JP2010091523A - Method and device for discriminating light and shade pattern, program for executing the method for discriminating light and shade pattern with computer, and computer-readable recording medium recording the program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method and device for accurately discriminating similarities, even if light and shade difference in a pattern is large or small, and unknown light and shade abnormality or failures are cotained in the pattern, by allowing brightness change to be robust in the method for discriminating light and shade patterns for discriminating whether a discrimination object pattern is similar to a comparison object pattern. <P>SOLUTION: Regular processing is applied on the light and shade values of the discrimination object patterns and the comparison object patterns to generate first and second regular light and shade values. The first and second regular light and shade values are made variables to set a two-dimensional space, making two mutually orthogonal coordinate axes. Points expressed by the first and second regular light and shade values at mutually corresponding positions on the discrimination object patterns and the comparison object patterns in the two-dimensional space are accumulated to make two-dimensional histograms. Gathering of the points making the two-dimensional histograms is classified to at least a single cluster. The method discriminates whether the discrimination object patterns are similar to the comparison object patterns, on the basis of the shapes, positions or frequencies relating to the cluster. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a light / dark pattern discriminating method, and more specifically, a light / dark pattern as a discrimination target (referred to as a “discrimination target pattern”) is used as a light / dark pattern as a comparison target (reference) (referred to as a “comparison target pattern”). It is related with the shading pattern discrimination method which discriminate | determines whether it is similar.

  The present invention also relates to a shading pattern discriminating apparatus that executes such a shading pattern discrimination method.

  The present invention also relates to a program for causing a computer to execute such a shading pattern discrimination method.

  The present invention also relates to a computer-readable recording medium recording such a program.

  In recent years, film thickness distribution of thin films formed by liquid phase growth method, vapor phase growth method, etc. is observed with a camera, and film formation is performed by automatically detecting whether the film thickness distribution is normal or abnormal. Systems that detect process abnormalities early have been developed. In general, when a thin film is measured by a camera using an interference optical system, the film formation state (film thickness distribution state) appears as a shading pattern on an image. For example, if there is a difference between the film thickness of a certain area and the film thickness of the surrounding area, a difference in density occurs between these areas on the image. In addition, if the film thickness distribution changes due to unexpected disturbance of the film formation conditions, the difference between the thin film light and dark patterns on the image (the surface of the light and dark values) Difference in internal distribution).

  In a system that automatically monitors a film formation state using an image captured by a camera, monitoring is performed using the above-described difference. As an automatic monitoring process, first, a density pattern of an image relating to a film thickness distribution of “good” is previously determined as a reference density pattern. In addition, the degree to which the shade pattern of the image relating to the film thickness distribution to be monitored varies with respect to the reference shade pattern is quantified and managed as a difference between the shade patterns in the image. Furthermore, based on the quantitative value, it is determined whether the film thickness distribution to be monitored is normal (the same level as the reference film formation state) or abnormal, and is determined to be abnormal. If an alarm occurs, an alarm is issued to the person in charge of monitoring.

  Conventionally, the problem of determining whether a gray pattern is similar or abnormal (the problem of determining whether the comparison target pattern and the determination target pattern are the same or the comparison target pattern is significantly different from the comparison target pattern) Various methods have been proposed regarding whether or not there is a problem (whether an abnormality has occurred). For example, as the simplest method, there is a method of detecting the presence / absence of an abnormality / defect on a pattern by taking a difference between both patterns and analyzing the difference image. However, such a method based on the difference deteriorates the accuracy with respect to the change in the brightness of the light and shade pattern.

  On the other hand, as another approach, there is a method of calculating a normalized correlation between the comparison target pattern and the determination target pattern, and performing pattern determination using the correlation value as the similarity between the patterns. This method has an advantage that the similarity between patterns can be calculated robustly even with respect to variations in brightness across the entire pattern, and the patterns can be discriminated stably. On the other hand, the normalized correlation itself is equivalent to calculating the correlation between noise components when the gradation difference in the pattern is small (in the case of a pattern close to flat), and the correlation value becomes unstable. There's a problem. A pattern comparison method that avoids this problem is disclosed in Patent Document 1 (Japanese Patent Application Laid-Open No. 2004-108779) (referred to as “Prior Art 1”).

Another approach is to create a two-dimensional histogram consisting of the gray value of the comparison target pattern and the gray value of the discrimination target pattern, and analyze the distribution on this two-dimensional histogram, so that the difference between the two patterns can be determined. There is also a way to determine. A two-dimensional histogram based on gray values is a gray value (x coordinate) at a certain position on one pattern in a two-dimensional space (xy space) in which the scale of the gray value is the x axis and the y axis, respectively. It is generated by plotting a point represented by a pair with a gray value (y coordinate) at the same position of the other pattern. Therefore, if both patterns are completely the same pattern, all plot points are plotted on the straight line y = x on the two-dimensional histogram. For example, in Patent Document 2 (Japanese Patent Laid-Open No. 2001-133418), in the comparison of two grayscale patterns, a two-dimensional histogram composed of these grayscale values is generated, and the region constituting the matrix distribution on the distribution is the two patterns. And a method of detecting a region deviating from the similar region as a difference between patterns (pattern anomaly) is disclosed (in the literature, it is treated as a defect detection problem). Called Technology 2 ”). Further, in Patent Document 3 (Japanese Patent Laid-Open No. 2003-337945), as a color image classification method, a distribution of RGB color space of one color image is projected onto a two-dimensional space to form a two-dimensional histogram, There is disclosed a method for classifying a color image by performing processing for obtaining a feature for each predetermined angle centered on the origin of the distribution, and using the feature amount for each angle as the feature amount of the color image (this is referred to as “ This is called “prior art 3”).
JP 2004-108779 A JP 2001-133418 A JP 2003-337945 A

  The following characteristics are required for a method of discriminating a gray pattern in a system for monitoring the film thickness distribution of a thin film. First, as with many other systems, it is required to be able to distinguish patterns robustly against changes in brightness across the entire grayscale pattern due to changes in lighting conditions and the influence of the characteristics of the target model (factory). This is a problem that cannot be ignored even in an environment where an imaging environment is prepared to some extent.

  In addition, the shading pattern itself representing the film formation state (film thickness distribution) does not have a clear edge, such as a semiconductor wafer, but spreads with a relatively gentle gradient. It is required to be a method capable of accurately discriminating a simple pattern. For example, consider a case in which patterns having a light and shade distribution as shown in FIG. 1A and FIG. 1 (a) and FIG. 1 (b) are divided into 9 × 3 × 3 as shown in FIGS. 2-1 (a) and 2-1 (b), respectively. -1 (c), nine divided regions (i) to (ix) as shown in FIG. 2-1 (d) are obtained. Then, as shown in FIG. 2-2, the nine divided regions (i) to (ix) in FIG. 2-1 (c) and the nine divided regions (i) to (ix) in FIG. 2-1 (d). ) And the divided regions at the same position (for example, “upper left” divided regions), respectively, and the similarity determination of the light and shade pattern (determination of presence / absence of abnormality) is performed for each divided region. In this case, in the case where there is a large shade difference in the divided areas (for example, upper left area (i), lower left area (iii), upper right area (vii), lower right area (ix) in FIG. 2-1 (c)). Even in the case where the density difference in the divided area is small (for example, the central area (v) in FIG. 2-1 (c)), it is necessary to make a highly accurate discrimination for each density pattern.

