JPH03194415A - Device and method for pattern inspection - Google Patents

Abstract

PURPOSE:To remarkably shorten a generation time by providing the device itself with a function for generating the majority of a dictionary automatically when the dictionary is generated although inspection references suitable for respective patterns and the best inspection dictionary are necessary for the inspection of many kinds of printed board. CONSTITUTION:Image information which is obtained by scanning a pattern and then performing binary coding is inputted from a binary-coded image pattern generation part 9 to an image measurement part 1 and the obtained code information is stored in a storage means 2 with a memory buffer or register function and also supplied to a comparator 8. The means 2 classifies and inputs the patterns to a generation frequency counting means 3 by pieces of numerical information, its output is compared with the reference values, and the result is sent to a normal/defective judging means 4. Code information which is judged as normal information is registered in the inspection dictionary 7 consisting of a RAM through a dictionary data generation part 5 consisting of a nonvolatile memory and a dictionary rewriting part 6. Information which is judged as defective information through the comparison part 8 is rejected and not registered.

Description

[Detailed Description of the Invention] [Summary] The present invention relates to a pattern inspection device and a pattern inspection method, and provides a method for easily creating a dictionary for inspection in a short time regardless of the ability of the dictionary creator. means for scanning a pattern; means for measuring a unit pattern shape at a predetermined interval while scanning the pattern; and means for converting the measured unit pattern shape into numerical information according to predetermined conditions. , a storage means for storing all of the numerical information or excluding information previously registered in a normal dictionary, and a storage means for storing the numerical information stored in the storage means for each numerical information representing the same unit pattern shape. occurrence frequency counting means for counting the occurrence frequency for each of the classified unit pattern shapes; means for comparing the occurrence frequency for each of the classified unit pattern shapes with a predetermined reference value to determine whether the information is normal information or defect information; and the normal information. display means for displaying one of the unit pattern shapes determined to be defect information; selection means for selecting normal information or defect information for the unit pattern shape displayed on the display means; The apparatus is configured to include: means for registering numerical information regarding the unit pattern shape determined as defect information by the selection means in a defect dictionary; and means for inspecting an inspection pattern using the defect dictionary.

[Industrial application field]

The present invention relates to a pattern inspection method and a pattern inspection apparatus, and more particularly to a pattern inspection method and apparatus for visual inspection of wiring patterns of printed circuit boards and the like.

[Conventional technology]

There are a wide variety of wiring patterns on printed circuit boards, etc., and visual inspection equipment that inspects all parts of these patterns to see if they are normal or have defects can accurately determine whether the pattern is normal or defective. judgment is required.

For this reason, the inspection logic of the visual inspection apparatus needs to be configured according to the pattern of the printed board to be inspected, and also needs to be rearranged as the target pattern changes.

Also, since the printed circuit boards in general use are quite large, a huge amount of data must be processed in order to inspect the entire board surface. Otherwise, it would take extra time and cause inefficiency.

Another problem with such inspections is that, in addition to the fact that the shape of the pattern differs from product to product as mentioned above, the inspection standards of the users who use the printed circuit boards also differ from user to user. It was necessary to assemble inspection logic based on corresponding inspection standards, and it took a considerable amount of time to create a dictionary for inspection.

For example, when inspecting whether the wiring of a printed board formed with a certain pattern is normal or has a defective part, the pattern image is recognized while irradiating an inspection beam onto the printed board using an appropriate scanning means, The pattern image is coded into numerical information at predetermined intervals, and the coded numerical information is compared with a pre-prepared inspection dictionary to determine whether the specific pattern part is normal or has a defect. The purpose was to determine whether the

An example of such inspection logic is disclosed in Japanese Patent Application Laid-Open No. 62-26340.
The inspection logic related to visual inspection shown in 4 is utilized. In this inspection logic, the printed circuit board to be inspected is scanned by an appropriate scanning means and the pattern image is measured at predetermined intervals, and the image is scanned radially in, for example, four directions from a certain point on the image. After measuring and converting each direction and length into binary information according to predetermined rules,
This is inspection logic that determines whether a pattern is normal or defective based on the state of data combinations in each direction.

In this method, the information for determining whether a pattern is normal or defective is determined by a skilled worker who visually observes the shape of the pattern in the relevant area and determines which numerical information is considered normal or defective. , this determination information is stored in a rewritable memory (RAM, etc.).

Therefore, it is necessary to rewrite this data each time a different printed board is inspected and perform the inspection using the most suitable determination information.

