KR101782364B1 - Vision inspection method based on learning data - Google Patents

Abstract

The present invention relates to a learning-based vision test method using k-average clustering, which includes: a pre-learning step, a defect determination step, and an additional learning step. The pre-learning step includes a k-average clustering step of forming a first learning group including the same number of genuine samples being genuine defect samples and false samples being false defect samples; extracting a first data group including the same number of pieces of formulated genuine and false data from the first learning group; and executing a k-average clustering operation with k-average points by obtaining the average of the feature points in each group after grouping the first data group into k groups. The pre-learning step includes, after the k-average clustering step, the process of deriving classification references on the border of a genuine area with the genuine data and a false area with the false data by schematizing the first data group with the k-average points applied thereto on a feature space. The defect determination step includes the processes of: applying the inspection target data formulated and extracted from a display panel, to which a polarizing layer used as an inspection target is attached, into the feature space; determining an area with the inspection target data in the feature space by using the classification references as the borders; and determining genuine or false defects of the inspection target. The additional learning step applies the inspection target data from the pre-learning step to modify the classification references in the feature space. The value of k is set to maximize the distance between the average points in the genuine and false areas in the feature space. The defect determination step and the additional learning step are repeatedly executed.

Description

{VISION INSPECTION METHOD BASED ON LEARNING DATA}

The present invention relates to a learning-based vision inspection method using K average clustering, more specifically, to simplify a calculation process by deriving an optimal k value while minimizing computation in performing K average clustering, And more particularly, to a vision inspection method in which reliability is improved by making the classification more clear.

In general, clustering is a method of dividing a given data into several clusters. It corresponds to unsupervised learning which learns by itself without prior knowledge.

Clustering is a method that enables efficient understanding and utilization of given data by dividing into clusters of similar data entities.

There are various clustering techniques such as hierarchical clustering and partition clustering, which are clustered into several clusters according to characteristics of large-scale data. Among them, the k-means clustering algorithm is easy to implement, The amount of computation is large and the time required is increased.

As a method for detecting defects in the step of sorting good and defective parts by checking the adhesion state of the sheet or panel in the production process of the display panel, there is introduced a vision inspection device for photographing and inspecting the plane to be inspected of the display panel as an image .

In applying K average clustering as an algorithm to identify defects of vision inspection devices, clustering is performed by dividing classes into k clusters in the process of processing given data for learning.

At this time, the number k of clusters is a very important factor for clustering, and the clustering result greatly varies depending on the value of k.

However, as a method of obtaining the k value in the K average clustering algorithm, it is necessary to perform a number of operations by combining the iterative K average clustering and the external cluster association criterion information for evaluating the quality of the cluster, There is a disadvantage in that a long time investment and a complicated operation must be performed.

This is because it is difficult to apply the present invention to a process of resetting a criterion for determining a defect through an iterative learning process in determining whether a polarizing layer adhered to a display panel is defective or not.

Therefore, a method for solving such problems is required.

SUMMARY OF THE INVENTION The present invention has been made in order to solve the above-mentioned problems of the prior art, and it is an object of the present invention to provide a vision inspection apparatus which increases reliability and accuracy through prior learning and repeated learning, This is to provide a vision inspection method that can calculate the optimal k value derived from the K average clustering in a small amount of calculation and in a short time.

The problems of the present invention are not limited to the above-mentioned problems, and other problems not mentioned can be clearly understood by those skilled in the art from the following description.

In order to accomplish the above object, a learning-based vision inspection method based on k average clustering of the present invention is characterized in that a first learning group is formed by including an intrinsic sample, which is a sample of intrinsic defects, and a tentative sample, Extracting a first data group including the same number of the intrinsic data and the false data formulated from the first learning group, grouping the first data group into k groups, and obtaining a minutiae point average for each group, K average clustering is performed and the first data group assigned to the k average points is plotted on the feature space and the classification region which is the boundary between the intrinsic region where the intrinsic data is located and the pseudo region where the pseudo- , A pre-learning step of deriving a pre-learning step, a test target extracted and formulated from a display panel to which a polarizing layer A defect discrimination step of judging an inspection object as an intrinsic defect or a false defect by determining an area where the test data is located in the feature space with the classification criterion as a boundary and applying the test data to the pre- And the k value is set so that the distance between the mean points of the intrinsic region and the false region in the feature space is maximized, and the defect determination step and the additional learning step are repeatedly performed.

