JP2003317083A - Image classifying method, program, and image classifying device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image classifying method which can easily generate a preferable decision function. <P>SOLUTION: The image classifying method of classifying defects of a pattern formed on a substrate when the pattern is inspected includes a process of preparing tutor data (step S21), a process of generating a decision function (function tree) through GA processing (step S24), a process of evaluating the generated decision function (step S25), and a process of repeating the generation and evaluation of the decision function (step S28). Consequently, the preferable decision function is easily generated. Then the finally generated decision function is used to classify defects. <P>COPYRIGHT: (C)2004,JPO

Description

Detailed Description of the Invention

[0001]

TECHNICAL FIELD The present invention relates to a technique for automatically classifying images.

[0002]

2. Description of the Related Art In the field of inspecting a pattern formed on a semiconductor substrate, a reticle, a printed wiring board or the like, a comparative inspection method has been mainly used conventionally. For example, in the case of a binary image, the image to be inspected (hereinafter,
It is called "target image". ) And the reference image, the image of the exclusive OR is obtained, and if there is a lump of a certain size or more in the obtained image, the region is detected as a defect.
In the case of a multi-valued image, a difference image showing the absolute value of the difference between the target image and the reference image is obtained, and an area having a pixel value larger than a predetermined threshold value in the difference image is detected as a defect.

In defect detection, defects may be classified (that is, the type (category) of defects is determined) in order to correct the manufacturing process that causes a reduction in the yield in each process. . In general, the classification process calculates a feature amount (area, brightness, etc.) from the target image used in the inspection, the reference image, and the vicinity of the defect of the image generated in the inspection, and calculates the feature amount as the feature amount. Based on this, a discriminant analysis or a method such as a neural network model is used.

[0004]

By the way, in the discriminant analysis method, the defect classification result is obtained by substituting the feature amount into a predetermined discriminant function. If the relationship between the feature quantity and the classification result is very simple, it is possible to perform correct classification using the discriminant function with fixed parameters, but when dealing with many feature quantities and many classification results. In the case of performing complicated classification in which the relationship between the feature amount and the classification result is complicated, it is difficult to perform appropriate classification by the discriminant analysis method.

When classification is performed using a neural net model, higher classification than the discriminant analysis method is expected, but the neural net model has the constraint that the input and output have the same relationship in each neuron. Yes,
Further, there is a problem that it is difficult to predetermine the optimal values of the number of steps in the intermediate layer and the number of neurons in the intermediate layer. That is, it can be said that there is a very high possibility that the advanced classification cannot be easily realized by the neural network model.

The present invention has been made in view of the above problems, and an object of the present invention is to provide a classification method excellent in image classification performance.

[0007]

An invention according to claim 1 is an image classification method for classifying images, which includes a step of preparing a teacher feature amount group of teacher images, an operator group and a variable group. Generating a discriminant function that is a combination of functions selected from a group of function elements, obtaining an evaluation value by substituting the teacher feature amount group into a variable group of the discriminant function, until a predetermined criterion is satisfied. The step of generating the discriminant function and the step of acquiring the evaluation value are repeated, the step of obtaining a target feature amount group from the target image, and the variable group of the discriminant function generated by the repeating step is added to the target feature amount group. To obtain the classification result of the target image.

The invention described in claim 2 is the image classification method according to claim 1, wherein the function element group includes a constant group.

The invention according to claim 3 is the image classification method according to claim 1 or 2, wherein the teacher image and the target image are images of a pattern formed on a substrate.

The invention according to claim 4 is the image classification method according to claim 3, wherein the classification result indicates the type of the defect of the pattern formed on the substrate.

The invention according to claim 5 is the image classification method according to any one of claims 1 to 4, wherein the step of generating the discriminant function generates a discriminant function by a genetic algorithm.

