUMMARY OF THE INVENTION A visual inspection method and apparatus according to the present invention receive an inspection image of an object and a reference image corresponding to the inspection image, and for each pixel of the inspection image, the inspection pixel and its neighboring pixels. Inspection image feature data is generated based on the luminance difference between the pixels in the inspection pixel region consisting of, and for each pixel of the reference image, based on the luminance difference between the pixels in the reference pixel region consisting of the reference pixel and its neighboring pixels Generate reference image feature data. Then, the defect of the object is determined by comparing the inspection image feature data with the reference image feature data for each pixel.

[0008]

FIG. 1 is a schematic block diagram of an inspection apparatus for carrying out a visual inspection method according to the present invention. LS
I A test object 101 having a fine pattern such as a wafer, a reticle, or a photomask for manufacturing is irradiated by a scanning beam from a light source 102, and an image of the test object 101 and an inspection are inspected based on the reflected light or transmitted light. An inspection reference image serving as a reference is captured.

Here, a part of the pattern of the object 101 is converted into an electric signal by the optical sensor 103, and A
After being converted into 8-bit grayscale data by the / D converter 104, the data is stored in the image memory 105 as inspection target image data. On the other hand, the A / D converter 1 captures an image of another part of the inspection subject 101 having the same pattern as the inspection part by the optical sensor 106.
After being converted into 8-bit gray scale data by 07, the image data is stored in the image memory 108 as reference image data. In other words, the optical sensor 103 captures the inspection target image and the optical sensor 106 captures the reference image using different portions of the test subject 101 having the same pattern. Of course, a non-defective object having the same pattern as the object 101 to be inspected may be prepared and the reference image may be captured by the optical sensor 106.

The image data of the test object 101 can be captured by scanning the laser beam from the light source 102 and moving the test object 101 (for example, moving by an XY stage). Alternatively, the optical sensor 1
If a two-dimensional (area) sensor is used as 03 and 106, image data of each part can be captured while moving the test subject 101.

The inspection target image and the reference image stored in the image memories 105 and 108 are aligned by the matching unit 109 so that the respective pixels correspond to each other. These are stored in the image memory 110 and the reference image memory 111, respectively. The inspection target image and the reference image matched in this way are input, and as described later, the feature extraction unit 112 extracts the respective feature data, and outputs the target feature memory 113 and the target feature memory 113 in the form of a feature data map corresponding to each pixel. Each is stored in the reference feature memory 114.

The comparison / determination unit 115 compares the target feature data map and the reference feature data map for each pixel, and determines whether or not the difference is within an allowable range set in the allowable value memory 116 in advance. That is, if the difference corresponding to a certain pixel is within the allowable value, it is determined to be “good”, and if it exceeds the allowable value, it is determined to be “bad”. Will be displayed.

The control unit 118 includes a program memory 119
The operation of the entire inspection apparatus is controlled according to the inspection program stored in the storage device. For example, the scanning timing of the laser beam from the light source 102, the XY
The control of the stage, the timing control and the input / output control of each functional unit and the memory described above are executed. Each of the above functional units can be realized as hardware by a logic circuit, but can also be realized as software using a program control processor such as a DSP. Needless to say, a higher speed is achieved by using a logic circuit.

FIG. 2 is a schematic block diagram showing another example of the inspection apparatus for performing the appearance inspection method according to the present invention. As shown in the figure, the design data is stored in the design data memory 120 in advance, and the control unit 118 reads out the design data of the same part as the image to be inspected, develops it into one image, and stores it in the image memory 108. It may be stored as a reference image. In other words, the reference image stored in the image memory 108 can be captured by the optical sensor 106 as described above. However, an ideal pattern is generated from the design data of the test object 101 and used as the reference image. Can also.

Next, an embodiment of the inspection method according to the present invention will be described in detail.

(First Embodiment) According to the first embodiment of the present invention, the feature extracting unit 112 determines the horizontal (x-direction), vertical (y-direction), and oblique direction for each pixel of the inspection target image or the reference image. Are calculated, and their absolute values are added in each direction to generate feature data of the pixel.
Then, by executing this operation on each pixel of one image, a feature data map corresponding to each pixel can be generated. Hereinafter, a specific description will be given.

