CN105654108A - Classifying method, inspection method, and inspection apparatus - Google Patents
Classifying method, inspection method, and inspection apparatus Download PDFInfo
- Publication number
- CN105654108A CN105654108A CN201510823163.4A CN201510823163A CN105654108A CN 105654108 A CN105654108 A CN 105654108A CN 201510823163 A CN201510823163 A CN 201510823163A CN 105654108 A CN105654108 A CN 105654108A
- Authority
- CN
- China
- Prior art keywords
- evaluation
- image
- estimate
- sample image
- article
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 102
- 238000007689 inspection Methods 0.000 title claims abstract description 33
- 238000011156 evaluation Methods 0.000 claims abstract description 180
- 238000012545 processing Methods 0.000 claims description 74
- 230000008569 process Effects 0.000 claims description 37
- 230000008859 change Effects 0.000 claims description 22
- 230000002950 deficient Effects 0.000 description 35
- 230000001186 cumulative effect Effects 0.000 description 17
- 238000006243 chemical reaction Methods 0.000 description 10
- 238000010606 normalization Methods 0.000 description 10
- 230000007547 defect Effects 0.000 description 9
- 238000010586 diagram Methods 0.000 description 8
- 230000006870 function Effects 0.000 description 7
- 239000000203 mixture Substances 0.000 description 6
- 230000008901 benefit Effects 0.000 description 5
- 238000005259 measurement Methods 0.000 description 5
- 238000012795 verification Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000001174 ascending effect Effects 0.000 description 2
- 230000000052 comparative effect Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000003825 pressing Methods 0.000 description 2
- 239000004576 sand Substances 0.000 description 2
- 238000007790 scraping Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000004304 visual acuity Effects 0.000 description 2
- 238000011179 visual inspection Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000003750 conditioning effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 239000011347 resin Substances 0.000 description 1
- 229920005989 resin Polymers 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a classifying method, an inspection method and an inspection apparatus. The present invention provides a classifying method of classifying an article into one of a plurality of groups based on an image of the article, comprising determining an evaluation method for obtaining an evaluation value of an image by using at least some of sample images, obtaining evaluation values for the sample images by the determined evaluation method, changing the evaluation method so as to increase a degree of dissimilarity in an evaluation value range for sample images between the plurality of groups by changing a evaluation value of at least one sample image having a singular evaluation value among the sample images, obtaining an evaluation value for the image of the article using the changed evaluation method, and classifying the article into one of the plurality of groups based on the evaluation value for the image of the article.
Description
Technical field
The present invention relates to by taxonomy of goods to one of multiple groups sorting technique, inspection method and inspection units.
Background technology
As the device of visual inspection or internal examination for performing article, it is provided for use by catching, with image capturing unit, the inspection units that the image (target image) that article obtain performs to check. Inspection units performs the image (study image) with the use of the multiple samples in being categorized into one of multiple groups respectively, determines obtaining so-called " study " of the evaluation method of the evaluation of estimate of image. The evaluation of estimate of the image of article is obtained based on the evaluation method determined, and, during article are classified into one of multiple groups based on the evaluation of estimate obtained. Therefore, inspection units determines evaluation method by study, to improve the performance of taxonomy of goods.
Patent documentation 1 (Japanese Patent Publication No.2010-102690) proposes to determine the method for the combination of the characteristic quantity that evaluation method to be used with the use of multiple study image. In addition, patent documentation 2 (Japanese Patent Publication No.2010-157154) proposes one method like this: in the method, specify by device and users classification to the sample in different classifications, and, prompting user changes evaluation method, so that by this device and this user by this sample classification in same classification.
Such as, in the method described by patent documentation 1, if multiple study image comprises the image with little defect or low contrast defect etc., so evaluation method can not be determined with by the sample classification relevant to image in the former group that should be categorized into of this sample. In this case, patent documentation 1 does not describe and changes evaluation method to improve the technology of the performance of taxonomy of goods.
In the method described by patent documentation 2, owing to being designated to the sample in different classifications by device and users classification, therefore, whenever evaluation method changes, it is necessary to make device by all sample classifications. This can make to determine complicated for the process (learning process) of the evaluation method of taxonomy of goods. In addition, in the method described by patent documentation 2, prompting user based on each in device and user by sample classification to classification characteristic quantity distribution change evaluation method use characteristic quantity. This needs user to have the technical ability changing evaluation method. Along with the quantity of selective image feature increases, the determination of user becomes more difficult, thus needs the plenty of time to perform to determine.