The invention according to claim 6 is the image classification method according to any one of claims 1 to 5, wherein each of the teacher image and the target image is an image to be inspected, and the teacher feature Each of the quantity group and the target feature quantity group further includes a feature quantity of a reference image that is referred to when an inspection of an inspection target is performed.

The invention according to claim 7 is the image classification method according to any one of claims 1 to 6, wherein each of the teacher image and the target image is an image to be inspected, and the teacher feature Each of the quantity group and the target feature quantity group further includes the feature quantity of the image generated during the inspection of the inspection target.

According to an eighth aspect of the present invention, there is provided a program for causing a computer to generate a discriminant function for image classification, and the execution of the program by the computer prepares a group of teacher feature amounts of teacher images in the computer. A step of generating a discriminant function which is a combination of a step and a function element group including an operator group and a variable group; and an evaluation value obtained by substituting the teacher feature amount group into the variable group of the discriminant function. The step of acquiring and the step of generating the discriminant function and the step of acquiring the evaluation value are repeated until a predetermined criterion is satisfied.

According to a ninth aspect of the present invention, there is provided an image classification device for classifying images, wherein a discriminant function determining section for determining a discriminant function for image classification, and a characteristic amount for acquiring a target characteristic amount group of a target image. An acquisition unit and a classification unit that acquires a classification result of the target image by substituting the target feature amount group into a discriminant function variable group, and the discriminant function determination unit determines the teacher feature amount group of the teacher image. Evaluation by assigning the teacher feature amount group to the variable group of the discriminant function, a step of preparing, a step of generating a discriminant function which is a combination of function element groups including an operator group and a variable group, and A step of acquiring a value, a step of generating the discriminant function and a step of acquiring the evaluation value are repeated until a predetermined criterion is satisfied.

[0016]

1 is a diagram showing a configuration of an inspection system 1 according to a first embodiment of the present invention. The inspection system 1 includes an imaging device 2 that acquires an image of a pattern formed on a semiconductor substrate (hereinafter referred to as “substrate”) 9, an inspection unit 3 that inspects a pattern for defects based on the acquired image, and , Having a computer 4 connected to the inspection unit 3. The computer 4 classifies the defects detected by the inspection unit 3 (that is, classifies the defect images) and also determines a discriminant function necessary for the defect classification as preparation for the classification process.

The image pickup device 2 has a camera 21 having a line sensor, a two-dimensional image pickup device, etc., and a stage 22 for moving the substrate 9 in the horizontal direction, and can pick up an image of a desired area on the substrate 9. It is said that. Imaging device 2
Can also be incorporated in the manufacturing line of the substrate 9, and the inspection system 1 can be used as a so-called in-line system.

The inspection unit 3 has a reference image memory 31 for storing reference image data in advance, and a comparison inspection circuit 32 for performing inspection by comparing the reference image with a target image which is an image to be inspected. The input from the image pickup device 2 can be switched between the reference image memory 31 and the comparison inspection circuit 32 by the changeover switch 33, and the changeover switch 33 connects the image pickup device 2 and the reference image memory 31 in advance. Data of the reference image is stored in the reference image memory 31 by the camera 21 capturing an image of the substrate 9 serving as a reference.

When the inspection is performed, the image pickup device 2 and the comparison inspection circuit 32 are connected by the changeover switch 33, and the data of the target image is input from the image pickup device 2 to the comparison inspection circuit 32 and the reference image memory 31. The reference image data is synchronously input from. In the comparison inspection circuit 32, for example, a difference image having the absolute value of the difference in pixel units between the reference image and the target image as a pixel value is obtained, and binarization is performed with a predetermined threshold to obtain a large difference area. Is detected as a defect. In the following description, the binarized difference image is referred to as an “inspection result image”. The inspection result image is, for example, 2 in which the pixel value of the defective area is 1 and the pixel value of other areas is 0.
Value image.