  In addition, when the product is continuously produced in the factory, the film formation state becomes, for example, the state of FIG. 3-1 (b) after one month has elapsed from the state of FIG. 3-1 (a) (initial state), It changes over time during a period of several months, such as the state shown in FIG. 3-2 (c) after one month and the state shown in FIG. 3-2 (d) after another month. To go. When it is desired to detect such a sudden change in the film formation state as abnormal or defective while permitting such a change in the film formation state over time, in order to eliminate the influence of the change over time, the change is made immediately in the pattern to be discriminated. It is necessary to determine whether the light / dark pattern is good or bad by using the light / dark pattern of the target that has passed through the film process as a comparison target pattern. For example, as shown in FIG. 3A, in the initial state, the pattern shown in FIG. Further, as shown in FIG. 3A, after two months have elapsed, the pattern of FIG. For this reason, the comparison target pattern is not fixed, but is sequentially updated. Therefore, the gradation pattern discrimination method in the system for monitoring the film thickness distribution of the thin film must be such that the discrimination accuracy is not impaired by such influence.

  Furthermore, when there is no detailed knowledge about the tendency of changes in film formation over time, such as when a factory or line is started up, or abnormalities or defects that occur (the characteristics and types of the light and shade patterns are Even in the case where it is not fully understood, it is necessary to be able to discriminate the pattern stably and accurately.

  In the prior art 1, in order to be able to stably calculate the normalized correlation even when there is a small difference in shading within the pattern area, the arbitrary pattern is made the same for each of the comparison target pattern and the discrimination target pattern. The method of calculating the normalized correlation is shown. However, when detecting an abnormality or defect of an unknown grayscale pattern, it is difficult to determine in advance what pattern should be embedded, and thus it is difficult to apply the prior art 1. Furthermore, another problem with normalized correlation is (1) When a change appears in a large range within the pattern area, a difference appears in the correlation value, but for a change in a small range within the pattern, Even if it is a strong contrast change, the problem that the difference does not appear in the correlation value and (2) the problem that the difference in the correlation value does not easily appear with respect to the change in the low-contrast light and shade pattern cannot be solved. It is hard to say that it is a method that can stably determine abnormalities and defects. For example, three different types of shade patterns A, B, and C of a square shown in FIGS. 4-1 (a) to (c) are set as a reference. The pattern A is a pattern in which the center is bright and gradually darkens as the distance from the center increases. Pattern B is a pattern in which the lower right corner is brighter and gradually becomes darker as it moves away from the lower right corner. Pattern C is a pattern in which each part has the same brightness. Then, as shown in FIGS. 4-2-1 and 4-2-2, shade patterns A-1, A-2,..., A-13 are formed by adding shading or the like to the pattern A or overlapping abnormal elements. Set. As shown in FIGS. 4-3-1 and 4-3-2, shade patterns B-1, B-2,..., B-13 are set by adding shading or overlapping abnormal elements to the pattern B. . Further, as shown in FIG. 4-4-1 and FIG. 4-4-2, shade patterns C-1, C-2,. Set. Furthermore, as shown in FIGS. 5-1-1, 5-1-2, and 5-1-3, the pattern A and the patterns A-1, A-2,. Sets A-1, A-2,..., A-13. As shown in FIG. 5-2-1, FIG. 5-2-2, and FIG. 5-2-3, the data set B is obtained by combining the pattern B and the patterns B-1, B-2,. -1, B-2, ..., B-13 are set. Further, as shown in FIG. 5-3-1, FIG. 5-3-2, and FIG. 5-3-3, the pattern C and the patterns C-1, C-2,. Sets C-1, C-2,..., C-13. For the sake of simplicity, the same reference numerals are used for the shading patterns shown in FIGS. 4-2-1 to 4-4-2 and the data sets shown in FIGS. 5-1-1 to 5-3-3. Yes. When considering the problem of pattern discrimination by each data set shown in FIG. 5-1-1 to FIG. 5-3-3, that is, each combination of shading patterns, the discrimination based on the above-described normalized correlation is a good discrimination. It can't be said. Specifically, as shown in the tables from FIGS. 5-1-1 to 5-3-3, when the data sets C-2 to C-6 to be determined as “similar” are determined by normalized correlation As a result, the result of discrimination becomes “different” and an error occurs. When the data sets A-8, A-12, and A-13 that are to be discriminated as “different” are discriminated by normalized correlation, the discrimination result is “similar” and an error occurs. In addition, when the data sets B-7, B-8, B-12, and B-13 that are to be determined as “different” are determined by normalized correlation, the determination result is “similar” and an error occurs. As described above, in the discrimination based on the normalized correlation, the similarity is reversed, and erroneous discrimination cannot be avoided. Among the two problems of the normalized correlation in this example, the former problem (1) corresponds to the results for the data sets A-12, A-13 and the data sets B-12, B-13. Corresponding to the latter problem (2) is the result concerning the data set A-8, the data sets B-7, and B-8.

  In the prior art 2, for example, among the data sets shown in FIGS. 5-1 to 5-3-3, steep in the pattern like the data sets A-13, B-13, and C-13 In the case where shading changes occur, when the respective two-dimensional histograms are generated, as shown in FIGS. 6-1, 6-2, and 6-3, the matrix distribution portion on the two-dimensional histograms and from the respective matrix distribution portions are obtained. It is possible to clearly distinguish the “displaced region” existing at a distant position (see FIGS. 6-1 (b) (c), FIGS. 6-2 (b) (c), and FIGS. 6-3 (b) (c) .) Therefore, a good discrimination result can be expected. However, in the case of data set C-9 or data set A-8, as shown in FIGS. 7-1 and 7-2, there is no clear “outlier region” on the two-dimensional histogram (FIG. 7). -1 (b) (c), see FIGS. 7-2 (b) and (c). For this reason, it is difficult to accurately discriminate.

  Prior art 3 is a pattern classification method based on feature vectors listing feature values obtained for each angle around the origin of the distribution on the two-dimensional histogram. In this classification method, a two-dimensional histogram is generated from one color image, and a feature vector is calculated for the distribution. Therefore, the classification of the two color images is made by comparing the respective feature vectors. When this classification method is applied to the gray pattern discrimination problem assumed by the present application, the following procedure is considered. First, a data set in which the shading pattern is determined as “similar”-“similar” is regarded as one set, and a feature vector obtained from the two-dimensional histogram is preliminarily set as a feature vector of this data set, that is, a “feature vector representing a similar”. It is decided. In the stage of actually discriminating the light and shade pattern, the feature vector obtained from the two-dimensional histogram of the discrimination target pattern and the comparison target pattern is calculated as the “feature vector related to the discrimination target pattern”, The similarity between feature vectors is compared between “feature vector” and “feature vector related to discrimination target pattern”, and whether the discrimination target pattern and the comparison target pattern are “similar” or not (“different” It is a procedure to determine whether there is. Here, for example, among the data sets from FIG. 5-1-1 to FIG. 5-3-3, data sets in which the shading pattern is discriminated as “same”-“same” are listed, and the respective two-dimensional histograms are listed. When generated, the result is as shown in FIGS. 8-1 to 8-6. For the “similar”-“similar” data sets illustrated here, low contrast shading as shown in FIGS. 9A, 9B, and 9C is included in the discrimination target pattern for each comparison target pattern. It is a data set. As can be seen from this, the distribution of each two-dimensional histogram is not uniform. On the other hand, when the feature vectors shown in the prior art 3 are used, the feature vectors for each of them are greatly different. That is, the “similar feature vectors” vary greatly depending on the type of grayscale pattern (a set of data sets in this example). As mentioned earlier, in the problem assumed in this application, it is necessary to update the comparison target pattern sequentially when it is possible to detect a change in film formation over time and to detect a sudden change as abnormal or defective. In this case, it is necessary to prepare in advance “feature vectors representing the same type” as many types as possible for the comparison target pattern. This is extremely difficult when knowledge about the detailed characteristics regarding the tendency of the change in the film formation state and the characteristics is not sufficiently obtained. Therefore, it can be said that the gradation pattern cannot be accurately discriminated even by using the prior art 3.