Conventionally, the method for creating this judgment information is for a skilled worker to extract normal patterns for each printed board with different patterns.
After converting it into a numerical value, it is stored in a non-volatile memory called a dictionary (
Inspection data was created by inputting it onto a floppy disk (such as a floppy disk).

In other words, in the past, each printed board with a different wiring pattern was inspected, and the inspection execution standards (for example, users who adopted standards that recognized all defects as defects) were inspected for each printed board with a different wiring pattern. It is necessary to create test dictionaries with different test logics depending on the differences in the criteria (some users may adopt criteria that exclude them from the target), but anyone can easily create such a dictionary. It takes a huge amount of time because it is not something that can be drawn, but is created by a skilled person who visually inspects each pattern and makes decisions while considering the pattern.

Moreover, as mentioned above, in order to inspect the entire printed board from the front, it is necessary to process a huge amount of data, which not only takes time to create a dictionary for judgment, but also requires the capacity of the storage device memory required for processing. The amount of data has become enormous, and the arithmetic processing means and software used to process it have also become complex.

[Problem to be solved by the invention]

The present invention improves the above-mentioned drawbacks of the prior art and provides a method and apparatus that can create a proper testing dictionary in a short time regardless of the proficiency and ability of the creator. It is.

[Means for Solving the Problems] The present invention employs the following technical configuration in order to achieve the above-mentioned objects. That is, the pattern inspection apparatus according to the present invention includes means for scanning a pattern, means for measuring a unit pattern shape at a predetermined interval while scanning the pattern, and measuring the measured unit pattern shape according to predetermined conditions. means for converting into numerical information; storage means for storing all of the numerical information or excluding information previously registered in a normal dictionary; and each of the numerical information stored in the storage means represents the same unit pattern shape. an occurrence frequency counting means for classifying each numerical information and counting the frequency of occurrence for each, and comparing the frequency of occurrence for each classified unit pattern shape with a predetermined reference value to determine whether it is normal information or defect information. means for registering the normal information in a normal dictionary; display means for displaying one of the unit pattern shapes determined to be defect information; and whether the unit pattern shape displayed on the display means is normal information or defect information. a selection means for selecting, a means for registering numerical information regarding the unit pattern shape determined as defect information by the selection means in a defect dictionary;
and means for inspecting the inspection pattern using the defect dictionary, and the pattern inspection method according to the present invention includes means for inspecting the inspection pattern using the defect dictionary. A step of measuring the unit pattern shape at each location at each interval, a step of converting the measured unit pattern shape into numerical information according to predetermined conditions, and a step of storing all the numerical information in the storage means as defect information. , a step of classifying the numerical information stored in the storage means into numerical information representing the same unit pattern shape, and counting the frequency of occurrence for each of the numerical information, and counting the frequency of occurrence for each of the classified unit pattern shapes; determining whether the information is normal or defective by comparing it with a predetermined reference value, and registering the normal information in a normal dictionary; a step of repeating each of the above steps under the condition of storing the numerical information as defect information in the storage means; and after each of the above steps is repeated for all areas of the inspection pattern, the information remaining as defect information is stored in the storage means; The process of leaving it in the means,
a displaying step of individually displaying the unit pattern shapes on the display means in order of occurrence frequency from among the unit pattern shapes determined to be the defect information left in the storage means; The steps of selecting normal information or defect information for the unit pattern shape and registering numerical information regarding the unit pattern shape determined to be defect information in a defect dictionary, displaying and selecting the information remaining in the storage means. It consists of a step of repeating for all of the unit pattern shapes that are determined to be defect information and having an occurrence frequency equal to or higher than a predetermined standard, and a step of inspecting the inspection pattern using the defect dictionary. This is a pattern inspection method.

[For production]

In the present invention, by adopting the above-mentioned technical configuration, it is possible to automatically create most of the inspection dictionary that constitutes the inspection logic in pattern inspection using hardware, which improves judgment. Because it introduces communication between humans and machines to make judgments interactively in delicate areas, that is, gray areas, an extremely accurate inspection dictionary can be used even by workers without specialized circuit knowledge. It can be created easily and in a short time.

Further, in the present invention, the inspection dictionary can be created in a short time without increasing the capacity of the memory or buffer that stores coded information obtained by converting patterns into numerical information.

〔Example〕

Specific examples of the pattern inspection apparatus and pattern inspection method of the present invention will be described in detail below with reference to the drawings.