FIG. 3 shows a 5 × 5 neighborhood of the target pixel (x, y).
FIG. 4 is a diagram illustrating a luminance distribution of a pixel region. The luminance value of the target pixel (x, y) located at the center of this distribution map is d (x, y)
It is. The horizontal (x-direction) feature data Fx in the area near the pixel can be obtained by the following equation (Equation 1), that is, by adding the absolute value of the amount of change between pixels.

[0018]

Fx = | d (x-2, y) -d (x-1, y) | + | d (x-1, y) -d (x,
y) | + | d (x, y) -d (x + 1, y) | + | d (x + 1, y) -d (x + 2, y) |.

Similarly, the vertical (y-direction) feature data Fy can be obtained by the following equation (Equation 2).

[0020]

Fy = | d (x, y-2) -d (x, y-1) | + | d (x, y-1) -d (x,
y) | + | d (x, y) -d (x, y + 1) | + | d (x, y + 1) -d (x, y + 2) |.

In the diagonal direction, the characteristic data Fxy + having a positive inclination can be obtained by (Equation 3), and the characteristic data Fxy- having a negative inclination can be obtained by (Equation 4).

[0022]

Fxy + = | d (x + 2, y-2) -d (x + 1, y-1) | + | d (x + 1, y-
1) -d (x, y) | + | d (x, y) -d (x-1, y + 1) | + | d (x-1, y + 1) -d (x-
2, y + 2) |.

[0023]

## EQU4 ## Fxy- = | d (x-2, y-2) -d (x-1, y-1) | + | d (x-1, y-
1) -d (x, y) | + | d (x, y) -d (x + 1, y + 1) | + | d (x + 1, y + 1) -d (x +
2, y + 2) |.

The thus obtained target pixel (x,
The feature data (Fx, Fy, Fxy +, Fxy-) for y) is stored corresponding to the position (x, y) of the target pixel. Hereinafter, similarly, the characteristic data is calculated for each pixel to create a characteristic data map.

The feature extracting unit 112, the inspection object characteristic data map MF _S and the reference characteristic data map MF _REF by performing the operation on the target image and the reference image generated each object characteristic memory 113 and the reference Each is stored in the feature memory 114.

The comparison determination unit 115 reads the inspection object characteristic data map MF _S and the reference characteristic data map MF _REF characteristic data of the corresponding pixels from, for example, x-direction of the object characteristic data F _XS and the reference feature data F _XREF respectively, It is determined by comparing whether or not the difference is within an allowable value.

As described above, since the feature data map is generated by adding the absolute values of the amounts of change in luminance in the vicinity of each pixel, a certain luminance difference exists between the inspection image and the reference image to be compared. Even in this case, an accurate test result can be obtained. Further, since only the absolute value of the amount of change is added, high-speed processing can be performed by only adding or subtracting. For this reason, it is possible to greatly reduce the time required for inspecting a defect such as a wafer, a reticle, or a photomask for manufacturing an LSI, which greatly contributes to an improvement in productivity. Further, this leads to a reduction in the cost of the LSI itself.

(Second Embodiment) According to a second embodiment of the present invention, the feature extracting unit 112 performs first-order horizontal (x-direction) and first-order vertical (y-direction) differentiation for each pixel of an inspection object image or a reference image. Are calculated, and their absolute values are added for each direction to generate feature data of the pixel. And
By executing this operation on each pixel of one image, a feature data map corresponding to each pixel can be generated. Hereinafter, a specific description will be given.

FIG. 4 shows a 3 × 3 neighborhood of the target pixel (x, y).
FIG. 4 is a diagram illustrating a luminance distribution of a pixel region. The luminance value of the target pixel (x, y) located at the center of this distribution map is d (x, y)
It is. The feature data F of the pixel can be obtained by the following equation (Equation 5), that is, by adding the absolute value of the first derivative in each direction.

[0030]

F = | d (x, y-1) -d (x, y + 1) | + | d (x-1, y) -d (x + 1, y) |

The target pixel (x,
The feature data F for y) is stored corresponding to the position (x, y) of the target pixel. Similarly, the above-described feature data is calculated for each pixel, and a feature data map (here, a primary differential data map) can be created.

The feature extraction unit 112 generates the inspection target feature data map MF _S and the reference feature data map MF _REF by performing the above-described calculation on the inspection target image and the reference image, and outputs the target feature memory 113 and the reference feature data map MF _REF. Each is stored in the feature memory 114.