The inspection unit 3 includes two timing adjustment circuits 34a and 34b and two AND circuits 35a and 35b.
Further, three cutout circuits 36a, 36b, 36c are further provided, and the data of the target image from the timing adjustment circuit 34a and the data of the inspection result image from the comparison inspection circuit 32 are input to one AND circuit 35a.
An image in which the target image is masked with the inspection result image (that is,
An image in which the pixel value of the target image is given only to the defective area) is acquired. The masked image is sent to the cutout circuit 36a, and the vicinity of the defective area is cut out into a rectangle.

Data of the reference image from the timing adjusting circuit 34b and data of the inspection result image from the comparison inspection circuit 32 are input to the other AND circuit 35b, and the image obtained by masking the reference image with the inspection result image (that is, a defect). An image in which the pixel value of the reference image is given only to the region) is acquired. The masked image is sent to the cutout circuit 36c, and the vicinity of the defective area is cut out into a rectangle. Data of the inspection result image is input to the cutout circuit 36b, and the vicinity of the defective area is cut out into a rectangle.

The image data cut out by the three cutting circuits 36a to 36c is stored in the image memory 37.
In the following description, the target image and the reference image after the masking and the clipping are simply referred to as the “target image” and the “reference image”, respectively, and the inspection result image after the clipping is also simply referred to as the “inspection result image”. Call.

As shown in FIG. 2, the computer 4 has a general computer system configuration in which a CPU 41 for performing various arithmetic processes, a ROM 42 for storing basic programs, and a RAM 43 for storing various information are connected to a bus line. There is. The bus line further includes a fixed disk 44 for storing information, a display 45 for displaying various kinds of information such as images, a keyboard 46a and a mouse 46b for accepting input from an operator (hereinafter, referred to as "input unit 46").
Collectively. ), A reading device 47 that reads information from a computer-readable recording medium 8 such as an optical disc, a magnetic disc, and a magneto-optical disc, and a communication unit 48 that transmits and receives signals to and from other components of the inspection system 1.
, Are appropriately connected via an interface (I / F) or the like.

The computer 4 has a reading device 47 in advance.
The program 80 is read from the recording medium 8 via
It is stored on the fixed disk 44. And program 8
0 is copied to RAM 43 and CPU 41 reads R
By executing the arithmetic processing according to the program in the AM 43 (that is, the computer executing the program), the computer 4 determines the classification processing and the discriminant function used for the classification processing.

In FIG. 3, when the CPU 41 operates according to the program 80, the CPU 41, the ROM 42, the RA
It is a block diagram which shows the functional structure which M43, the fixed disk 44, etc. implement | achieve. In FIG. 3, the feature amount calculation unit 51 and the classification unit 52 show the functions when the computer 4 operates as a classification device, and the feature amount calculation unit 53 and the function determination unit 54 make the computer 4 prepare for the classification process. The function at the time of determining a discriminant function is shown. In addition,
These functions may be constructed by a dedicated electric circuit, or a dedicated electric circuit may be partially used.

FIG. 4 is a diagram showing a flow of classification processing by the computer 4. When the classification process is performed, first,
A target image obtained by imaging the inspection target from the image memory 37 of the inspection unit 3, a reference image referred to during the inspection,
Also, the data of the inspection result image generated during the inspection (target image data 601 and reference image data 60 in FIG. 3).
2 and inspection result image data 603. ) Is input to the feature amount calculation unit 51. Then, the characteristic amount of each image is obtained by performing an operation on each image data (step S11). Note that the feature amount is the average brightness,
The area and entropy of a region having a pixel value are values obtained by subjecting the pixel value in the image to a predetermined calculation process. For example, a value obtained by a calculation process in which a filter acts on an image may be used as the feature amount.

The obtained feature amount is input to the classification unit 52, and the classification unit 52 performs the classification based on the feature amount. A discriminant function in which the feature amount is substituted into a variable is prepared in the classification unit 52, a category is selected from the category data 604 according to the calculation result of the discriminant function, and the selected category (that is, defect type). ) Is output as the classification result (step S12).