  Therefore, an object of the present invention is a light / dark pattern determination method for determining whether or not a determination target pattern is similar to a comparison target pattern, which is robust to a change in brightness and has a light / dark difference in the pattern. An object of the present invention is to provide an apparatus capable of accurately determining similarity even if the pattern is large or small, and whether the pattern includes an unknown density abnormality or defect.

  Another object of the present invention is to provide a shading pattern discriminating apparatus that executes such a shading pattern discrimination method.

  Another object of the present invention is to provide a program for causing a computer to execute such a shading pattern discrimination method.

  Moreover, this invention is providing the computer-readable recording medium which recorded such a program.

  In order to solve the above problems, the present inventor has considered as follows.

  In the case where the discrimination target pattern and the comparison target pattern are equal to each other, for example, the data sets A-1, B-1, shown in FIGS. In the case of C-1, the two-dimensional histogram generated from each data set is on a straight line y = x in the two-dimensional space as shown in FIGS. 10-1, 10-2, and 10-3. Distributed. Even if there is a uniform brightness change between the light and shade pattern as the discrimination target pattern and the light and shade pattern as the comparison target pattern, if a two-dimensional histogram is generated by the light and dark pattern obtained by normalizing them, as described above Distributed. In addition, even when an abnormality or a defect is included in the light and shade pattern as the discrimination target pattern, the occurrence position and shape (whether linear or circular) of the light and shade pattern are different. However, the abnormality or defect is any one of the following: y = x straight line in the xy two-dimensional space is “swelling”, “bending”, “changing direction”, “changing shape”, or It appears as a geometric change that combines them. In addition, the size of the change area in the light and shade pattern, contrast strength, and the like appear as the amount (degree) of such geometric change (however, as a special exceptional change, FIG. -1, FIG. 6-2, and the example of FIG. Therefore, by measuring how much the y = x line on the xy plane of the two-dimensional histogram space changes geometrically and how it is complexly determined, abnormalities and defects of unknown gray patterns can be determined. Even if this occurs, it can be expected that the similarity of the light and shade pattern can be accurately determined.

  Also, when there are few shading differences in the shading pattern, as illustrated in FIGS. 8-1 to 8-6, the distribution of the two-dimensional histogram changes drastically with only slight noise and low-contrast shading. To do. This property makes it difficult to apply the prior art. As described above, when this phenomenon is considered as a geometric change of a straight line of y = x, the straight line swells greatly and its direction changes variously. However, noise, shading, and the like are generally gradual (not abrupt changes in shade value) over the entire shade pattern. Accordingly, even if the straight line y = x swells as a whole, it can be considered that the entire distribution is close to an ellipse or a circle without being damaged. Therefore, it can be said that the feature value related to the shape of this distribution (ellipse / circle) is a feature value that is not easily affected by noise and shading in the comparison of patterns with small shading differences. That is, it can be expected that the similarity of the light and shade pattern can be determined stably even if the patterns having a small light and shade difference are compared by using the feature of this shape.

  In order to quantitatively measure the geometric change amount regarding the line y = x on the xy plane of such a two-dimensional histogram space, first, the distribution on the two-dimensional histogram is labeled, and clusters (concatenated) are connected. It can be quantified by calculating the shape feature of the cluster. For example, the phenomenon that the straight line “swells” can be quantified as the variance of the cluster with respect to the y = x linear axis, and the phenomenon such as the y = x straight line “bends” or “changes the shape” It can be quantified by features such as uniqueness. In addition, regarding the event that the “extra area” completely deviating from the line y = x, which is an exceptional case as described above, occurs, the number of clusters to be generated (whether two or more clusters have been generated). ) And the degree can be determined by the position and frequency of the cluster.

  The light / dark pattern discrimination method of the present invention was created based on the above consideration.

In order to solve the above-described problem, the light / dark pattern discrimination method of the present invention is:
A density pattern determination method for determining whether or not a determination target pattern indicating a gray value distribution spreading in a certain area is similar to a comparison target pattern indicating a gray value distribution extending in a certain area,
A normalization process is performed on the gray value of the discrimination target pattern to generate a first normalized gray value, and a normalization process is performed on the gray value of the comparison target pattern to generate a second normalized gray value. ,
A two-dimensional space in which the first normalized gray value and the second normalized gray value are variables and two coordinate axes are orthogonal to each other is set, and the discrimination target pattern and the comparison are set in the two-dimensional space. The points represented by the first normalized gray value and the second normalized gray value at positions corresponding to each other on the target pattern are accumulated, and the first normalized gray value on the discrimination target pattern and the above Creating a two-dimensional histogram representing the correlation with the second normalized gray value on the comparison target pattern;
Classifying the set of points forming the two-dimensional histogram into at least one cluster;
It is determined whether or not the determination target pattern is similar to the comparison target pattern based on the shape related to the cluster, the feature related to the position or frequency, or the combination of the features related to the shape, position or frequency.

  In the shading pattern discriminating method of the present invention, the shading value of the discrimination target pattern is normalized to generate a first normalized shading value, and the shading value of the comparison target pattern is subjected to normalization processing. A normalized gray value of 2 is generated. Therefore, the determination result described later as to whether or not the determination target pattern is similar to the comparison target pattern is robust against the brightness change. Further, in the grayscale pattern discrimination method of the present invention, the set of points forming the two-dimensional histogram is classified into at least one cluster, and the shape, position, or frequency characteristics relating to the cluster, or the shape, position, or frequency are related. Based on the combination of features, it is determined whether or not the determination target pattern is similar to the comparison target pattern. Even if an abnormality or a defect is included, it can be accurately determined whether or not the determination target pattern is similar to the comparison target pattern.

  In one embodiment, the normalization process is a process of normalizing the gray value distributions of the discrimination target pattern and the comparison target pattern to a gray value distribution having a magnitude of 1 and an average of 0, respectively. Features.

  In the light / dark pattern discrimination method according to the embodiment, the normalization process is in a state suitable for comparing the discrimination target pattern and the comparison target pattern. Therefore, it is robust against changes in brightness.

  In the grayscale pattern discrimination method according to an embodiment, a filtering process for removing a noise component is performed on the two-dimensional histogram before the process of classifying the set of points forming the two-dimensional histogram into the cluster. It is characterized by.

In the light / dark pattern discriminating method of this embodiment, noise components such as isolated points and missing points on the two-dimensional histogram are removed by the filtering process. Therefore, it can be determined with higher accuracy whether the determination target pattern is similar to the comparison target pattern.
it can.

  In the shading pattern discriminating method of one embodiment, the filtering process converts the frequency value to 0 when the frequency value on the two-dimensional histogram is less than or equal to a certain threshold value, while the frequency value is larger than the threshold value. A binarization process for converting the frequency value into one or more constants is included.

  In the shading pattern discrimination method of this embodiment, information on noise-like elements (unimportant elements) on the two-dimensional histogram can be deleted by the binarization process.

In the light and shade pattern discrimination method of one embodiment,
The process of classifying the set of points forming the two-dimensional histogram into the clusters extracts a main cluster that forms the main part of the set of points,
The process of determining whether or not the determination target pattern is similar to the comparison target pattern includes, as the characteristics serving as a reference for the determination, a spread with respect to the main axis of the main cluster, an unevenness degree of the main cluster, and / Or the direction of the main axis of the main cluster is used.

  In the light / dark pattern discriminating method according to this embodiment, as the features serving as the criteria for the discrimination, the spread of the main cluster with respect to the main axis, the unevenness of the main cluster, and / or the direction of the main axis of the main cluster are used. Therefore, based on such characteristics of the main cluster, it can be determined with higher accuracy whether or not the determination target pattern is similar to the comparison target pattern.