The pre-learning step includes an image collecting step of collecting a first learning image group composed of a genuine image and a false image extracted from each of the intrinsic specimens included in the first learning group and the image information from the specimen, The pseudo-data is derived as n feature points from each of the true image and the false image included in the first learning image group to form a first data group, a first data group is schematized, and adjacent feature points are grouped into k groups A dictionary creation step of creating a dictionary composed of k average points by obtaining feature point averages for each group of the feature points and the vital data and the false data, And a classifying step of determining a classifying criterion that is a boundary of the pseudo-zone in which the categorizing section is located.

Alternatively, the preprocessing step and the classifying step are performed first by assigning arbitrary k values, re-performed with k values at which the distance between the average points of the intrinsic region and the intrinsic region derived from the arbitrary k value becomes maximum, .

In the defect determination step, the test data, which is n feature points obtained from the image of the subject extracted from the image information, is compared with the dictionary, and the test data is classified based on k average points and displayed on the feature space.

Alternatively, the intrinsic image, the false image, and the image to be inspected are image information of a point at which a modulated signal of the photographed image is abruptly changed, moving the inspected surface of the inspected object along a predetermined pattern.

1 is a flowchart illustrating a vision-based vision inspection method using K average clustering according to an exemplary embodiment of the present invention.
FIG. 2 is a flowchart illustrating data for defining a classification criterion in a learning-based vision inspection method using K average clustering according to an exemplary embodiment of the present invention.
FIG. 3 is an image showing an intrinsic image of an intrinsic specimen in a learning-based vision inspection method using K average clustering according to an embodiment of the present invention.
FIG. 4 is an image showing a false image of a false sample in a learning-based vision inspection method using K average clustering according to an embodiment of the present invention.
FIG. 5 is a schematic diagram showing a process of deriving a dictionary in a learning-based vision inspection method using K average clustering according to an embodiment of the present invention.
FIG. 6 is a schematic diagram illustrating a process of assigning a first image group in advance in a learning-based vision inspection method using K average clustering according to an exemplary embodiment of the present invention.
FIG. 7 is a graph illustrating a distance between averages of classes in a learning-based vision inspection method using K average clustering according to an exemplary embodiment of the present invention.
FIG. 8 is a schematic diagram showing a learning step and a defect determination step in a learning-based vision inspection method using K average clustering according to an embodiment of the present invention.
FIG. 9 is a graph illustrating a three-dimensional state in which an intrinsic region and a false region are derived on a feature space in a learning-based vision inspection method using K average clustering according to an exemplary embodiment of the present invention.
FIG. 10 is a state diagram showing an apparatus using a vision-based vision inspection method using K average clustering according to an exemplary embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings. In describing the present embodiment, the same designations and the same reference numerals are used for the same components, and further description thereof will be omitted.

The learning-based vision inspection method through K average clustering according to the present invention can be performed as follows.

FIG. 1 is a flowchart illustrating a vision-based vision inspection method based on K average clustering according to an exemplary embodiment of the present invention. FIG. 2 is a flowchart illustrating a vision-based vision inspection method using K average clustering according to an exemplary embodiment of the present invention. Fig. 6 is a flowchart showing data for defining a classification criterion. Fig.

The learning-based vision inspection method using K average clustering according to an exemplary embodiment of the present invention is an example of a learning-based vision inspection method for inspecting defects of the polarizing layer 304 adhered to the display panel 302. FIG.

The learning-based vision inspection method based on K average clustering according to an embodiment of the present invention includes a pre-learning step (S100), a defect determination step (S200), and an additional learning step (S300) And the additional learning step S300 are repeatedly performed.

In the pre-learning step (SlOO), the first learning group is prepared by including the intrinsic specimen 100, which is a sample of intrinsic defective product, and the tentative specimen 200, which is a sample of a false defect, in the same number.

The first data group including the same number of the intrinsic data 120 and the pseudo data 220 extracted from the first learning group is extracted.

The first data group is grouped into k groups, and a feature point average is obtained for each group, and a k average clustering step is performed, which is composed of k average points. The first data group assigned to k average points is divided into a feature space 400 To derive a classification criterion 410 that is the boundary between the intrinsic region 130 where the intrinsic data 120 is located and the false region 230 where the false data 220 is located.

At this time, the k value is set so that the distance between the average points of the intrinsic region 130 and the pseudo region 230 in the feature space is maximized.

In the defect determination step S200, the inspection data 320 extracted from the display panel 302 on which the polarizing layer 304, which is the inspection object 300, is adhered and formulated is substituted on the feature space 400.