FIG. 5 is a diagram showing a flow of processing for determining a discriminant function used in the classification processing in the computer 4.

In determining the discriminant function, first, teacher data is prepared on the fixed disk 44 (step S21). The teacher data includes a teacher image showing a pattern formed on the substrate 9, a reference image corresponding to the teacher image (an image referred to during inspection when the teacher image is regarded as a target image), and a teacher image. Of the inspection result image (the image generated at the time of inspection when the teacher image is regarded as the target image) (the teacher image data 611 and the reference image data 6 in FIG. 3)
12 and the inspection result image data 613), and the teaching signal 61 indicating the category corresponding to the teacher image.
4 is included.

Teacher image data 611 and reference image data 6
12 and the inspection result image data 613 are the inspection unit 3
The one used in may be diverted for teachers,
It may be separately prepared by the operator via the input unit 46. The teaching signal 614 is input by the operator via the input unit 46. It should be noted that the various image data are image data in which the vicinity of the defect area is cut out, like the image data in which the classification process is actually performed.

When the discriminant function is determined, in FIG. 3, the feature amount calculation unit 53 causes the teacher image data 611 for the teacher,
Reference image data 612 and inspection result image data 613
Then, the function determining unit 54 determines the discriminant function using the feature amount and the teaching signal 614. The function determining unit 54 generates a discriminant function candidate by using a genetic algorithm, and then repeatedly evaluates the discriminant function candidate and regenerates the discriminant function to determine a final discriminant function. It In FIG. 5, the process after step S22 is a loop process using a genetic algorithm.

In the first loop processing after the teacher data is prepared, the GA discriminator 541 of the function determiner 54 generates and registers the discriminant function of initialization (step S).
22, S23).

The discriminant function is at least 1 in which at least one operator and the feature amount obtained for teachers are substituted.
It is a combination of two variables. Further, at least one constant is combined as needed. In other words, the discriminant function has a structure in which ones selected from the function element group including the operator group, the variable group, and the constant group are combined.

Note that when determining the discriminant function shown in FIG. 5, the variable to which the feature amount of the teacher image, the feature amount of the reference image for teacher, and the feature amount of the inspection result image for teacher are input is shown in FIG.
In the classification process shown in (1), the feature amount of the target image, the feature amount of the reference image for inspection, and the feature amount of the inspection result image at the time of inspection are substituted.

The operator is selected from a group of operators such as four arithmetic operators (addition, subtraction, multiplication, division), tanh, exp, and sin. The four arithmetic operators are combined with two variables (or a combination of operators and variables (hereinafter, referred to as “variable etc.”)), and tanh, ex
p and sin are combined with one variable or the like. The operators are not limited to the above example, and the more kinds of operators there are, the higher the degree of freedom in selecting a discriminant function is. Further, the operator may be a predetermined function (so-called macro) defined by the operator.

The feature amount substituted into the variable of the discriminant function is, for example, a defect region (however, not necessarily a defect region) obtained from the inspection result image for teacher (the inspection result image of the target image in the classification processing shown in FIG. 4). It does not need to be present, to be precise, it is a region where a defect may occur.), The average brightness of the teacher image (target image in the classification process), the reference image for the teacher (reference image at the time of inspection in the classification process) Brightness average is used. These feature amounts may be normalized (for example, minimum value 0 and maximum value 1) in advance. In the following description, the variable into which the area of the defect area is substituted is “Area”, the variable into which the average brightness of the teacher image (or the target image) is substituted into “ObjAveL”, and the variable into which the average brightness of the reference image is substituted into. It is called "RefAveL". Although not used in the following specific example, a more optimal discriminant function can be determined by using a feature amount (for example, the average brightness of difference images) from the inspection result image.