In the light and shade pattern discrimination method of one embodiment,
In the process of classifying the set of points forming the two-dimensional histogram into the clusters, in addition to the main cluster, a sub-cluster representing a part different from the main body part is extracted from the set of points,
The process of determining whether or not the determination target pattern is similar to the comparison target pattern includes, as the characteristics serving as a reference for the determination, the distance between the sub-cluster and the main cluster, the sub-cluster Centroid position, frequency sum of the points in the sub-cluster, and / or sub-cluster area.

  In the light / dark pattern discrimination method according to this embodiment, the features serving as the criteria for the discrimination include the distance between the sub-cluster and the main cluster, the position of the center of gravity of the sub-cluster, and the frequency of the points in the sub-cluster. Use the sum and / or the sub-cluster area. Therefore, based on such characteristics of the sub-cluster, it can be determined with higher accuracy whether or not the determination target pattern is similar to the comparison target pattern.

  In the shading pattern discrimination method according to an embodiment, the process of discriminating whether or not the discrimination target pattern is similar to the comparison target pattern is sequentially performed for each cluster, and is based on any one of the clusters. When the determination target pattern is determined not to be similar to the comparison target pattern, the determination process is immediately terminated.

  In the shading pattern discrimination method of this embodiment, discrimination processing for useless clusters can be omitted. Therefore, the time for the discrimination process can be shortened as a whole.

  It is desirable to include a labeling process for assigning identification numbers to the classified clusters. In this case, it is possible to easily execute processing for determining whether or not the determination target pattern is similar to the comparison target pattern sequentially for each cluster.

The gray pattern discriminating apparatus of the present invention is
A light / dark pattern determining apparatus for determining whether or not a determination target pattern indicating a gray value distribution spreading in a certain area is similar to a comparison target pattern indicating a gray value distribution extending in a certain area,
A data input unit for inputting gray value data of the discrimination target pattern and the comparison target pattern;
Normalization processing is performed on the gray value of the discrimination target pattern to generate a first normalized gray value, and normalization processing is performed on the gray value of the comparison target pattern to generate a second normalized gray value. A normalization processing unit;
A two-dimensional space in which the first normalized gray value and the second normalized gray value are variables and two coordinate axes are orthogonal to each other is set, and the discrimination target pattern and the comparison are set in the two-dimensional space. The points represented by the first normalized gray value and the second normalized gray value at positions corresponding to each other on the target pattern are accumulated, and the first normalized gray value on the discrimination target pattern and the above A two-dimensional histogram creation unit that creates a two-dimensional histogram representing a correlation with the second normalized gray value on the comparison target pattern;
The set of points constituting the two-dimensional histogram is classified into at least one cluster, and the discrimination target pattern is based on a feature related to the shape, position or frequency related to the cluster, or a combination of features related to the shape, position or frequency. A similarity determination unit that determines whether or not is similar to the comparison target pattern.

  In the light / dark pattern discriminating apparatus according to the present invention, the normalization processing unit generates a first normalized light / dark value by performing normalization processing on the light / dark value of the pattern to be discriminated, and sets the light and dark value of the comparison target pattern. A normalization process is performed to generate a second normalized gray value. Therefore, the determination result by the similarity determination unit described later is robust against the brightness change. In the light / dark pattern discriminating apparatus according to the present invention, the two-dimensional histogram creating unit uses the first normalized gray value and the second normalized gray value as variables and uses two coordinate axes orthogonal to each other. A dimension space is set, and the points represented by the first normalized gray value and the second normalized gray value at positions corresponding to each other on the discrimination target pattern and the comparison target pattern are accumulated in the two-dimensional space. Then, a two-dimensional histogram representing the correlation between the first normalized gray value on the discrimination target pattern and the second normalized gray value on the comparison target pattern is created. The similarity determination unit classifies the set of points forming the two-dimensional histogram into at least one cluster, and combines the shape, position, or frequency with respect to the cluster, or a combination of features with respect to the shape, position, or frequency. Based on this, it is determined whether or not the determination target pattern is similar to the comparison target pattern. Therefore, whether the pattern to be discriminated is similar to the pattern to be compared or not, regardless of whether the shading difference in the pattern is large or small, or whether the pattern contains unknown shading abnormalities or defects it can.

  In addition, it is preferable that the light / dark pattern discriminating apparatus includes a data input unit for inputting light / dark value data of the light discrimination value pattern and the comparison target pattern. In that case, the data input unit inputs gray value data of the discrimination target pattern and the comparison target pattern. Whether or not the discrimination target pattern is similar to the comparison target pattern is discriminated with respect to the gray value data.

  In addition, it is desirable that the light / dark pattern discriminating apparatus includes a display device or a data output unit that outputs a discrimination result by the similarity judgment unit. Thereby, the user who uses the said light / dark pattern determination apparatus can know easily the determination result as to whether or not the determination target pattern is similar to the comparison target pattern.

  The program of this invention is a program for making a computer perform the said light / dark pattern discrimination method.

  The recording medium of the present invention is a computer-readable recording medium on which the program is recorded.

  Hereinafter, the present invention will be described in detail with reference to the illustrated embodiments.

  FIG. 11 illustrates the flow of the shading pattern discrimination method of one embodiment. Here, it is assumed that the shading patterns handled by this method, that is, the comparison target pattern and the discrimination target pattern are all aligned with the same size. Note that the “position” in the comparison target pattern and the discrimination target pattern indicates a rectangular small region (pixel in this example) designated by the XY coordinates, as shown in FIGS.

  First, a normalization process is performed on the grayscale value of the discrimination target pattern to generate a first normalized grayscale value, and a normalization process is performed on the grayscale value of the comparison target pattern to generate a first normalized grayscale value. To do. Specifically, in this normalization process, the grayscale value distribution of the discrimination target pattern and the grayscale value distribution of the comparison target pattern are each normalized to a pattern having a grayscale value distribution of size 1 and average 0 (S001, S002). ). This is a process for robustly discriminating patterns even with uniform brightness changes.

  Subsequently, a two-dimensional histogram representing the correlation between the first normalized gray value on the discrimination target pattern and the second normalized gray value on the comparison target pattern is generated (S003). Hereinafter, for the sake of simplicity, the normalized gray value is simply referred to as a “light value”.

  Here, the method for generating the two-dimensional histogram will be specifically described with reference to FIG. As shown in FIG. 12B, the gray value at a certain position (i, j) of the comparison target pattern is Lb. As shown in FIG. 12A, the gray value of the position (i, j) corresponding to the position in the discrimination target pattern is La (here, since both the gray patterns are aligned with the same size, The position (i, j) of the light / dark value La on the XY coordinates coincides with the position (i, j) of the light / dark value Lb). Here, as shown in FIG. 12C, a two-dimensional space is set in which the gray value on the discrimination target pattern and the gray value on the comparison target pattern are variables, and the vertical axis and the horizontal axis are orthogonal to each other. Then, based on the gray values La and Lb of the position (i, j) on the XY coordinates in FIGS. 12A and 12B, the position (La, Lb) in the two-dimensional space of FIG. Add 1 to the frequency of That is, one point d1 (indicated by a circle) is added. In this way, the positions corresponding to each other on the discrimination target pattern and the comparison target pattern in the two-dimensional space for all the positions (i, j) on the XY coordinates in FIGS. 12 (a) and 12 (b). The points represented by the normalized gray value La and the normalized gray value Lb are accumulated. Thereby, a two-dimensional histogram relating to the discrimination target pattern and the comparison target pattern is generated. This two-dimensional histogram can be regarded as a two-dimensional image (the frequency is a pixel value) representing the correlation between the gray value of the comparison target pattern and the gray value of the discrimination target pattern.