Although the following specific example shows an example of inspecting whether the wiring of a printed circuit board is normal or defective, the present invention is not limited to inspecting the appearance of such a printed circuit board, but also inspecting the shape of a pattern mask. It is clear that this method can be applied to all methods that process images of parts with multiple patterns, recognize the patterns, and compare them with a reference image to determine whether the image part is normal or defective. be.

First, the basic technology of the pattern recognition technology, that is, the image recognition means and the image processing method used in the pattern inspection method according to the present invention will be explained.

FIGS. 3(a) and 3(b) show examples of typical wiring patterns formed on printed wiring boards, and 30 is what is called a pad or land, and the pattern used here is It has not been specifically considered as a subject for inspection. Further, 31 indicates a straight wiring portion extending from the land, and 32 similarly indicates a wiring portion bent at a certain angle.

Other wiring portions are continuously constructed of various shapes such as right-angled portions and curved portions. Furthermore, although the width of each wiring portion varies from printed board to printed board, it is designed to be approximately constant within one printed board. That is, the normal (non-defective) pattern of a printed board is drawn according to certain rules, and each of these unit patterns occurs repeatedly within the same board.

Next, when inspecting a printed board on which such a pattern is formed, as described above, the pattern on the printed board is scanned using an appropriate scanning means, the image is captured, and the image is captured at the time of scanning. Convert the image of the pattern into binary information, then measure the length in a specific direction and that direction from the binary information, and combine them to recognize the shape of the pattern. A code for each unit pattern. Create. The code, ie, digitized information, created in this way is compared with a reference code group, ie, an inspection dictionary, stored in an appropriate memory to determine whether each unit pattern portion is a normal pattern or a defective pattern.

An outline of such a conventional inspection method is shown in FIG. FIG. 4 is a block diagram of the pattern inspection device. lens 4
By scanning the test sample 40 with a line sensor consisting of a CCD 2 and a CCD 43, the wiring pattern 41 on the sample is converted into an electrical signal.

This signal is binarized by the binarization circuit 44 and then stored in the storage circuit 45. The output of the storage circuit 45 is sent to a length measurement circuit 46, and the output of the length measurement circuit 46 is sent to a reference dictionary storage circuit 47 or a comparison/judgment circuit. Comparison judgment circuit 4
8 is a data from the length measurement circuit 46 and a reference dictionary storage circuit 47;
A comparison is made with the contents of , and the result is output as an output 49 of the defect detection result.

There are particular limitations on the method of converting the image information of such patterns into numerical values, for example, converting them into binary information, and then combining the direction and length information to create numerical information that specifies the shape of each unit pattern, that is, encoding it. Although any suitable method can be employed, as mentioned above, a typical numerical method is the method described in Japanese Patent Application Laid-Open No. 62-263404.

That is, the method disclosed in the publication first determines the center point at each measurement point when scanning the pattern to be inspected, and then scans the pattern in four or eight directions equally angularly divided from the center point. Measure the length and add the measured length to the fifth
For example, concentric circles 51 having three different diameters as shown in the figure
53 for short (s), correct (C), and long (Lon).
g (L) ) and Over (OV)
This is a four-stage judgment.

The example in FIG. 5 shows an example in which the length of the pattern is measured in four directions, and the length in the 0° direction is between the circle 51 with the minimum diameter and the circle 52 with the intermediate diameter. Since there is, it is determined to be C, and the length in the 45° direction is determined to be between the circle 52 with the intermediate diameter and the circle 53 with the maximum diameter, and the 90° direction is the same as the direction in which the wiring pattern is formed. Therefore, it becomes infinite, and of course it is outside the circle 53 with the maximum diameter, so it is 0.
■It is judged that.

Also, the direction of 135° is determined to be L, the same as the direction of 45°. Therefore, the unit pattern at the time of FIG. 5 is coded as a combination of numerical values C, L, OV, and L for each direction.

This combination is stored in a memory as 12-bit information, for example, and used for arithmetic processing.

Such coded information is shown in FIG. 6(a).

As shown in (b), measurements are taken at each point, and each time the shape of the pattern differs, it is stored and accumulated in a memory, buffer, etc. as different coded information.

If the wiring is normal, similar numerical information (i.e., code information) will occur continuously, but as shown in Figure 7(a)
(b) If a defective turn portion as shown in (c) exists, numerical information different from the numerical information generated up to now will be generated. For example, in Figure 7(a), s, s, ov
, s are generated, and in FIG. 7(b), s, s,
The code becomes s, s, and further, in Figure 7 (C), C,
It is conceivable that the code will be an OVC-like code.