The comparison determination unit 115 reads the inspection object characteristic data map MF _S and the reference characteristic data map MF _REF characteristic data of the corresponding pixels from, for example, x-direction of the object characteristic data F _XS and the reference feature data F _XREF respectively, It is determined by comparing whether or not the difference is within an allowable value.

In order to generate a feature data map by adding the absolute values of the first derivative in the vicinity of each pixel as described above, when a certain luminance difference exists between the inspection image to be compared and the reference image, However, accurate test results can be obtained. Furthermore, since only the absolute value of the first derivative is added, high-speed processing can be performed only by addition and subtraction. For this reason, it is possible to greatly reduce the time required for inspecting a defect such as a wafer, a reticle, or a photomask for manufacturing an LSI, which greatly contributes to an improvement in productivity. Further, this leads to a reduction in the cost of the LSI itself.

(Third Embodiment) According to a third embodiment of the present invention, the feature extracting unit 112 calculates the second derivative for each pixel of the inspection target image or the reference image, and calculates the absolute value of the second derivative. Is generated as feature data of the pixel. Then, by performing this operation on each pixel of one image,
A feature data map corresponding to each pixel can be generated. Hereinafter, a specific description will be given.

In FIG. 4, the characteristic data F of the target pixel (x, y) is calculated by the following equation (Equation 6), that is, the second derivative is calculated using the pixels in the neighboring area, and is calculated as its absolute value. Can be.

[0037]

F = | d (x, y-1) + d (x, y + 1) + d (x-1, y) + d (x + 1, y)-
4d (x, y) |.

The target pixel (x,
The feature data F for y) is stored corresponding to the position (x, y) of the target pixel. Similarly, the above-described feature data is calculated for each pixel, and a feature data map (here, a secondary differential data map) can be created.

The feature extracting unit 112, the inspection object characteristic data map MF _S and the reference characteristic data map MF _REF by performing the operation on the target image and the reference image generated each object characteristic memory 113 and the reference Each is stored in the feature memory 114.

The comparison determination unit 115 reads the inspection object characteristic data map MF _S and the reference characteristic data map MF _REF characteristic data of the corresponding pixels from, for example, x-direction of the object characteristic data F _XS and the reference feature data F _XREF respectively, It is determined by comparing whether or not the difference is within an allowable value.

As described above, since the feature data map is generated by adding the absolute values of the second derivative in the vicinity of each pixel, a certain luminance difference exists between the inspection image and the reference image to be compared. Even in this case, an accurate test result can be obtained. Further, since only the absolute value of the second derivative is added, high-speed processing can be performed by only adding and subtracting. For this reason, it is possible to greatly reduce the time required for inspecting a defect such as a wafer, a reticle, or a photomask for manufacturing an LSI, which greatly contributes to an improvement in productivity. Further, this leads to a reduction in the cost of the LSI itself.

(Fourth Embodiment) FIG. 5 is a schematic block diagram of an inspection apparatus according to a fourth embodiment of the present invention. However, the same functional units as those of the inspection apparatus shown in FIG.

In FIG. 5, a direction pattern template memory 121 is provided in the inspection apparatus, and a plurality of direction pattern templates (luminance patterns) are set in advance. The feature extraction unit 112 reads a pixel region in the vicinity of each pixel from the target image memory 110 or the reference image memory 111, selects a template that is most similar to the directional pattern template, and outputs its registration number as feature data. By executing this comparison and selection on each pixel of one image, a feature data map corresponding to each pixel can be generated. Hereinafter, a specific description will be given.

FIG. 6 is a diagram exemplifying template types stored in the direction pattern template memory 121. As shown in the figure, several direction patterns are registered together with registration numbers in the direction pattern template memory 12.
1 is stored. Here, the size of the template is 3 × 3 pixels. Referring to the 3 × 3 pixel area shown in FIG. 4, the direction pattern # 1 shown in FIG. 5 is, for example, the luminance of the pixels in the first column, that is, d (x-1, y-1), d (x-1, y ) And d (x-1, y +
1) is low, the luminance of the pixels in the second column and the luminance of the pixels in the third column,
That is, d (x, y-1), d (x, y) and d (x, y + 1) and d (x + 1, y-1), d (x +
1, y) and d (x + 1, y + 1) are set high.