  Next, as shown in FIG. 13, two-dimensional expansion / contraction filtering is performed on the two-dimensional histogram generated in this way in order to remove noise components (isolated points and missing points) on the distribution. Implement (S004). Thereby, the isolated points d2 and d3 shown in FIG. 13A are removed and the missing points v1 and v2 are filled. Further, the entire distribution may be smoothed by a two-dimensional smoothing filter or the like. Thereby, the discrimination process described later becomes more accurate.

  As a pre-process for the next process (labeling process), a binarization process may be performed on the two-dimensional histogram. In this case, the expansion / contraction process is performed after the binarization process. In this binarization processing, the frequency value on the two-dimensional histogram is set to 0 when the frequency value is below a certain threshold value, and set to a constant other than 0 when the frequency value is greater than the threshold value (for example, when treating as an image, the maximum gradation value) 255). If this binarization processing is performed, information on noise-like elements (unimportant elements) on the two-dimensional histogram can be deleted. Therefore, it can be expected that the discrimination process described later will be more accurate.

  Next, as shown in FIG. 14, the set of points forming the two-dimensional histogram in the two-dimensional space is classified into at least one cluster, and a labeling process is performed to assign an ID (identification number) to each cluster. (S005).

Next, the feature regarding the shape, position, and frequency is calculated for each cluster to which the ID is assigned (S006). Here are some examples of features
I) Features related to the shape ・ Spread with respect to the main axis ・ An angle that the direction of the main axis forms a straight line of y = x ・ Roughness of the cluster ・ Area of the cluster
II) Features related to position-Cluster center of gravity
III) Features related to frequency-Sum of frequencies in cluster (If binarization processing was performed before this step S006, the sum of frequencies in each cluster in the frequency distribution before the binarization processing)
And so on.

  In this example, these features are used for discrimination. That is, after obtaining these features for each cluster, pattern discrimination is performed using these features (details regarding these features will be described in the following description).

  First, each cluster is classified into either “main cluster” or “sub-cluster” (S007). The “main cluster” refers to a cluster corresponding to a main part of a set of points of a two-dimensional histogram, like the cluster MC illustrated in FIG. The “main cluster” is limited to one. If there is one cluster, it becomes the main cluster. The “sub-cluster” refers to a cluster corresponding to a portion different from the main body portion corresponding to the main cluster, like the sub-cluster SC illustrated in FIG. There may be a plurality of “sub-clusters”. This classification can be performed using the above-described features relating to the “cluster area” and “cluster centroid position”. For example, the cluster having the largest area may be classified as the main cluster based on the characteristics related to the “cluster area”, and among the clusters having a predetermined area, the “cluster centroid position” is the largest at the coordinate origin. You may classify a near cluster as a main cluster.

  Next, discrimination processing based on the “feature” of the classified main cluster is performed (S008). In the determination processing for the main cluster, when it is determined that “the determination target pattern is different from the comparison target pattern” (YES in S009), the pattern determination processing is completed with this as a final result (S015). Details of the discrimination processing for the main cluster will be described later.

  If the “discrimination target pattern is not different from the comparison target pattern” in the discrimination process (NO in S009), pattern discrimination processing based on the “feature” of the subcluster is performed (S010 to S013). For example, when there are N sub-clusters, the discrimination processing of each sub-cluster is sequentially performed as k = 1, 2, 3,. If it is determined that “the determination target pattern is different from the comparison target pattern” in the determination result of a certain sub-cluster k (YES in S012), this is used as a final result, and the pattern determination process is completed at that time (S015). ). Even if the discrimination processing (S010 to S013) for all the sub-clusters (N) is completed, if it is not discriminated that “the discrimination target pattern is different from the comparison target pattern”, the final result is “the discrimination target pattern is compared”. The pattern discrimination process is completed as “similar to target pattern” (S014). In this case, a provisional discrimination result that “the discrimination target pattern is similar to the comparison target pattern” is obtained in the discrimination processing for all the subclusters k = 1, 2, 3,. Details of the discrimination processing based on the characteristics of individual sub-clusters will be described later.

  The above is the overall flow of the pattern discrimination process.

  FIG. 15 shows the flow of the discrimination process for the main cluster (S008 in FIG. 11).

  In this determination process, first, it is determined whether or not the spread of the main cluster with respect to the main axis is large based on the feature related to the shape of the main cluster (S101).

  In this example, the “principal axis” is determined by calculating the inertial principal axis of the two-dimensional histogram and matching the inertial principal axis. The degree of “spreading” with respect to the main axis means the degree of dispersion of elements in the main cluster with respect to the main axis. As this “spread”, for example, the variance may be calculated by the distance to the point farthest from the main axis, the Ferret system may be calculated, and all elements constituting the main cluster may be calculated. The variance may be calculated from the frequency value. In the example of FIG. 16A, this “spread” is shown as a distance E between points farthest from the main axis M of the main cluster MC. In this example, the description will be continued as a spread E with respect to the main axis M.

  The reason for performing this determination (S101 in FIG. 15) is that when the spread E with respect to the main axis M is large, “there is any significant change in the shading pattern” between the comparison target pattern and the determination target pattern, or the comparison target pattern This is because it is estimated that the pattern to be discriminated is “there is little difference in shading in the pattern area”.

  If it is determined in step S101 that the spread E with respect to the main axis M of the main cluster is large (YES in S101), it is subsequently determined whether or not the unevenness of the main cluster is high (S102).

  Here, “degree of unevenness” means an index representing the degree of unevenness of the shape of the main cluster. “High unevenness” means that the shape of the main cluster is “more bumpy”. In order to quantify the “concave / convex degree”, for example, as shown in FIG. 16B, the convex hull of the main cluster (the smallest convex polygon including the cluster) is obtained, and then the convex hull is obtained. The ratio of the perimeter of the area (convex hull area) J occupied by the area to the perimeter of the area (cluster area) F occupied by the actual main cluster, or the area F of the area J occupied by the convex hull and the area F occupied by the actual main cluster It is defined by the ratio with the area. As described above, when the “concave / convex degree” is defined by the ratio of the peripheral length or the ratio of the area, the more uneven the main cluster shape is, the larger (higher) the unevenness value is. Further, if there is almost no unevenness in the shape of the cluster, the value of the unevenness degree becomes small (approaching 1.0). In this example, the description will be continued assuming that the “degree of unevenness” is defined in this way.

  If it is determined in step S102 in FIG. 15 that the unevenness degree of the main cluster is high (YES in S102), there is a high possibility that “there is any significant change in the shading pattern”. Therefore, it is determined that “the determination target pattern is different from the comparison target pattern” (S105). As a result, the determination process for the main cluster is completed, and the pattern determination process is also ended as a whole.

  On the other hand, when it is determined that the unevenness degree of the main cluster is low (NO in S102), it can be interpreted as “there is little difference in shading within the pattern region”, and thus “the determination target pattern is similar to the comparison target pattern”. A provisional determination is made (S104). As a result, the main cluster discrimination processing is completed.

  On the other hand, if it is determined in step S101 that the spread E with respect to the main axis M of the main cluster is small (NO in S101), then the direction θ of the main axis M (angle measured counterclockwise from the horizontal axis) is 45 degrees. It is discriminated whether or not it is out of the range (S103).

  The reason for performing this determination is that it cannot be said that there is almost no change in light and shade in the entire light and shade pattern just by saying “the spread E with respect to the main axis M is small”. Whether or not the direction θ of the main axis M deviates from 45 degrees means whether or not the main axis M deviates from the linear direction of y = x as Δθ shown in the drawing.