As the constant of the discriminant function, for example, (-1
00) to (+100) selected from a constant group of 100 kinds of random numbers are used. In the following description, 100 constants are used as Const00 to Con
Call st99. The value of the constant is, for example, Const00.
= 0.5, Const01 = (-3.2) and the like.

Since the discriminant function is a combination of variables and constants with operators, it can be expressed in a tree format with variables and constants as terminals and operators as branch points. Hereinafter, the discriminant function handled in the tree format will be referred to as “function tree”. For example, the discriminant function represented by Equation 1 which outputs z is handled as the function tree shown in FIG.

[0039]

[Equation 1]

In step S23, a variable group, an operator group, and a constant group are randomly selected and combined as an initial setting, and a plurality of function trees are generated and registered. The larger the number of function trees, the better.

Next, the function evaluation section 542 of the function determination section 54.
Thus, the plurality of function trees are individually evaluated and ranked (step S25). The evaluation of the function tree is performed by substituting the feature amount obtained from the data of various images in the teacher data into the variable of the function tree and comparing the output (calculation result) with the teaching signal. Hereinafter, two types of defects (defect α,
It is assumed that the discriminant function for classifying the defect β) is finally determined. If the discriminant function output is positive, it is determined to be the defect α, and if it is negative, it is determined to be the defect β. The teaching signals shown in Table 1 are prepared.
A specific example will be described on the assumption that the three function trees 71 to 73 shown in FIGS. 7 to 9 are registered as the initial setting function trees. The column of "ID" in Table 1 is the identification number of the defect sample from which the teacher data was generated.

[0042]

[Table 1]

Table 2 is a table showing a list of values obtained when the function tree is evaluated. In the “tree number” column of Table 2, “1”, “2”, and “3” are the function trees 71 and 7, respectively.
It corresponds to 2,73. As shown in Table 2, when the feature value for teacher of each ID is substituted into the variable of each function tree to obtain the output value, the square of the difference between the output value and the teaching signal (“square error in Table 2” ).) Is required. Then, in each function tree, an average value of a plurality of squared errors corresponding to a plurality of IDs is obtained, the reciprocal of (average value + 1) is set as an evaluation value, and ranking is performed in order from the largest evaluation value ( In Table 2, the higher the rank, the greater the number.)

[0044]

[Table 2]

Next, the highest evaluation value among the obtained evaluation values is stored, and the corresponding function tree is also stored as a finally used function tree candidate (steps S26 and S27).
In the example of Table 2, the function tree 73 with tree number “3” is stored.

When the above process is completed, the process returns to step S22 (step S28), and in the second and subsequent loop processes, the GA processing unit 541 performs the process using the genetic algorithm (GA process) (step S24). Genetic algorithms generally use "crossover" or "mutation".
And "copy" processing is performed on the function tree. The GA process is performed with higher frequency for a function tree with a higher rank to generate a next-generation function tree. FIG. 10 is a diagram showing the flow of GA processing.

In the GA process, first, which of the "crossover", "mutation" and "copy" is to be performed is selected (step S241). GA is the choice of these processes
It is performed by the processing unit 541 at random.

When "Cross" is selected (step S2)
42), two function trees having a relatively high rank are selected from the plurality of registered function trees (step S24).
3), a subtree of each function tree is randomly selected. Then, by replacing the selected subtrees with each other, two new function trees are generated and registered as next-generation function trees (step S244). For example, as shown in FIG. 11, a subtree indicated by a thick frame is selected in the function trees 72 and 73 of the current generation in the upper stage, and the subtrees are exchanged to generate a next-generation function tree in the lower stage.

If "crossover" is not selected, one function tree having a relatively high rank is selected (step S2).
45). If the selected process is “mutation” (step S246), a subtree of the function tree is randomly selected, and the selected subtree is replaced with a newly randomly generated subtree. The new function tree is registered as the next-generation function tree (step S247). For example, as shown in FIG. 12, the subtree indicated by a thick frame is selected in the function tree 73 of the current generation in the upper stage, and the subtree is replaced with a new subtree to generate a next-generation function tree in the lower stage. To be done.