The feature extracting unit 112 is used for the target image memory 11
0 or a 3 × 3 pixel area near each pixel is read from the reference image memory 111, the most similar template is selected by comparing with the direction pattern template, and the registration number is output as feature data. And this comparison selection is 1
By executing the process on each pixel of one image, a feature data map corresponding to each pixel can be generated. Thus,
Inspected characteristic data map MF _S and the reference characteristic data map MF _REF is stored in each of the target features memory 113 and the reference feature memory 114.

The comparison determination unit 115 reads the inspection object characteristic data map MF _S and the reference characteristic data map MF _REF characteristic data of the corresponding pixel from, that direction pattern registration number, respectively, when their registration numbers are different,
The corresponding direction pattern is read from the memory 121 and compared, and it is determined whether or not the difference between them is within an allowable value.

As described above, the direction pattern of the most similar template is selected by comparing the vicinity area of each pixel with the preset direction pattern template, and these patterns are compared for each pixel of the target screen and the reference screen for inspection. Therefore, even when a certain luminance difference exists between the inspection image to be compared and the reference image, an accurate inspection result can be obtained. Furthermore,
Even in the selection of the optimal direction pattern and the comparison of the direction patterns, high-speed processing can be performed by performing only addition and subtraction as calculations. For this reason, it is possible to greatly reduce the time required for inspecting a defect such as a wafer, a reticle, or a photomask for manufacturing an LSI, which greatly contributes to an improvement in productivity. Further, this leads to a reduction in the cost of the LSI itself.

(Fifth Embodiment) FIG. 7 is a partial block diagram of an inspection apparatus according to a fifth embodiment of the present invention. The inspection device shown in the figure has a configuration in which a pattern classification unit 201 and an allowable value offset setting unit 202 are further added to the inspection device shown in FIG. 1, 2 or 5.

The pattern classifying unit 201 reads a pixel region in the vicinity of each pixel from the target image memory 110 and the reference image memory 111, and obtains a difference between the maximum luminance and the minimum luminance in the pixel region in the vicinity of the pixel. According to the magnitude of the difference, the luminance change of the portion where the pixel is located is classified into a plurality of predetermined patterns. For example, a neighboring pixel region as shown in FIG. 3 or FIG. 4 is read out for a certain pixel, and the neighboring pixel region is defined as an edge portion depending on the difference between the maximum luminance and the minimum luminance in the neighboring pixel region. It is determined whether the luminance change is a sharp part or a flat part with little luminance change.

The allowable value offset setting unit 202 compares the classification patterns of the respective pixels of the target image and the reference image thus obtained, and compares and determines the offset value for adjusting the setting of the allowable value according to the magnitude of the difference. Output to the unit 115. For example, if the classification pattern of a certain pixel of the target image and the reference image is the same, set the offset value to zero,
If they are different, the offset value is increased. Comparison determination unit 115 reads the characteristic data of the corresponding pixels from the examined object characteristic data map MF _S and the reference characteristic data map MF _REF (variation, first derivative, second derivative, or direction pattern feature data), respectively, further The offset value is input from the allowable value offset setting unit 202, and it is determined by comparing whether or not the difference of the feature data is within the offset set allowable value. Since the allowable value is adjusted by the offset value, it is possible to adjust the comparison / determination result depending on whether the pixel is in a portion where the luminance change is large or small.

As described above, since the inspection can be performed by changing the preset allowable value according to the state of the luminance change in the vicinity of each pixel, an accurate inspection result can be obtained, The defect inspection time of a wafer, a reticle, a photomask, or the like for manufacturing an LSI can be significantly reduced, which greatly contributes to an improvement in productivity. Further, this leads to a reduction in the cost of the LSI itself.

(Sixth Embodiment) FIG. 8 is a partial block diagram of an inspection apparatus according to a sixth embodiment of the present invention. The inspection apparatus shown in FIG. 11 further includes a connection processing unit 301 in addition to the inspection apparatus shown in FIG. 1, FIG. 2, FIG. 5 or FIG.
And an external interface 117. In accordance with the control timing from the control unit 118, the connection processing unit 301 detects a defect extending over adjacent inspection result images based on the inspection result for each image input from the comparison and determination unit 115, and connects the defects to one another. It has the function of organizing as one defect. Hereinafter, a specific description will be given.