An area where the test data 320 is located is determined on the feature space 400 with the classification reference 410 as a boundary so that the test object 300 is judged to be an intrinsic defect or a false defect.

In the additional learning step S300, the test data 320 is applied to the pre-learning step S100 to modify the classification criteria 410 on the feature space 400. The test data 320 includes intrinsic data 120, The data 220 is corrected through the data balancing step S310 so as to form the first data group with the same number.

Each of the above steps will be described in detail below.

The first learning group includes the same number as the intrinsic specimen 100, which is a sample of the intrinsic defective product, and the simulated specimen 200, which is a sample of the false defective product.

The first learning image group includes the same number of the intrinsic image 110 derived from the intrinsic specimen 100 and the intrinsic image 210 derived from the pseudo-specimen 200. [

The first data group includes the same number of the intrinsic data 120 in which the intrinsic image 110 is formulated and the false data 220 in which the virgin image 210 is formulated.

Therefore, it is assumed in the embodiment of the present invention that the intrinsic specimen 100 and the intrinsic image 110 are the same A, and the false specimen 200 and the false image 210 are also the same A.

The feature space 400 is a multi-dimensional space having a plurality of variables, and refers to a space in which specific data according to an embodiment of the present invention is depicted.

The pre-learning step S100 includes an image collecting step S110 for obtaining an intrinsic image 110 and a false image 210, which are image images of suspected bad points from the intrinsic specimen 100 and the false specimen 200, Creating step S120 and class classification step S130.

FIG. 3 is an image showing an intrinsic image of an intrinsic specimen in a learning-based vision inspection method based on the K mean clustering according to an exemplary embodiment of the present invention. FIG. Based visual inspection method.

3 to 4, in the image capturing step S110, an image is collected through a photographing unit for photographing the intrinsic specimen 100 and the tentative specimen 200 by moving the measurement target plane in a predetermined pattern, And acquiring the intrinsic image 110 or the false image 210 by capturing an image of the position at the time of deriving a result satisfying predetermined conditions.

In an embodiment of the present invention, when a change in brightness of an image picked up through a photographing unit is rapidly increased or decreased, the point may be photographed and acquired as a video image as shown in FIGS.

The preprocessing step S120 derives the intrinsic data 130 and the pseudo data 230 derived from the intrinsic image 110 and the voiced image 210, respectively, into n feature points.

The intrinsic data 130 is obtained by A x n, and the permutation data 230 is also obtained by A x n.

FIG. 5 is a schematic diagram illustrating a process of deriving a dictionary in a learning-based vision inspection method using K average clustering according to an exemplary embodiment of the present invention. FIG. FIG. 7 is a schematic diagram illustrating a process of assigning a first learning image group in a vision-based vision inspection method according to an exemplary embodiment of the present invention. FIG. Fig.

5 through 7, the intrinsic data 130 and the falsification data 230 derived from the intrinsic image 110 and the tentative image 210 are used to generate color information in the intrinsic image 110 and the tentative image 210, Such as brightness, saturation, shape, curvature, etc., can be extracted from n points.

The intrinsic data 130 and the pseudo data 230 are converted into a vector quantity having characteristics such as color, lightness, saturation, shape, curvature, etc. as variables, and a multidimensional space Lt; / RTI >

FIG. 5 schematically shows a process in which n feature points are derived as vector quantities from the intrinsic image 110 and the tentative image 210 and are expressed in a multidimensional space.

It should be noted that the intrinsic data 130 and the pseudo data 230 may be extracted based on various factors according to an embodiment to which the present invention is applied.

As described above, the A.sub.n intrinsic data 130 from the A intrinsic images 110, the A.times.n pseudonymous data 230 from the A pseudonym images 210 are collected as vector quantities, and the intrinsic data 130 ) And fuzzy data 230 are graphically depicted on a multidimensional space and grouped into k groups by grouping adjacent intrinsic data 130 and fuzzy data 230 respectively into intrinsic data 130 and pseudo data 230 belonging to each group, The average point of the data 230 is calculated to calculate k average points.

At this time, each group is defined as a word, and a k-word dictionary is created by defining k average points as a dictionary.

라 하고, w는 클러스터에 속한 진성데이터(130) 및 가성데이터(230)의 패턴집합이며,

는 클러스터에 속한 특정 패턴으로서

와 같은 수식으로 정의될 수 있다.Specifically, each group is an individual cluster in the K average clustering algorithm, and k average points are centers of the respective clusters. The mean or centroid of the patterns belonging to the cluster,

W is a pattern set of the intrinsic data 130 and the false data 230 belonging to the cluster,

Is a specific pattern belonging to the cluster

And so on.