When "Duplicate" is selected, as illustrated in FIG. 13, the function tree selected in step S245 is registered as it is as a next-generation function tree without being changed (step S248). .

When the above-described processing is repeated and a predetermined number of next-generation function trees are registered, the GA processing is once ended (step S249), and the processing returns to the processing of FIG. Then, the plurality of function trees registered by the method described above are evaluated and ranked (step S25). When the highest evaluation value of each obtained function tree is higher than the highest evaluation value of the past (larger value), the function tree corresponding to the highest evaluation value of the past is the optimal function tree. Is registered as a candidate for (steps S26 and S27), and the process returns to the GA process again (step S28). On the other hand, when the obtained evaluation value does not exceed the highest evaluation value in the past, the processing directly returns to the GA processing (step S28).

When the GA process and the evaluation of the function tree are repeated a predetermined number of times (for example, 50 times) (may be repeated until another condition such as convergence of the evaluation value is satisfied), the highest in these processes. The function tree corresponding to the evaluation value of is determined as the optimum function tree.

FIG. 14 is a diagram showing an example of a final function tree (that is, a discriminant function) obtained by the above processing.
When the most preferable function tree is obtained, the output value of the function tree for the teacher data, the squared error, and the evaluation value are as shown in Table 3.

[0054]

[Table 3]

The final discriminant function is sent from the function determining section 54 shown in FIG. 3 to the classifying section 52 and registered in the classifying section 52 (step S29). As a result, the classification process shown in FIG. 4 is made executable by the computer 4.
Then, the feature amount of each image related to the inspection target obtained by the feature amount calculation unit 51 is input to the classification unit 52, and the feature amount is substituted into the variable group of the discriminant function to obtain the classification result.

That is, the target image, the reference image for inspection, and the reference image for inspection are set in the variables into which the feature amounts of the teacher image, the reference image for teacher, and the inspection result image for teacher are substituted when the discriminant function is generated. When the feature amount of the inspection result image is substituted and, for example, an output value close to 1 is obtained as shown in Table 4, it is determined that the defect α exists in the inspection area of the substrate 9 by referring to the category data 604. , (-1), it is determined that the defect β exists in the inspection area.

[0057]

[Table 4]

As described above, the computer 4 can easily generate a preferable discriminant function for classifying images using a genetic algorithm with a high degree of freedom,
High-performance classification processing can be easily realized.

Next, a second operation example of the inspection system 1 will be described.
Will be described as an embodiment. In the second embodiment,
Three discriminant functions A, B, for classifying three types of defects
C is determined and input to the classification unit 52. Where Table 5
As shown in, the relationship between each category of defects α, β, γ and the teaching signal for obtaining each discriminant function A, B, C is determined, and teaching data (various images and teaching signals for obtaining the feature amount Combination) is prepared. In addition, by preparing the teacher data illustrated in Table 5, after the discriminant function is determined, if the output value of the discriminant function A is the maximum, the defect α, and if the output value of the discriminant function B is the maximum, the defect β, the discrimination is performed. If the output value of the function C is the maximum, it will be judged as a defect γ.

[0060]

[Table 5]

The function determining process of the inspection system 1 according to the second embodiment is basically the same as that shown in FIG. 5, but Table 6 is shown for each discriminant function A to C after the teacher data is prepared. As shown, three function trees A1 to A3, B1 to B3, C1
~ C3 are initially generated by a random combination of variables and operators (steps S21 to S23).

[0062]

[Table 6]

In the evaluation of the function tree (step S25), as shown in Table 7, the feature amount is substituted into the function tree group corresponding to each discriminant function to obtain the evaluation value, and the ranking is performed in each function tree group. Be seen. After that, in each function tree group, the above-mentioned G
By the processing A, a new function tree is generated as needed (step S24). That is, the function tree is generated and evaluated in the function tree group corresponding to each discriminant function, and finally one discriminant function (a total of three discriminant functions A to C) is determined from each function tree group.