FIG. 9 is a diagram schematically showing a case where there is a defect extending over an adjacent inspection screen. For example, when a defect 302 is detected on the entire right end of a certain inspection screen N based on the inspection result input from the comparison determination unit 115, the connection processing unit 301 checks the adjacent inspection screen N + 1.
It is presumed that the defect 302 is also continuous at the left end of.
Then, when the existence of the defect 302 at the left end of the subsequent inspection screen N + 1 is confirmed, the link information of the defect 302 is output to sort the defect data. Such a linking process is performed in the x direction and the y direction, and a defect existing over a plurality of inspection screens is also recognized as one defect.

As described above, in order to recognize a defect continuing over a plurality of screens as one defect, the confirmation work performed by the operator after the completion of the inspection is greatly reduced. The time required for inspecting defects such as a reticle or a photomask can be greatly reduced, which greatly contributes to improvement in productivity. Further, this leads to a reduction in the cost of the LSI itself.

[Brief description of the drawings]

FIG. 1 is a schematic block diagram of an inspection apparatus for performing a visual inspection method according to the present invention.

FIG. 2 is a schematic block diagram showing another example of the inspection apparatus for performing the appearance inspection method according to the present invention.

FIG. 3 is a diagram showing a luminance distribution in a 5 × 5 pixel area near a target pixel (x, y).

FIG. 4 is a diagram showing a luminance distribution in a 3 × 3 pixel area near a target pixel (x, y).

FIG. 5 is a schematic block diagram of an inspection device according to a fourth embodiment of the present invention.

FIG. 6 is a diagram illustrating template types stored in a direction pattern template memory 121;

FIG. 7 is a partial block diagram of an inspection device according to a fifth embodiment of the present invention.

FIG. 8 is a partial block diagram of an inspection device according to a sixth embodiment of the present invention.

FIG. 9 is a diagram schematically showing a case where a defect that extends over an adjacent inspection screen exists.

[Explanation of symbols]

 Reference Signs List 101 subject 102 light source 103 optical sensor 104 A / D converter 105 image memory 106 optical sensor 107 A / D converter 108 image memory 109 matching unit 110 target image memory 111 reference image memory 112 feature extraction unit 113 target feature memory 114 Reference feature memory 115 Comparison / determination unit 116 Permissible value memory 117 External interface 118 Control unit 119 Program memory 120 Design data memory 121 Directional pattern template memory 201 Pattern classification unit 202 Permissible value offset setting unit 301 Link processing unit 302 Defect unit

Claims (17)