As mentioned above, collecting numerical information and creating an appropriate test dictionary requires a huge amount of data processing and judgment by experts, and it takes a considerable amount of time to create a test dictionary. In order to improve Numerical information (code)
This method automatically determines whether the code is normal (high frequency of occurrence) or defective (low frequency of occurrence) by collecting and analyzing the code and calculating the frequency of occurrence of the same code.

That is, in the present invention, after all the patterns of one printed circuit board to be inspected have been inspected, the code information (digitized information) of each unit pattern stored in the memory circuit is converted into the code type (at least four types). For example, if you classify each combination of numerical values, count the frequency of occurrence, and arrange them by frequency of occurrence,
A distribution map as shown in the figure is obtained.

Therefore, a unit pattern with a code having a predetermined frequency of occurrence, for example 100, is selected as the standard The work can be performed automatically using appropriate computing means completely using only hardware. However, it is often difficult to automatically determine whether a unit pattern has a code with an occurrence frequency below the criterion X as being normal or defective, so human judgment is combined to determine whether it is normal or defective. . In other words, the aim is to create a more accurate test dictionary using communication between machines and humans.

In this mode, code information (numeric information) that has not yet been determined to be normal or defective, that is, code information that is still left in the inspection dictionary as defect data, is specified one by one in order of the frequency of occurrence. Part or all of the unit patterns having the code are displayed individually or in combination on a suitable pattern display display.

Since the code information is stored in advance together with the coordinate values, it can be easily displayed on a plotter, CRT screen, etc. In this case, instead of showing only one measurement value at the measurement point mentioned above, by displaying the pattern before and after that measurement point, it is possible to grasp the shape of the area around the part that has the code, so it is possible for humans to It becomes easier to make decisions. A plurality of individuals visually inspect the unit pattern portion having the code displayed using this method and determine whether the pattern is normal or defective.

Another standard value Y(
For example, the codes with 5) are repeated individually, and those determined to be normal are registered in the test dictionary as normal unit patterns as described above.

The defects that remain until the end are registered as defect patterns in the inspection dictionary, and are used to perform the original pattern inspection.

In the present invention, the side on which the reference value Y is determined depends on the existence of a unique pattern whose occurrence frequency is extremely low. If a unique pattern is included, it will be about θ, and if it is not included, a pattern that occurs infrequently can be defined as a defect, so 5
It is enough.

Therefore, if the value is set to about 5, inspection will be performed even if some defects are ignored.

The basic test dictionary creation method according to the present invention has been described above, and a more detailed example for carrying out the present invention will be described below.

FIG. 1 is a principle diagram of a pattern inspection device according to the present invention and 1
FIG. 2 is a block diagram showing a specific example.

The image information that has been binarized by scanning the pattern is sent to a binarized image pattern forming circuit 9 corresponding to the binarizing circuit 44 in FIG.
is input to the image measuring section 1 from The image measurement unit 1 calculates the direction and length at each predetermined interval of scanning a pattern from the input binary data, and simply measures the pattern shape, and converts the measured unit pattern shape into a predetermined and means for converting into numerical information (code information) according to conditions.

Next, the numerical information (code information) is stored in the storage means 2, which is a numerical information collecting section having the function of a memory buffer or register.

When storing, the storage means can register all numerical information as defect information (B^0), or store normal information (GOO) in the normal dictionary section of the inspection dictionary as described later.
It is also possible to exclude the numerical information registered as D) and store and register the remaining numerical information, which is defect information.

In the former case, by registering all the numerical information input as defect information (BAD) in the inspection dictionary 7, which will be described later, all of the numerical information can be stored by comparing with the comparator 8, which will be described later. .

Next, the numerical information stored in the storage means is classified by each piece of numerical information representing the same unit pattern shape, and the occurrence frequency counting means 3 counts the frequency of occurrence for each of the pieces of numerical information. Means 4 is provided for comparing the occurrence frequency of each unit pattern shape with a predetermined reference value to determine whether the information is normal information or defective information.

Numerical information (code information) representing unit patterns determined to be normal by the normality/defective judgment means is transferred to an inspection dictionary formed from a RAM or the like through a dictionary data creation section 5 consisting of a non-volatile memory or the like and a dictionary rewriting section 6. 7 in the normal dictionary section. Since all numerical information is registered as defect information as described above, such registration is executed by rewriting the numerical information as normal. That is, the defect dictionary is updated. At this time, it is also possible to erase the numerical information corresponding to the normal information stored in the buffer of the storage unit 2.