Summary of the invention
The present invention such as provides the complicacy being beneficial to reduction learning process and improves the technology of the performance of taxonomy of goods.
According to an aspect of the present invention, there is provided a kind of image based on article by taxonomy of goods to the sorting technique in one of multiple groups, the method comprises: with the use of at least some in sample image, determining obtaining the evaluation method of evaluation of estimate of image, sample image refers to respectively to control oneself the image of the multiple samples in being classified into one of multiple groups; The evaluation of estimate of sample image is obtained by the evaluation method determined; By changing the evaluation of estimate of at least one sample image with unusual (singular) evaluation of estimate among sample image, change evaluation method, with the different degree of the evaluation of estimate scope of sample image increased between multiple groups; And with the use of the evaluation method changed, obtain the evaluation of estimate of image of article, and the evaluation of estimate of the image based on article, by taxonomy of goods in one of multiple groups.
According to an aspect of the present invention, it is provided that the inspection method of a kind of inspection performing article, the method comprises: the image obtaining article by catching article; And use sorting technique, by taxonomy of goods in one of multiple groups, wherein, sorting technique is the image based on article by taxonomy of goods to the method in one of multiple groups, and comprise: with the use of at least some in sample image, determining obtaining the evaluation method of evaluation of estimate of image, sample image refers to respectively to control oneself the image of the multiple samples in being classified into one of multiple groups; The evaluation of estimate of sample image is obtained by the evaluation method determined; By changing the evaluation of estimate of at least one sample image with unusual evaluation of estimate among sample image, change evaluation method, with the different degree of the evaluation of estimate scope of sample image increased between multiple groups; And with the use of the evaluation method changed, obtain the evaluation of estimate of image of article, and the evaluation of estimate of the image based on article, by taxonomy of goods in one of multiple groups.
According to an aspect of the present invention, it is provided that a kind of for performing the inspection units that article check, this inspection units comprises: be configured to the image capturing unit obtaining the image of article by catching article; And it is configured to the image based on article by taxonomy of goods to the processing unit in one of multiple groups, wherein, processing unit performs following operation: with the use of at least some in sample image, determining obtaining the evaluation method of evaluation of estimate of image, sample image refers to respectively to control oneself the image of the multiple samples in being classified into one of multiple groups; The evaluation of estimate of sample image is obtained by the evaluation method determined; By changing the evaluation of estimate of at least one sample image with unusual evaluation of estimate among sample image, change evaluation method, with the different degree of the evaluation of estimate scope of sample image increased between multiple groups; And with the use of the evaluation method changed, obtain the evaluation of estimate of image of article, and the evaluation of estimate of the image based on article, by taxonomy of goods in one of multiple groups.
Read the following description of exemplary embodiment with reference to accompanying drawing, the further feature of the present invention will become clear.
Accompanying drawing explanation
Fig. 1 is the schematic diagram representing the inspection units according to the first embodiment;
Fig. 2 is the schema of the method for the classification process illustrating in processing unit;
Fig. 3 is the schema of the method (learning method) illustrating the acquisition classified information according to the first embodiment;
Fig. 4 illustrates about each in multiple study image, extracts the table of the result of the characteristic quantity of each characteristics of image being contained in feature list;
Fig. 5 is the diagram of the show example illustrating on the picture of display unit;
Fig. 6 is the schema of the method (learning method) illustrating the acquisition classified information according to the 2nd embodiment;
Fig. 7 is the diagram illustrating the cumulative amount distribution shown on the display unit;
Fig. 8 is the diagram illustrating the comparative example that cumulative amount distributes; And
The diagram that Fig. 9 is the modification of cumulative amount distribution illustrates.
Embodiment
The exemplary embodiment of the present invention is described referring to accompanying drawing. Noting, identical Reference numeral represents identical parts in all the drawings, and, do not provide the description that it repeats.
<the first embodiment>
With reference to Fig. 1, the inspection units 1 according to the first embodiment of the present invention is described. Fig. 1 is the schematic diagram illustrating the inspection units 1 according to the first embodiment. Such as, inspection units 1 performs the visual inspection of the article 2 of the such as metal parts or resin component that are used in Industrial products. Such as, the surface of article 2 may occur the defect of such as scraping, uneven (uneven color) or convex-concave (roughness). The defect that inspection units 1 occurs on the surface of article 2 based on the image detection of article 2, and, during article 2 are categorized into one of multiple groups. First embodiment will describe such example: wherein, multiple groups comprise no defective product group (first group) and defectiveness product group (the 2nd group), and article 2 are categorized in one of no defective product group and defectiveness product group by inspection units 1. Although description inspection units 1 is checked the example of the outward appearance (surfaces of article 2) of article 2 by the first embodiment, but the present embodiment is also applicable to check the situation of the inside of article 2 with the use of X-ray etc.