[Claims] 1. An inspection apparatus for inspecting the appearance of an object by image processing, an inspection input unit for inputting an inspection image of the object, a reference input unit for inputting a reference image corresponding to the inspection image, and the inspection image A first feature extraction unit that generates test image feature data based on a luminance difference between pixels in a test pixel region including the test pixel and its neighboring pixels, A second feature extraction unit that generates reference image feature data based on a luminance difference between pixels in a reference pixel region including a pixel and its neighboring pixels; and compares the inspection image feature data with the reference image feature data for each pixel. And a determination unit for determining a defect of the object by performing the inspection. 2. The method according to claim 1, wherein the first feature extracting unit generates the inspection image feature data based on an absolute value of a luminance difference between pixels in the inspection pixel region. The appearance inspection apparatus according to claim 1, wherein the reference image feature data is generated based on an absolute value of a luminance difference between pixels. 3. The first feature extraction unit generates the inspection image feature data by adding absolute values of luminance change amounts in a plurality of directions in the inspection pixel region. The appearance inspection apparatus according to claim 2, wherein the reference image feature data is generated by adding absolute values of the luminance change amounts in the plurality of directions in a reference pixel area. 4. The first feature extraction unit generates the inspection image feature data by adding the absolute values of the primary differential values of the inspection pixel in the x direction and the y direction. The appearance inspection apparatus according to claim 2, wherein the reference image feature data is generated by adding absolute values of first derivatives of the reference pixel in the x direction and the y direction. 5. The first feature extracting unit adds second derivative values in the x direction and the y direction of the inspection pixel and generates the inspection image feature data from its absolute value. 3. The appearance inspection apparatus according to claim 2, wherein the reference image feature data is generated by adding absolute values of the second derivative in the x direction and the y direction of the reference pixel. 6. A pattern storage unit for storing a plurality of registered luminance patterns in advance, wherein the first feature extracting unit selects a registered luminance pattern most similar to a luminance pattern in the inspection pixel area to perform the inspection. The image feature data is generated, and the second feature extracting unit generates the reference image feature data by selecting a registered brightness pattern most similar to a brightness pattern in the reference pixel area. Visual inspection device as described. 7. The determining means calculates a difference between the inspection image feature data and the reference image feature data for each pixel,
The visual inspection apparatus according to claim 1, wherein the defect is determined for the object by comparing the difference with a predetermined allowable value. 8. A maximum luminance difference in an inspection pixel region including the inspection pixel and its neighboring pixels for each pixel of the inspection image, and a maximum luminance difference in a reference pixel region including the reference pixel and its neighboring pixels for each pixel of the reference image. A maximum luminance difference detecting means for detecting a maximum luminance difference, calculating a difference between a maximum luminance difference in the inspection pixel area and a maximum luminance difference in the reference pixel area, and changing the allowable value according to the magnitude of the difference The visual inspection apparatus according to claim 7, further comprising: an allowable value adjusting unit. 9. The inspection input unit inputs the appearance of the object by a plurality of inspection images, the reference input unit inputs a plurality of reference images corresponding to the plurality of inspection images, and the determination unit determines the inspection. 9. The visual inspection device according to claim 1, wherein a defect is determined for each image. 10. In a first inspection image and a second inspection image adjacent to each other, a defect existing at an end adjacent to the first inspection image and a second inspection image are determined.
10. The visual inspection apparatus according to claim 9, further comprising a connection processing unit that connects a defect existing at an adjacent end of the inspection image as one defect. 11. An inspection method for inspecting the appearance of an object by image processing, comprising: inputting an inspection image of the object; inputting a reference image corresponding to the inspection image; Inspection image feature data is generated based on a luminance difference between pixels in an inspection pixel region including the neighboring pixels. For each pixel of the reference image, luminance between pixels in the reference pixel and a reference pixel region including the neighboring pixels is generated. A visual inspection method comprising: generating reference image feature data based on a difference; and comparing the inspection image feature data with the reference image feature data for each pixel to determine a defect of the object. 12. The inspection image feature data is generated by adding absolute values of luminance change amounts in a plurality of directions in the inspection pixel region, and the reference image feature data is a luminance in the plurality of directions in the reference pixel region. The visual inspection method according to claim 11, wherein the visual inspection method is generated by adding an absolute value of a change amount. 13. The inspection image feature data is generated by adding the absolute values of the primary differential values in the x and y directions of the inspection pixel, and the reference image feature data is generated in the x and y directions of the reference pixel. The appearance inspection method according to claim 11, wherein the appearance inspection method is generated by adding an absolute value of the first derivative. 14. The inspection image feature data is generated by adding absolute values of second derivative values in the x direction and the y direction of the inspection pixel, and the reference image feature data is generated in the x direction and the y direction of the reference pixel. 12. The appearance inspection method according to claim 11, wherein the method is generated by adding the absolute value of the second derivative of 15. A method for storing a plurality of registered luminance patterns in advance, wherein the inspection image feature data is generated by selecting a registered luminance pattern most similar to a luminance pattern in the inspection pixel area; The visual inspection method according to claim 11, wherein the reference image is generated by selecting a registered luminance pattern most similar to the luminance pattern in the reference pixel area. 16. A method for calculating a difference between the inspection image feature data and the reference image feature data for each pixel, and comparing the difference with a predetermined allowable value to determine a defect of the object. The appearance inspection method according to any one of claims 11 to 15, wherein: 17. A maximum luminance difference in an inspection pixel region including the inspection pixel and its neighboring pixels for each pixel of the inspection image, and a maximum luminance difference in a reference pixel region including the reference pixel and its neighboring pixels for each pixel of the reference image. Detecting a maximum luminance difference, calculating a difference between the maximum luminance difference in the inspection pixel area and the maximum luminance difference in the reference pixel area, and changing the allowable value according to the magnitude of the difference. The appearance inspection method according to claim 16.