In the embodiment of the present invention, the number of k is arbitrarily selected and executed when the preprocessing step (S120) is performed for the first time, and the number of k derived through the k-value optimization process to be described later is applied again at the time of execution.

The preprocessing step (S120) may be performed using K average clustering, which is an algorithm for grouping the given data into k clusters.

The class classification step S130 substitutes the intrinsic data 130 into the dictionary created through the pre-creation step S120 to histogram the intrinsic data 130 belonging to each of the k words as shown in Fig. 6 The histogramming step is performed first.

The above-described pre-creation step S120 is a method for determining a variable k, which is the number of groups and the number of averages, in utilizing the K average clustering, and the intrinsic region 130 and the false region 130, which are disposed on the feature space 400, (K) in which the distance (d) between the averaging points of the reference points 230 is maximized.

As shown in FIG. 6, when the feature space 400 is shown in two dimensions, the value of k can be determined so that the distance d shown by the straight line is maximized, and the value of k formed in the intrinsic region 130 and the false region 230 When classifying three or more classes rather than two classes, it is necessary to sum the distances between the average points of the respective classes to determine the k value at which the maximum value is obtained.

The method for obtaining such an optimized k value is as shown in FIG. 7, and M in FIG. 7 is the number of classes to be learned.

와 같은 수식으로 클래스간의 평균점간 거리의 개수를 구할 수 있고,

로 표현한 바와 같이 클래스간의 평균점간 거리를 모두 합한 수치인 정도(accuracy)가 최대가 되는 k값을 최적의 k값이라 볼 수 있다.therefore ,

The number of the distance between the average points of the classes can be obtained by the equation such as < RTI ID = 0.0 >

The k value at which the accuracy, which is the sum of the average distances between classes, is maximized, can be regarded as the optimum k value.

Further, the number of k determined as described above is applied to the dictionary creation step (S120) so that a more accurate dictionary can be created.

The determination process of k may omit the verification procedure to be performed to determine the k value minimizing the error. The verification process may be performed by substituting the intrinsic specimen 100 and the tentative specimen 200 in advance It is possible to derive an optimum k value without a process of obtaining an optimal k value by trial and error by repeating the process of drawing on the space 400.

FIG. 8 is a schematic view showing a learning step and a defect determination step in a learning-based vision inspection method using K average clustering according to an embodiment of the present invention, and FIG. 9 is a schematic diagram illustrating a K mean clustering according to an embodiment of the present invention. FIG. 10 is a graph illustrating an example in which the intrinsic region and the false region are derived on the feature space in the learning-based vision inspection method based on the learning-based vision through the K average clustering according to an embodiment of the present invention. And a device using the inspection method.

9, the intrinsic data 130 belonging to a word and each word is used as a variable on the basis of the histogram obtained through the histogramming step described above, and the feature space The intrinsic region 130 is derived.

In addition, the histogramming step of substituting the pseudo-data 230 into the dictionary created through the preprocessing step S120 to histogramize the pseudo-data 230 belonging to each of the k words as shown in Fig. 6 is performed do.

The words and the pseudo-data 130 belonging to the respective words are shown on the feature space 400 as variables and the feature space 400 as shown in Fig. 8 based on the histogram obtained through the pseudo- 400). ≪ / RTI >

The classification reference 410 may be expressed as a line, a plane, or a hyperplane which is a boundary between the intrinsic region 130 and the pseudo region 230 on the feature space 400, and the intrinsic region 130 and the pseudo- A support vector machine (SVM) technique may be used to determine the boundary between the feature space 400 and the feature space 400 to maximize the space between them.

In the defect determination step S200, the surface to which the polarizing layer 304 is adhered is continuously photographed on the display panel 302, which is the inspection object 300, along a predetermined path through the photographing unit, (S210) of acquiring an image 310 of the image of the position of the subject 310 and the intrinsic image 110 and the false image 210 describing the image of the subject 310 as the intrinsic data 120 and the false data The image 310 is transformed into the image 310 in the same manner as the method of transforming the transformed image 220 into the feature space 400 and the transformed image 310 is displayed on the feature space 400 in the class classification step S130 It is judged whether or not it belongs to the intrinsic region 130 or the false region 230 with the obtained classification reference 410 as a boundary and a failure discrimination step S220 according to a classification criterion judging as a genuine defect or a false defect according to the region to which the genuine region 130 or the false region 230 belongs .