Inspection units 1 can comprise image capturing unit 11, processing unit 12, display unit 13 and input unit 14. Image capturing unit 11 such as comprises lighting unit and photographic camera, and, the image (target image) of article 2 is obtained by catching article 2. The image of the article 2 obtained by image capturing unit 11 is sent to processing unit 12. Processing unit 12 can be realized by the signal conditioning package comprising CPU12a (central processing unit), RAM12b (random access memory) and HDD12c (hard disk drive). Processing unit 12 performs following process (classification process): obtain the evaluation of estimate of target image obtained by image capturing unit 11, and in article 2 being categorized into one of multiple groups based on the evaluation of estimate scope (threshold value) in the evaluation of estimate obtained and each group. CPU12a perform to be categorized into article 2 one of multiple groups in program, and, RAM12b and HDD12c stores program and data. Display unit 13 comprises such as watch-dog, and, the result of the classification process that display is performed by processing unit 12. Input unit 14 comprises such as keyboard and mouse, and, send the instruction from user to processing unit 12.
[the classification process in processing unit 12]
Hereinafter with reference to Fig. 2, the process of the classification in processing unit 12 is described. Fig. 2 is the schema of the method for the classification process illustrated in processing unit 12. In step sl, the condition of the image capturing unit 11 when article 2 are caught in processing unit 12 setting, to obtain image visual for the defect on the surface of article 2. Described condition can comprise time shutter of such as throw light on angle and photographic camera, focusing and aperture. In step s 2, processing unit 12 obtains the image (study image, sample image) of the multiple samples in being classified into one of multiple groups of respectively controlling oneself.Multiple study image can be the image caught by image capturing unit 11 and preserved in advance. If there is no the image preserved, so newly obtain image by making image capturing unit 11 catch multiple sample. Each in multiple sample is such as classified in one of multiple groups (no defective product group and defectiveness product groups) based on its study image by user. In the first embodiment, the image of the sample being categorized in one of two groups (no defective product group and defectiveness product group) is used as study image. But, the present invention is not limited thereto. Such as, as study image, it is possible to use according to the image of the sample that the type of defectiveness product (scraping or uneven etc.) more carefully is classified.
In step s3, processing unit 12 performs so-called " study " that obtains the information (following, to be called classified information) for being classified by target image with the use of at least some in the multiple study images obtained in step s 2. Classified information can comprise the evaluation of estimate for obtaining image evaluation method and for the threshold value by the evaluation of estimate of taxonomy of goods. Evaluation method is such as the function of the evaluation of estimate for obtaining image, and, by least one characteristic quantity in image being updated to the evaluation of estimate using at least one characteristic quantity to obtain image in the evaluation method of parameter. The size of the feature (following, to be called characteristics of image) in characteristic quantity indicator card picture. In step s3, multiple characteristics of image that processing unit 12 is estimated as is used to each study image is categorized in one of no defective product group and defectiveness product group by automatically extracting from many characteristics of image, create feature list. The characteristic quantity of the multiple characteristics of image being contained in feature list is used to be confirmed as a part for classified information as the evaluation method of parameter. The method obtaining classified information will be described in detail later. In step s 4 which, processing unit 12 makes image capturing unit 11 catch article 2, further, based on the classified information obtained in step s3, the image (target image) of article 2 obtained is categorized in one of no defective product group and defectiveness product group. Such as, processing unit 12 obtains the evaluation of estimate of target image with the use of the evaluation method determined in step s3, and, compare evaluation of estimate and threshold value to determine the group that article are classified into. In step s 5, processing unit 12 shows check result on display unit 13. Except determining that article 2 are no defective product or defectiveness product, the characteristic quantity of the image of the defectiveness part that processing unit 12 also can be displayed in article 2 to occur on display unit 13, each characteristics of image being contained in feature list and evaluation of estimate etc. are as check result.
[acquisition of classified information]
The acquisition (study) of the classified information in the process in the step S3 of the schema shown in Fig. 2 is described with reference to Fig. 3. Fig. 3 is the schema illustrating the method (learning method) obtaining classified information.