  If it is determined in step S103 that the direction θ of the spindle M is not deviated from 45 degrees (NO in S103), there is almost no change in light and shade in the entire light and shade pattern, that is, “the determination target pattern is similar to the comparison target pattern”. It is tentatively determined that there is (S104). Thereby, the discrimination process for the main cluster is completed.

  On the other hand, if it is determined in step S103 that the direction θ of the spindle M is out of 45 degrees (YES in S103), the reason for the deviation is that the gray value changes in the gray pattern area (for example, Or the sub-cluster SC is present in addition to the main cluster MC, and the main cluster M is tilted due to the sub-cluster (the sub-cluster is so affected that the tilt of the main axis M is affected). Or the sum of the frequencies in the cluster SC is large). In any case, since there is a change in the entire shade pattern, it is determined that “the determination target pattern is different from the comparison target pattern” at this point (S105). As a result, the determination process for the main cluster is completed, and the pattern determination process is also ended as a whole.

  FIG. 17 shows the flow of discrimination processing (S011 in FIG. 11) for the individual sub-clusters.

When a sub-cluster exists in the above-described two-dimensional histogram, the following two types can be considered as the properties of the sub-cluster.
(A) What should originally be part of the main cluster (discontinued)
(A) Completely independent of the main cluster In the discrimination processing of FIG. 17, the sub-cluster shown in FIG. 18 is considered in consideration of the nature of the sub-cluster (a) or (b). A determination is made using a distance D between the SC and the main cluster MC (hereinafter simply referred to as “distance to the main cluster” D) and a “frequency sum” in the sub-cluster. Specifically, it is as follows.

  First, it is determined whether or not the “distance from the main cluster” is large (S201).

  Here, when it is determined that the “distance to the main cluster” is small (NO in S201), the subcluster is considered to be a noise component of the main cluster, and the subcluster is originally a part of the main cluster. It is determined that what should have been interrupted (above a). Therefore, the sub-cluster is not used for pattern discrimination, and at this time, it is provisionally determined that “the discrimination target pattern is similar to the comparison target pattern” (S203). Then, the process proceeds to analysis of another subcluster.

  In this step S201, not only the “distance to the main cluster” but also a criterion of whether or not the center of gravity is distributed on a straight line of y = x may be used (the subcluster is This is an interpretation that the main cluster existing on the straight line y = x is interrupted.)

  On the other hand, if it is determined in step S201 that the “distance to the main cluster” is large (YES in S201), it is then determined whether the “frequency sum” in the sub-cluster is large (S202).

  The reason for this determination is as follows. That is, if it is determined in step S201 that the “distance to the main cluster” is large (YES in S201), it can be considered that the sub-cluster represents a strong light / dark variation of the light / dark pattern. Furthermore, the sub-cluster is either a noisy component (variation that does not affect pattern similarity determination) or not (variation that affects density pattern similarity determination). It is believed that there is. Here, it is considered that a case where the “frequency sum” in the sub-cluster is small is not a serious change in the original shading pattern.

  Therefore, when it is determined that the “frequency sum” of the subcluster is small (NO in S202), the subcluster is considered to be a noise component. Therefore, the sub-cluster is not used for pattern discrimination, and at this time, it is provisionally determined that “the discrimination target pattern is similar to the comparison target pattern” (S203). Then, the process proceeds to analysis of another subcluster.

  On the other hand, if it is determined that the “frequency sum” in the sub-cluster is large (YES in S202), the sub-cluster is considered to be a fluctuation that affects the similarity determination of the density pattern. It is determined that the target pattern is different from the comparison target pattern (S204). Then, the discrimination processing for the sub-cluster is finished, and the pattern discrimination processing is also finished as a whole.

  However, when it is determined that the sub-cluster is a variation related to the similarity determination of the density pattern, a more accurate determination may be performed. Specifically, it is confirmed at which position (coordinates) the gray value forming the subcluster appears on the original gray pattern, and the appearing position is continuous or periodic. Or whether it is dispersed in a spot manner. For this purpose, a specific gray value is extracted from the sub-cluster by performing clipping processing, or a connected component of the gray value is labeled, and the area / proximity of the labeled connected component is calculated. As a result, it is confirmed whether the sub-cluster is a fluctuation that really affects pattern similarity determination or a noise component. If it is determined that the sub-cluster is a noisy component in this more accurate determination, the sub-cluster is not used for pattern determination, and it is tentative that "the determination target pattern is similar to the comparison target pattern". (S203). Then, the process proceeds to analysis of another subcluster.

  19-1-1 to 19-1-3 show that the data sets A-2, A-3,..., A-13 shown in FIGS. It shows how the pattern is discriminated by the shape shading pattern discrimination method. 19-2-1 to 19-2-3 show that the data sets B-2, B-3,..., B-13 shown in FIGS. It shows how the pattern is discriminated by the shape shading pattern discrimination method. In addition, FIGS. 19-3-1 to 19-3-3 show that the data sets C-2, C-3,..., C-13 shown in FIGS. It shows how it is discriminated by the shading pattern discrimination method of this embodiment. In FIGS. 19-1-1 to 19-3-3, the leftmost column (a) indicates the number of the target data set. The column (b) in the second column from the left shows a two-dimensional histogram created from the corresponding data set (normalized gray value and before binarization). The (c) column in the third column from the left shows a binarized two-dimensional histogram created from the corresponding data set. The (d) column in the fourth column from the left shows a state where the corresponding two-dimensional histogram is classified into clusters (clustering). The determination result is shown in the rightmost column.

  In these FIG. 19-1-1 to FIG. 19-3-3, attention is first focused on a data set to be determined as “a discrimination target pattern is similar to a comparison target pattern”.

  For example, regarding data set A-2 to data set A-6 and data set B-2 to data set B-6, there is only one cluster (main cluster MC). According to the discrimination processing for the main cluster, the spread of the main cluster with respect to the main axis is small, and the direction of the main axis is substantially equal to the direction of 45 degrees (y = x). Therefore, it is determined as “similar” and a correct determination result is obtained.

  Further, regarding data set C-2 to data set C-6, there is only one cluster (main cluster MC). According to the discrimination processing for the main cluster, the main cluster has a large spread with respect to the main axis, but the degree of unevenness of the shape is low. Therefore, it is determined as “similar” and a correct determination result is obtained.

  Next, attention is focused on the data set that should be determined as “the determination target pattern is different from the comparison target pattern”.

  For example, regarding the data set A-7 to the data set A-10 and the data set B-7 to the data set B-11, according to the determination process for the main cluster, the main cluster has a large spread with respect to the main axis, and the shape unevenness degree Is expensive. Therefore, it is determined to be “different”. Note that some of these data sets actually have sub-clusters, but no determination is made regarding those sub-clusters (the determination result is “no sub-cluster”). The result of being “different” as it is is the final discrimination result, and the correct discrimination result is obtained.

  Each of the data sets A-11, A-12, A-13 and the data sets B-12, B-13 includes one main cluster MC and at least one subcluster SC (data For set A-11, there are two sub-clusters SC1, SC2.) With respect to these data sets, according to the discrimination processing for the main cluster, the spread of the main cluster with respect to the main axis is small, and the direction of the main axis is close to the direction of 45 degrees (y = x). Therefore, it is determined as “similar”. However, in the discrimination processing for the sub-cluster, the sub-cluster is determined as a separate one independent from the main cluster. From the sum of the frequencies in the sub-cluster, it is finally discriminated to be different and a correct discrimination result is obtained.

  For data sets C-12 and C-13, there is only one cluster (main cluster MC). According to the determination processing for the main cluster, in the determination of the main cluster, the spread of the main cluster with respect to the main axis is small, but is greatly deviated from the 45 degree direction. Therefore, it is determined to be “different”. The determination regarding the sub-cluster is not performed, and the determination result becomes the final determination result, and the correct determination result is obtained.