In this way, a normal dictionary in which normal information is registered in the inspection dictionary 7 and a defect dictionary section in which defect information previously designated as defective are registered as described above are formed. Therefore, when numerical information about the unit pattern is input from the image measurement section in the next repeating process, the numerical information indicating the unit pattern registered in the normal dictionary is not stored and is excluded from the defect dictionary by the action of the comparator 8. Only the registered numerical information is stored and goes to 0.

In other words, when inspecting a given printed circuit board in its entirety, a huge amount of data is generated as mentioned above, so a huge amount of memory must be used, but this is not economically viable, so only a limited amount of data is required. Memory must be used. In this case as well, the present invention exhibits extremely useful effects.

That is, in the present invention, one printed board is divided into strips in one direction, and the strip-shaped section at one end is scanned from the top or bottom end to the other end for inspection, and then the adjacent strip-shaped section is scanned for inspection. The entire surface is scanned by sequential scanning.

An example of an inspection method using such a device will be explained according to the flowchart shown in FIG. 2. At the start of FIG. (step a),
Therefore, all of the numerical information from the image measurement unit 1 is stored in the storage unit 2, so the buffer becomes empty immediately after the start (step B), and at that point, the test operation ends (step C). . In other words, the area that can be inspected for the first time is extremely small. Here, as mentioned above, defect information (
All information) is sent to the occurrence frequency counting unit 3 (step d)
Compare the frequency of occurrence with a predetermined reference value to determine whether it is normal or defective (step e). After performing the operation and registering what is determined to be normal information in the normal dictionary (step n), the second Start the second inspection operation (return to step b). At this time, due to the action of the comparator 8, the numerical information indicating the unit pattern registered in the normal dictionary is not input into the storage unit 2, and the same numerical information that was input last time is erased. Therefore, a considerable amount of numerical information can be stored in the storage section 2. In other words, the second inspection area is much larger than the first inspection area. In addition, in the second operation, since the occurrence frequency in the first operation is lower than the predetermined value, the numerical information left in the defect dictionary of the inspection dictionary as defect information is accumulated as it is, so the predetermined value is It is possible that the standard value X may be exceeded. Therefore, when the frequency of occurrence exceeds the predetermined value X, steps d and e are applied in the same manner as described above, and the normal information is rewritten and registered in the normal dictionary, and the normal information is increased.

In this way, the above operation is repeated n times until the entire surface of the pudding tffi is inspected. At this point, no new numerical information will be generated even if the test is repeated any further, and the test ends here. (Step f) If there is no numerical information left in the defect dictionary at this point that exceeds the reference value X of the frequency of occurrence, the inspection is terminated, and the remaining numerical information is taken as a candidate for defect information.

Since such candidate data is of a field in which it is difficult to determine whether it is normal or defective by automatic calculation processing by a machine, the next inspection mode is executed in order to further reduce the candidate data and create a more accurate inspection dictionary. That is, one by one is selected from the defect candidate codes in descending order of frequency of occurrence, a dictionary is created in which only these codes are considered defects, and the inspection is performed. At this time, a human confirms the image of the defective area and determines whether it is normal or defective.

To explain more specifically, first, from among the unit pattern shapes determined to be the defect information left in the storage means, only numerical information indicating the unit pattern shape with the highest frequency of occurrence is stored in the defect dictionary as defect information. Register. (Step g
) This step may use data already in the test dictionary.

Next, the inspection is started (step h), and a part or all of the unit pattern having the numerical information is displayed on a suitable die display (step i). Next, the operator selects and judges whether the unit pattern shape displayed on the display means is normal information or defect information. At the same time, those determined to be normal are registered in the normal dictionary (step 0).

Next, the process returns to step g and the same operation is repeated for the next most frequently occurring numerical information. This operation repeats the display and selection process for all unit pattern shapes that are determined to be defective information and that have an occurrence frequency of 7 or more based on a predetermined criterion, which are left in the storage means.

In this way, the numerical information left in the storage means 2 and the defect dictionary until the end becomes official defect information, which becomes the contents of the inspection dictionary, and this is used to carry out the actual printed board inspection. become.

〔effect〕

In order to inspect a wide variety of printed circuit boards, there are inspection standards that are optimal for each pattern, and inspection dictionaries that are optimal for each pattern are also required.

According to the present invention, when creating this dictionary, the device mill automatically creates most of the inspection dictionary, making it possible to significantly shorten the creation time. Further, it has the effect that even someone without specialized circuit knowledge can easily create a test dictionary. In addition, in creating a test dictionary according to the present invention, communication between humans and machines is introduced to solve the delicate part of judgment, that is, the gray part, and the judgment is carried out in an interactive manner. Even a worker without basic circuit knowledge can easily create the circuit in a short time.