[0064]

[Table 7]

By registering the three discriminant functions A to C finally determined as described above in the classifying section 52, the classification process shown in FIG. 4 becomes possible. That is, the feature amount is obtained from the data of the target image regarding the inspection target, the reference image for inspection, and the inspection result image at the time of inspection, and the discriminant function A
By substituting for ~ C, the output illustrated in Table 8 is obtained. If the output value of the discriminant function A is the maximum, the defect α is determined. If the output value of the discriminant function B is the maximum, the defect β is determined. If the output value of the discriminant function C is the maximum, the defect γ is determined.

[0066]

[Table 8]

When classifying a plurality of types of defects, it is possible to classify according to the range of the output value of the discriminant function. For example, when the output of the discriminant function is near 1, the defect is α, when it is close to 0, the defect is β, and when it is close to −1, the defect is γ. Of course, it is preferable to prepare a large number of discriminant functions so that a large number of types of defects can be accurately classified, and discriminant functions corresponding to the number of defect types (categories) are prepared.

Although the embodiments of the present invention have been described above, the present invention is not limited to the above embodiments, and various modifications can be made.

The inspection result image is not limited to the difference image,
It may be another type of image or a plurality of types of inspection result images. Further, as described above, the feature extraction may be performed from a plurality of inspection result images (for example, difference images, edge images, etc.).

Although the defect inspection is performed in the above embodiment, the category indicated by the classification result may include "normal (non-defect)". Further, the inspection target is not limited to the semiconductor substrate, and can be applied to a classification process in the visual inspection of various patterns such as a printed circuit board, a lead frame, a flat panel display and a glass substrate for a photomask. Note that by using the discriminant function, an image of an object having an abstract pattern (in particular, an object having a geometric pattern formed on a substrate) (that is,
Classification processing suitable for non-natural images) can be performed.

In the above-described embodiment, it has been described that the feature amount is calculated from the target image (teacher image), the reference image and the inspection result image, but the feature amount is calculated only from the target image (teacher image). Good. Furthermore, the above-described processing for determining the discriminant function can be applied to a classification device that classifies an object based on an image without performing a defect inspection. In this case, the image cutout process may be omitted.

In the above embodiment, the discriminant function is determined by using the genetic algorithm, but the discriminant function may be determined by another method. For example, a local search method that obtains an optimal solution by neighborhood search, a random multi-start local search method, an annealing method or a taboo search method that is an improvement of the local search method, or the like may be used.

The characteristic amount calculation unit 51 and the characteristic amount calculation unit 5 in FIG.
3 may be provided as the same processing routine, and the teacher data may be prepared not as the image data format but as a combination of the feature amount and the teaching signal.

In the inspection system 1, the inspection is performed by an electrical dedicated circuit, and the computer 4 performs the classification process and the determination of the discriminant function by software. Which function is provided by the dedicated circuit is determined by software. Which functions are built or which functions are executed by one independent computer may be appropriately changed. For example, the classification process may be provided as a dedicated circuit, and the inspection process, the classification process, and the determination of the discriminant function may be executed by software by individual computers.

[0075]

According to the present invention, it is possible to easily generate a preferable discriminant function and easily realize a high-performance classification process.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of an inspection system.

FIG. 2 is a diagram showing a configuration of a computer.

FIG. 3 is a block diagram showing a configuration of functions realized by a computer.

FIG. 4 is a diagram showing a flow of classification processing.

FIG. 5 is a diagram showing a flow of processing for determining a discriminant function.

FIG. 6 is a diagram illustrating a function tree.

FIG. 7 is a diagram illustrating a function tree that is initially registered.

FIG. 8 is a diagram illustrating a function tree that is initially registered.

FIG. 9 is a diagram illustrating a function tree that is initially registered.