  As described above, according to this light / dark pattern discrimination method, it is robust against changes in brightness, and even if the light / dark difference in the pattern is large or small, an unknown light / dark abnormality or defect is included in the pattern. Even if it is, the similarity can be determined with high accuracy.

  Further, the process of determining whether or not the determination target pattern is similar to the comparison target pattern is sequentially performed for each cluster. Then, when it is determined as “distinct” based on one of the main cluster and the sub-cluster, the determination process is immediately terminated. Therefore, it is possible to omit the discrimination processing for useless clusters. Therefore, the time for the discrimination process can be shortened as a whole.

  FIG. 20 shows a configuration of a system 100 including a light / dark pattern discriminating apparatus 110 according to an embodiment suitable for carrying out the light / dark pattern discriminating method.

  The system 100 includes a light / dark pattern discriminating device 110, a database 101 for collecting and storing data from a factory production line, an input device 102, a display device 108, and an output device 109.

  The database 101 is composed of an external storage device such as a hard disk drive. In this example, the database 101 stores data representing the discrimination target pattern and the comparison target pattern.

  The input device 102 is composed of, for example, a keyboard and a mouse. In this example, the input device 102 is used by the user to input a specification of a pattern to be analyzed to the gray pattern discriminating device 110.

  The display device 108 is a device that can display an image such as an LCD (liquid crystal display panel) or a CRT (cathode ray tube). It is used to display the pattern to be analyzed and the discrimination result as an image.

  The output device 109 outputs the result of characteristic distribution analysis by the light / dark pattern discriminating device 110 through paper output or a magnetic disk or portable semiconductor memory.

  The light / dark pattern discriminating apparatus 110 includes a data input unit 103, a normalization processing unit 104, a two-dimensional histogram creation unit 105, a similarity determination unit 106, and a discrimination result output unit 107.

  The data input unit 103 receives data representing the discrimination target pattern and the comparison target pattern from the database 101, processes the discrimination target pattern and the comparison target pattern into the same size, and aligns them. Further, the data input unit 103 receives the designation of the pattern to be analyzed from the input device 102.

  The normalization processing unit 104 performs a normalization process on the gray value of the discrimination target pattern to generate a first normalized gray value, and performs a normalization process on the gray value of the comparison target pattern to generate a second value. Generate normalized gray values. Therefore, the determination result obtained by the similarity determination unit 106 described later is robust against the brightness change.

  The two-dimensional histogram creation unit 105 sets a two-dimensional space having the first normalized gray value and the second normalized gray value as variables and two coordinate axes orthogonal to each other, and the two-dimensional histogram is stored in the two-dimensional space. The points represented by the first normalized gray value and the second normalized gray value at positions corresponding to each other on the discrimination target pattern and the comparison target pattern are accumulated, and the first on the discrimination target pattern is accumulated. A two-dimensional histogram representing a correlation between one normalized gray value and the second normalized gray value on the comparison target pattern is created.

  The similarity determination unit 106 classifies the set of the points forming the two-dimensional histogram into at least one cluster, and combines the feature related to the shape, position, or frequency with respect to the cluster, or a combination of features related to the shape, position, or frequency. Based on this, it is determined whether or not the determination target pattern is similar to the comparison target pattern.

  The discrimination result output unit 107 converts the discrimination result obtained by the similarity discrimination unit 106 into a data format suitable for the display device 108 and the output device 109 and outputs the data format.

  According to the shading pattern discriminating apparatus 110, whether the shading difference in the pattern is large or small, or whether an unknown shading abnormality or defect is included in the pattern, can be discriminated with high accuracy.

  In addition, you may construct | assemble as a program for making a computer perform the above-mentioned shade pattern discrimination method.

  Also, such a program may be recorded and distributed on a computer-readable recording medium such as the CD-ROM 130 shown in FIG. By installing the program on a general-purpose computer, the gray-scale pattern discrimination method can be executed by the general-purpose computer.

It is a figure which shows the example of the general shading pattern showing the film-forming state. It is a figure which illustrates the state which divided the light / dark pattern showing a normal film-forming state, and the light / dark pattern when abnormality occurs in the film-forming state into 9 divided vertically 3 × 3. It is a figure which shows the aspect which performs the similarity determination of a light / dark pattern for every division area in FIGS. It is a figure which shows the aspect which performs similarity determination of the shading pattern of an initial state, when the shading pattern showing a film-forming state changes with time. It is a figure which shows the aspect which performs the similarity determination of the light / dark pattern showing the film-forming state after two months. It is a figure which illustrates three different kinds of light and shade patterns A, B, and C of a square serving as a reference. It is a figure which shows the light / dark pattern which adds shading etc. to the pattern A of FIG. It is a figure which shows the light / dark pattern which adds shading etc. to the pattern A of FIG. It is a figure which shows the shading pattern formed by adding shading etc. to the pattern B of FIG. It is a figure which shows the shading pattern formed by adding shading etc. to the pattern B of FIG. It is a figure which shows the shading pattern formed by adding shading etc. to the pattern C of FIG. It is a figure which shows the shading pattern formed by adding shading etc. to the pattern C of FIG. It is a figure which illustrates the discrimination | determination result when the similarity is discriminate | determined using normalization correlation regarding data set A-1, ..., A-4 of a comparison object pattern and a discrimination | determination object pattern, and those data sets. It is a figure which illustrates the discrimination | determination result when the similarity is discriminate | determined using the normalization correlation regarding data set A-5, ..., A-9 of a comparison object pattern and a discrimination | determination object pattern, and those data sets. It is a figure which illustrates the discrimination | determination result at the time of discriminating similarity using the data set A-10, ..., A-13 of a comparison object pattern and a discrimination | determination object pattern, and normalization correlation regarding those data sets. It is a figure which illustrates the discrimination | determination result when the similarity is discriminate | determined using normalized correlation regarding the data sets B-1, ... B-4 of the comparison target pattern and the discrimination target pattern, and those data sets. It is a figure which illustrates the discrimination | determination result at the time of discriminating similarity using the data set B-5, ..., B-9 of a comparison object pattern and a discrimination | determination object pattern, and normalization correlation regarding those data sets. It is a figure which illustrates the discrimination | determination result at the time of discriminating similarity using normalization correlation regarding data set B-10, ..., B-13 of a comparison object pattern and a discrimination | determination object pattern, and those data sets. It is a figure which illustrates the discrimination | determination result when the similarity is discriminate | determined using the normalized correlation regarding the data sets C-1, ... C-4 of the comparison target pattern and the discrimination target pattern, and those data sets. It is a figure which illustrates the discrimination | determination result when the similarity is discriminate | determined using the normalized correlation regarding the data sets C-5, ..., C-9 of the comparison target pattern and the discrimination target pattern, and those data sets. It is a figure which illustrates the discrimination | determination result when the similarity is discriminate | determined using the normalization correlation regarding the data sets C-10, ..., C-13 of the comparison target pattern and the discrimination target pattern, and those data sets. It is a figure which shows the aspect in which the prior art 2 produces the correct discrimination result regarding data set A-13. It is a figure which shows the aspect in which the prior art 2 produces the correct discrimination result regarding data set B-13. It is a figure which shows the aspect in which the prior art 2 produces the correct discrimination result regarding data set C-13. It is a figure which shows the aspect in which the prior art 2 produces | generates the discrimination | determination result of an error regarding the data set C-9. It is a figure which shows the aspect in which the prior art 2 produces | generates the discrimination | determination result of an error regarding data set A-8. It is a figure which shows the aspect in which the prior art 3 produces the discrimination | determination result of an error regarding the data set C-2. It is a figure which shows the aspect in which the prior art 3 produces the discrimination | determination result of an error regarding the data set C-3. It is a figure which shows the aspect in which the prior art 3 produces the discrimination | determination result of an error regarding the data set C-4. It is a figure which shows the aspect in which the prior art 3 produces the discrimination | determination result of an error regarding the data set C-5. It is a figure which shows the aspect in which the prior art 3 produces | generates the discrimination | determination result of an error regarding data set A-2. It is a figure which shows the aspect in which the prior art 3 produces | generates the discrimination | determination result of an error regarding data set B-2. It is a figure explaining the shading provided with respect to a specific data set. It is the figure which showed the example of the two-dimensional histogram produced | generated from the data set A-1 which contrasted the same pattern. It is the figure which showed the example of the two-dimensional histogram produced | generated from the data set B-1 which contrasted the same pattern. It is the figure which showed the example of the two-dimensional histogram produced | generated from the data set C-1 which contrasted the same pattern. It is a figure which shows the flow of the light / dark pattern discrimination method of one Embodiment of this invention. It is a figure explaining the production | generation method of the two-dimensional histogram in the said light / dark pattern discrimination method. It is a figure explaining the filtering process with respect to the said two-dimensional histogram in the said shading pattern discrimination method. It is a figure explaining the labeling process to the cluster obtained from the said two-dimensional histogram in the said light / dark pattern discrimination method. It is a figure which shows the flow of the discrimination | determination process with respect to the main cluster obtained from the said two-dimensional histogram. It is a figure explaining the example of the characteristic regarding the said main cluster. It is a figure which shows the flow of the discrimination | determination process with respect to the subcluster obtained from the said two-dimensional histogram. It is a figure which illustrates and illustrates the distance between a subcluster and a main cluster as a feature of a subcluster. It is a figure explaining the discrimination | determination result by the said light / dark pattern discrimination method about the said data sets A-2, ..., A-5. It is a figure explaining the discrimination | determination result by the said shading pattern discrimination method about the said data sets A-6, ..., A-9. It is a figure explaining the discrimination | determination result by the said light / dark pattern discrimination method about the said data sets A-10, ..., and A-13. It is a figure explaining the discrimination | determination result by the said light / dark pattern discrimination method about the said data sets B-2, ..., B-5. It is a figure explaining the discrimination | determination result by the said shading pattern discrimination method about the said data sets B-6, ..., B-9. It is a figure explaining the discrimination | determination result by the said light / dark pattern discrimination method about the said data sets B-10, ..., B-13. It is a figure explaining the discrimination | determination result by the said light / dark pattern discrimination method about the said data sets C-2, ..., C-5. It is a figure explaining the discrimination | determination result by the said light / dark pattern discrimination method about the said data sets C-6, ..., C-9. It is a figure explaining the discrimination | determination result by the said shading pattern discrimination method about the said data sets C-10, ..., C-13. It is a figure which shows the structure of the system containing the light / dark pattern discrimination | determination apparatus of one Embodiment of this invention.