In the additional learning step S300, a step of correcting the dictionary through the process of adding the image 310 to be inspected obtained in the failure determination step S200 in advance and re-reproducing k average points may be added.

Or a classification criterion correction step of correcting the classification criterion 410 by adding the test data 320 obtained through the defect judgment step S200 to the feature space 400 and reflecting the added test data 320 S320) may be performed.

In the step of correcting the dictionary, the classification criterion correction step (S320) may be performed by performing defect determination on the learned defect criterion learned through the pre-learning step (SlOO), adding new information to be acquired, Can be obtained.

It is possible to increase the accuracy of performing the failure discrimination by increasing the number of learning samples and to reduce the phenomenon in which the false defect is detected by making the boundary between the true defect and the false defect more apparent in the feature space 400. [

It will be apparent to those skilled in the art that the present invention can be embodied in other specific forms without departing from the spirit or scope of the invention as defined in the appended claims. It is obvious to them. Therefore, the above-described embodiments are to be considered as illustrative rather than restrictive, and the present invention is not limited to the above description, but may be modified within the scope of the appended claims and equivalents thereof.

100: intrinsic specimen 110: intrinsic image
120: intrinsic data 130: intrinsic region
200: False sample 210: False image
220: falsifying data 230:
300: object to be inspected 302: display panel
304: polarization layer 310: image to be imaged
320: Test data
400: Feature Space 410: Classification Criteria
510: transfer means 520: photographing unit
530: Reading section 540:
550:

Claims (5)

A learning-based vision inspection method for inspecting defects of a polarizing layer adhered to a display panel,
Wherein a first learning group is formed by the same number of false samples that are samples of intrinsic defective products and a false sample that is a sample of false defects and the first data group including the same number of the intrinsic data and the false data formulated from the first learning group The first data group is grouped into k groups, the average of feature points is grouped for each group, and K average clustering consisting of k average points is performed, Learning step of drawing the first data group on the feature space and deriving a classification criterion that is a boundary between the intrinsic area where the intrinsic data is located and the pseudo area where the pseudo data is located;
The inspection data extracted and modified from the display panel to which the polarizing layer as the inspection object is adhered is substituted into the feature space and the region where the inspection data is located is set in the feature space with the classification reference as a boundary, A defect judgment step of judging that a true defect or a false defect is judged; And
And an additional learning step of applying the test data to the pre-learning step and correcting the classification criteria in the feature space,
Wherein the variable k, which is the number of each of the group and the average point, is set such that the distance between the mean point of the intrinsic region and the mean point of the pseudo region in the feature space is maximized, and wherein the failure discrimination step and the additional learning step are repeatedly performed, Learning Vision Inspection Method through.
The method according to claim 1,
Wherein the pre-
An image collecting step of collecting a first learning image group composed of a genuine image and a false image extracted from each of the intrinsic specimens included in the first learning group and image information from the pseudo specimen;
Wherein the intrinsic data and the pseudo data are derived from n feature points respectively from the respective intrinsic images and the pseudo images included in the first learning image group to form the first data group and the first data group A dictionary creation step of grouping adjacent feature points into k groups and obtaining a feature point average for each group to create a dictionary composed of k average points; And
A classifying unit for classifying the intrinsic data and the pseudo data into the dictionary and drawing it on the feature space and determining a classification reference that is a boundary between the intrinsic region where the intrinsic data is located and the pseudo- step;
A learning - based vision inspection method using K average clustering.
3. The method of claim 2,
Wherein the pre-creation step and the class classification step comprise:
K average clustering which is performed first by substituting an arbitrary k value and is performed again with a value of k which maximizes the distance between the mean value of the intrinsic region and the mean value of the false region derived through an arbitrary k value to determine the classification criterion Learning - based vision inspection method.
The method of claim 3,
In the failure determination step,
A K average clustering is performed in which the test data, which is n feature points obtained from the image of interest to be inspected, extracted from the image to be inspected, is compared with the dictionary, and the test data is classified based on the k average points, Learning - based vision inspection method.
5. The method of claim 4,
Wherein the intrinsic image, the false image,
Based vision inspection method by means of K average clustering, which is image information of a point where the inspected surface of the inspected object moves according to a certain pattern and the formulated signal of the captured image is abruptly changed.

Images