FIG. 10 is a diagram showing a flow of GA processing.

FIG. 11 is a diagram for explaining a “crossover” process.

FIG. 12 is a diagram for explaining processing of “mutation”.

FIG. 13 is a diagram for explaining processing of “copy”.

FIG. 14 is a diagram illustrating a finally generated function tree.

[Explanation of symbols]

4 computer 9 board 51 feature quantity calculation unit 52 classification unit 54 function determination unit 80 program 601 target image data 602, 612 reference image data 603, 613 inspection result image data 611 teacher image data S11, S12, S21, S24, S25, S28 Step

Front page continuation (51) Int.Cl. ⁷ Identification code FI theme code (reference) G06T 7/00 350 G01B 11/24 FF term (reference) 2F065 AA49 AA56 AA61 BB02 CC19 FF04 JJ02 JJ03 JJ25 JJ26 PP12 QQ04 QQ13 QQ25 RR08 5B057 AA03 BA02 CE06 CE12 DA03 DA12 DB02 DB09 DC04 DC36 DC40 5B075 ND06 NK06 NK39 NR12 5L096 AA06 BA03 CA02 EA35 EA43 FA59 GA08 GA17 GA51 GA55 HA13 JA11 JA22 KA04

Claims (9)

[Claims] 1. An image classification method for classifying images, the method comprising: preparing a teacher feature amount group of a teacher image; and a discriminant function which is a combination of a function element group including an operator group and a variable group. Generating an evaluation value by substituting the teacher feature amount group into a variable group of the discriminant function, generating the discriminant function until a predetermined criterion is satisfied, and acquiring the evaluation value And a step of obtaining a target feature amount group from the target image, and a step of obtaining the target image classification result by substituting the target feature amount group into the variable group of the discriminant function generated by the repeating step. An image classification method, comprising: 2. The image classification method according to claim 1, wherein the function element group includes a constant group. 3. The image classification method according to claim 1, wherein the teacher image and the target image are images of a pattern formed on a substrate. 4. The image classification method according to claim 3, wherein the classification result indicates a type of a defect of a pattern formed on a substrate. 5. The image classification method according to claim 1, wherein the step of generating the discriminant function generates a discriminant function by a genetic algorithm. 6. The image classification method according to claim 1, wherein each of the teacher image and the target image is an image to be inspected, and the teacher feature amount group and the target feature amount. An image classification method, wherein each of the groups further includes a feature amount of a reference image that is referred to when an inspection of an inspection target is performed. 7. The image classification method according to claim 1, wherein each of the teacher image and the target image is an image to be inspected, and the teacher feature amount group and the target feature amount. An image classification method, wherein each of the groups further includes a feature amount of an image generated at the time of inspection of an inspection target. 8. A program for causing a computer to generate a discriminant function for image classification, wherein the execution of the program by the computer comprises the steps of: preparing the computer with a teacher feature amount group of teacher images; A step of generating a discriminant function which is a combination of functions selected from a function element group including a variable group, a step of obtaining an evaluation value by substituting the teacher feature amount group into the variable group of the discriminant function, A program for executing a step of generating the discriminant function and a step of repeating the step of obtaining the evaluation value until a criterion is satisfied. 9. An image classification device for classifying images, comprising: a discriminant function determining unit for determining a discriminant function for image classification; a feature amount acquiring unit for obtaining a target feature amount group of a target image; A classification unit that acquires the classification result of the target image by substituting the target feature amount group into a variable group; and the discriminant function determination unit prepares a teacher feature amount group of the teacher image; A step of generating a discriminant function which is a combination of function element groups including a child group and a variable group, and a step of obtaining an evaluation value by substituting the teacher feature amount group into a variable group of the discriminant function, An image classification apparatus, wherein: the step of generating the discriminant function and the step of obtaining the evaluation value are repeated until a predetermined criterion is satisfied.