Furthermore, according to the present invention, a test dictionary can be created in a short time without increasing the capacity of the memory or buffer that stores coded information obtained by converting patterns into numerical information.

[Brief explanation of drawings]

FIG. 1 explains the principle of a pattern inspection apparatus according to the present invention and shows a specific example for implementing the present invention. FIG. 2 shows a flowchart for implementing the pattern inspection method according to the present invention. FIG. 3 is a diagram showing an example of a pattern shape. FIG. 4 is a diagram showing an example of a conventional pattern inspection device. FIG. 5 is a diagram showing an example of a method of converting a unit pattern shape into numerical information in a conventional pattern inspection method. FIG. 6 is a diagram in which the numerical information conversion means of FIG. 5 is applied to the pattern shape of FIG. 3. FIG. 7 is a diagram showing an example of a unit pattern having a defective portion. FIG. 8 shows the numerical information (code information) representing the unit pattern and the firing frequency. DESCRIPTION OF SYMBOLS 1... Pattern shape measurement means and numerical information conversion means, 2... Storage means, 3... Occurrence frequency counting means, 4... Means for determining normal information or defect information, 5...
Dictionary data creation unit, 6... Dictionary rewriting unit, 7... Inspection dictionary, 8...
- Comparator, 9... Binarized image pattern forming circuit, 30... Pad section, 31... Wiring section, 40... Sample to be inspected (printed board), 41... Wiring pattern, 42
...Lens, 43...CCD, 44...
・Binarization circuit, 45...Memory circuit, 46...
・Length measurement circuit, 47... Dictionary storage circuit, 48...
Comparison circuit, 49...output circuit. (a) (b) 4] Figure 4 shows the configuration of a conventional pattern inspection device Figure 4 shows an example of a method for converting unit pattern shapes into numerical information
Figure 5: Diagrams (a) and (c) showing the state of measurement of a unit pattern shape having a defective part when the numerical information processing means shown in Figure 5 is applied to each pattern.

Claims (2)

[Claims] 1. means for scanning the pattern; means for scanning the pattern and measuring the shape of the unit pattern at each predetermined interval; means for converting the measured shape of the unit pattern into numerical information according to predetermined conditions; A storage means for storing all information except for information registered in a normal dictionary in advance; a storage means for storing numerical information stored in the storage means for each numerical information representing the same unit pattern shape; an occurrence frequency counting means for counting the occurrence frequency; a means for comparing the occurrence frequency for each classified unit pattern shape with a predetermined reference value to determine whether it is normal information or defect information; and registering the normal information in a normal dictionary. means, display means for displaying one of the unit pattern shapes determined to be defect information; selection means for selecting normal information or defect information for the unit pattern shape displayed on the display means; defect information by the selection means. A pattern inspection device comprising: means for registering numerical information regarding the unit pattern shape determined to be in a defect dictionary; and means for inspecting an inspection pattern using the defect dictionary. 2. A step of scanning a pattern in a part of the inspection pattern and measuring a unit pattern shape at that location at predetermined intervals; a step of converting the measured unit pattern shape into numerical information according to predetermined conditions; A step of storing all numerical information as defect information in a storage means, classifying the numerical information stored in the storage means into each piece of numerical information representing the same unit pattern shape, and counting the frequency of occurrence for each piece of numerical information. a step of comparing the frequency of occurrence of each classified unit pattern shape with a predetermined reference value to determine whether it is normal information or defective information, and registering the normal information in a normal dictionary; repeating each of the above steps for other areas under the condition that the numerical information is stored in the storage means as defect information except for the normal information registered in the normal dictionary; and the above steps are repeated for all areas of the inspection pattern a step of leaving the information left behind as defect information in the storage means after the repetition; a unit with a high occurrence frequency among the unit pattern shapes determined to be the defect information left in the storage means; A display step of displaying pattern shapes individually on a display means, selecting normal information or defect information for the unit pattern shape displayed on the display means, and displaying numerical information regarding the unit pattern shape determined to be defect information. registering in the defect dictionary; and performing the display and selection step on all unit pattern shapes that are determined to be defect information left in the storage means and that have a frequency of occurrence equal to or higher than a predetermined standard. A pattern inspection method comprising: a repeating step; and a step of inspecting an inspection pattern using the defect dictionary.