In step S3-11, processing unit 12 loads the multiple study images obtained in step s 2. In step S3-12, processing unit 12 creates feature list with the use of at least some in multiple study image, and, it is determined that use the evaluation method of characteristic quantity as parameter of the multiple characteristics of image being contained in feature list. Hereinafter, use n study creation of image feature list will be explained and uses geneva (Mahalanobis) distance as the example of evaluation method. Such as, in order to emphasize in multiple study image each in defect, processing unit 12 to each study image perform the little wave conversion of Haar, the little wave conversion of Haar is as the one in the little wave conversion of the transform method transforming to frequency domain.The little wave conversion of Haar is the process that can perform frequency transformation keeping positional information while. Each execution inner product in multiple study image is calculated by processing unit 12 with the use of four kinds of wave filters of the provide by formula (1) first to the 4th wave filter. In formula (1), first wave filter is the wave filter of the high frequency composition for extracting in vertical direction, 2nd wave filter is for extracting the wave filter to the high frequency composition on angular direction, 3rd wave filter is the wave filter of the high frequency composition for extracting in horizontal direction, and the 4th wave filter is the wave filter for extracting low-frequency component.
This allows processing unit 12 to obtain four kinds of images: the image that the high frequency composition in the image that the high frequency composition in vertical direction has been extracted, the image being extracted by the high frequency composition on angular direction, horizontal direction has been extracted and the image that low-frequency component has been extracted. Compared with the image before conversion, each in the four kinds of images obtained like this has 1/2 resolving power. The image that processing unit 12 repeats to be extracted by low-frequency component performs the little wave conversion of Haar and obtains the process of four kinds of images with 1/2 resolving power, thus obtains multiple images that frequency level ground (hierarchically) reduces.
Processing unit 12 is from each the global image feature extracting the such as maximum value of all pixel values, mean value, variance value, kurtosis (kurtosis), the degree of bias (skewness) and geometric mean etc. in the image before the image each layer obtained by the little wave conversion of Haar and conversion. Processing unit 12 can extract the statistical value of the difference and standard deviation and so between such as contrast gradient, maxima and minima as global image feature. By this process, processing unit 12 can from multiple many characteristics of image of study image zooming-out. In the present embodiment, by using the little wave conversion of Haar to obtain many characteristics of image. But, such as, by using another transform method of such as another little wave conversion, edge extraction, Fourier transform or Gabor transformation, obtain many characteristics of image. Except global image feature, many characteristics of image also can comprise the topography's feature calculated by filtering process.
Processing unit 12 calculates the score of each in the characteristics of image extracted with the use of the study image in no defective product group, and, from the characteristics of image that the image feature selection of many extractions processes for classifying, thus create feature list. As the method selecting characteristics of image, such as, as described in patent documentation 1, it is provided that with the use of the method for compatibility of the combination of the study picture appraisal characteristics of image in no defective product group. In the present embodiment, the characteristics of image for process of classifying is selected with the use of the method. But, such as, it is possible to use the other method of such as principal component analytical method. A kind of decomposition by proper value (eigenvalue) principal component analytical method select the characteristics of image with high intrinsic value to avoid the method for the redundancy in multiple characteristics of image (redundancy). Use the method can prevent the characteristics of image selecting redundancy. Although more than explaining the example selecting characteristics of image with the use of the study image in no defective product group, but the present invention is not limited thereto. Such as, by using the study image in defectiveness product group or use the study image in two groups to select characteristics of image.
The method of the weight of the characteristic quantity of each characteristics of image determining to be contained in the feature list of establishment is described with reference to Fig. 4.Fig. 4 is the table of the result illustrating the characteristic quantity extracting each characteristics of image being contained in feature list about each in multiple study image. With reference to Fig. 4, each characteristic quantity in each in multiple study image is by XijRepresent, here, i represent study image numbering (i=1,2 ..., n), j represent the feature being contained in feature list numbering (j=1,2 ..., k). Noting, n represents the quantity of study image, and k represents the quantity of the characteristics of image being contained in feature list. Relation between n and k can meet n >=k. It should also be noted that MjRepresent the characteristic quantity X in multiple study imageijMean value, ��jRepresent the characteristic quantity X in multiple study imageijStandard deviation. For each characteristics of image of the characteristics of image of the process that is not selected as being used for classifying, characteristic quantity, mean value and standard deviation can be obtained. This is because, these results use in follow-up process (step S3-16).
Processing unit 12 passes through formula (2) by each characteristic quantity X in each in multiple study imageijNormalization method. In formula (2), YijRepresent each normalized characteristic quantity. Processing unit 12 obtains relative coefficient r by formula (3)pq, and, obtain formula (4) provide by each relative coefficient r11��rkkThe inverse matrix A of the correlation matrix R formed. Inverse matrix A is corresponding with the weight of the characteristic quantity of each characteristics of image being contained in feature list. This makes processing unit 12 can determine the mahalanobis distance MD that formula (5) representsiAs evaluation method, mahalanobis distance MDiUse the characteristic quantity (normalization characteristic amount) of each characteristics of image being contained in feature list as parameter.