Explanation of symbols

MC Main cluster SC Sub cluster 110 Shading pattern discrimination device

Claims (10)

A density pattern determination method for determining whether or not a determination target pattern indicating a gray value distribution spreading in a certain area is similar to a comparison target pattern indicating a gray value distribution extending in a certain area,
A normalization process is performed on the gray value of the discrimination target pattern to generate a first normalized gray value, and a normalization process is performed on the gray value of the comparison target pattern to generate a second normalized gray value. ,
A two-dimensional space in which the first normalized gray value and the second normalized gray value are variables and two coordinate axes are orthogonal to each other is set, and the discrimination target pattern and the comparison are set in the two-dimensional space. The points represented by the first normalized gray value and the second normalized gray value at positions corresponding to each other on the target pattern are accumulated, and the first normalized gray value on the discrimination target pattern and the above Creating a two-dimensional histogram representing the correlation with the second normalized gray value on the comparison target pattern;
Classifying the set of points forming the two-dimensional histogram into at least one cluster;
A gray pattern for determining whether or not the determination target pattern is similar to the comparison target pattern based on the shape, position, or frequency characteristics related to the cluster, or a combination of characteristics related to the shape, position, or frequency How to determine. In the light / dark pattern discrimination method according to claim 1,
The gradation pattern discrimination method characterized in that the normalization process is a process of normalizing the density value distributions of the discrimination target pattern and the comparison target pattern to a gray value distribution having a magnitude of 1 and an average of 0, respectively. In the light / dark pattern discrimination method according to claim 1 or 2,
A gray-scale pattern discriminating method comprising performing filtering processing for removing noise components on the two-dimensional histogram before the processing of classifying the set of points forming the two-dimensional histogram into the clusters. In the light and shade pattern discrimination method according to claim 3,
The filtering process converts the frequency value to 0 when the frequency value on the two-dimensional histogram is less than or equal to a certain threshold value, and converts the frequency value to a constant of 1 or more when the frequency value is greater than the threshold value. A gradation pattern discriminating method characterized by including binarization processing. In the light / dark pattern discrimination method according to any one of claims 1 to 4,
The process of classifying the set of points forming the two-dimensional histogram into the clusters extracts a main cluster that forms the main part of the set of points,
The process of determining whether or not the determination target pattern is similar to the comparison target pattern includes, as the characteristics serving as a reference for the determination, a spread with respect to the main axis of the main cluster, an unevenness degree of the main cluster, and A method for discriminating light and shade patterns, wherein the direction of the main axis of the main cluster is used. In the light and shade pattern discrimination method according to claim 5,
In the process of classifying the set of points forming the two-dimensional histogram into the clusters, in addition to the main cluster, a sub-cluster representing a part different from the main body part is extracted from the set of points,
The process of determining whether or not the determination target pattern is similar to the comparison target pattern includes, as the characteristics serving as a reference for the determination, the distance between the sub-cluster and the main cluster, the sub-cluster A gray-scale pattern discriminating method using the position of the center of gravity, the frequency sum of the points in the sub-cluster, and / or the sub-cluster area. In the light and shade pattern discrimination method according to any one of claims 1 to 6,
The process of determining whether or not the determination target pattern is similar to the comparison target pattern is sequentially performed for each cluster, and the determination target pattern is based on any one of the clusters. A gray-scale pattern discriminating method characterized in that the discriminating process is immediately terminated when it is judged that the image is not similar to the above. A light / dark pattern determining apparatus for determining whether or not a determination target pattern indicating a gray value distribution spreading in a certain area is similar to a comparison target pattern indicating a gray value distribution extending in a certain area,
A data input unit for inputting gray value data of the discrimination target pattern and the comparison target pattern;
Normalization processing is performed on the gray value of the discrimination target pattern to generate a first normalized gray value, and normalization processing is performed on the gray value of the comparison target pattern to generate a second normalized gray value. A normalization processing unit;
A two-dimensional space in which the first normalized gray value and the second normalized gray value are variables and two coordinate axes are orthogonal to each other is set, and the discrimination target pattern and the comparison are set in the two-dimensional space. The points represented by the first normalized gray value and the second normalized gray value at positions corresponding to each other on the target pattern are accumulated, and the first normalized gray value on the discrimination target pattern and the above A two-dimensional histogram creation unit that creates a two-dimensional histogram representing a correlation with the second normalized gray value on the comparison target pattern;
The set of points constituting the two-dimensional histogram is classified into at least one cluster, and the discrimination target pattern is based on a feature related to the shape, position or frequency related to the cluster, or a combination of features related to the shape, position or frequency. A gray pattern discriminating apparatus comprising: an analogy discriminating unit that discriminates whether or not the pattern is similar to the comparison target pattern.   A program for causing a computer to execute the shading pattern discrimination method according to any one of claims 1 to 7.   A computer-readable recording medium on which the program according to claim 9 is recorded.