Referring again to the schema shown in Fig. 3, in step S3-13, processing unit 12 is with the use of evaluation method (the mahalanobis distance MD determined in step S3-12i) obtain the evaluation of estimate of each in multiple study image. Processing unit 12 extracts multiple characteristic quantity according to feature list from each multiple study image, and, multiple characteristic quantity is updated in evaluation method, thus obtains the evaluation of estimate of each study image. In the first embodiment, the example using intensity of anomaly as the evaluation of estimate of each study image will be described. Although in the present embodiment, by mahalanobis distance MDiObtain intensity of anomaly (evaluation of estimate), but the projection distance also by Ou Shi (Euclidean) distance or as a kind of subspace method obtains it.
In step S3-14, processing unit 12 is the distribution of the intensity of anomaly (evaluation of estimate) that each group produces multiple study image, and, display unit 13 shows it. In step S3-15, processing unit 12 determines whether the different degree of the scope (evaluation of estimate scope) of the evaluation of estimate of the sample image between multiple groups meets permissible value. If different degree does not meet permissible value, so process proceeds to step S3-16; Otherwise, obtain (study) classified information and terminate. Such as, permissible value can by user preset.
Fig. 5 is the diagram of the show example illustrated on the picture of display unit 13. In the region 13a of display unit 13, show the distribution (histogram) of the intensity of anomaly (evaluation of estimate) of each study image for each group. In the histogram, informal voucher represents the quantity of the study image in no defective product group, and secret note represents the quantity of the study image in defectiveness product group. In the region 13b of display unit 13, show for determining the verification and measurement ratio of the defectiveness product of the threshold value of evaluation of estimate classified by target image and as the orthogonal rate (orthogonality) of different degree.
The ratio that the sample that the verification and measurement ratio of defectiveness product refers in should be classified in multiple groups predetermined group is classified in this predetermined group, and, refer to that such as defectiveness product is classified as the ratio of defectiveness product. Can according to the verification and measurement ratio definite threshold of defectiveness product. The verification and measurement ratio of defectiveness product can be set arbitrarily by user, but is generally set to 100%, to be not no defective product by defectiveness product classification. If the verification and measurement ratio of defectiveness product is 100%, so processing unit 12 sets the threshold to the value of the minimum value of the intensity of anomaly of the study image being less than in defectiveness product group. That is, processing unit 12 sets threshold value so that all study images in defectiveness product group are arranged in the right side of threshold value in the histogram shown in Fig. 5. In the example as shown in fig. 5, threshold value 13c is set smaller than the intensity of anomaly of study image 13h minimum in defectiveness product group by processing unit 12.
Refer to have separately the ratio of all study images in the study image of the intensity of anomaly being less than threshold value and no defective product group as the orthogonal rate of different degree. Generally, orthogonal rate can be high. When orthogonal rate is 100%, obtaining perfect condition, wherein, all study graphical arrangement in no defective product group are in the left side of threshold value, and all study graphical arrangement in defectiveness product group are in the right side of threshold value.
Although performing to determine whether different degree meets the process of permissible value by processing unit 12 in the present embodiment, but such as can performing it by user. In this case, if user determines different degree, (orthogonal rate) does not meet permissible value, and so he/her presses " adding study " button 13d by input unit 14; Otherwise, he/her presses " terminating study " button 13e by input unit 14. If user's pressing " adds and learn " button 13d, so processing unit 12 proceeds to step S3-16. If user's pressing " terminates to learn " button 13e, so processing unit 12 terminates the acquisition (study) of classified information.
In step S3-16, processing unit 12, based on the information of the intensity of anomaly (evaluation of estimate) representing multiple study image for each group, specifies at least one sample with the unusual intensity of anomaly among the intensity of anomaly of multiple study image. In the first embodiment, as described information, it may also be useful to the histogram of the intensity of anomaly in each group. The samples selection that processing unit 12 can be subordinated to the scope of the evaluation of estimate of the image of the sample of the no defective product group part overlapping with the scope of the evaluation of estimate of the image of the sample of defectiveness product group (following, to be called overlapping part) has at least one sample of unusual evaluation of estimate. Such as, overlapping part refers to the scope 13f of the intensity of anomaly in the histogram of display in the region 13a of display unit 13. Such as, the intensity of anomaly among sample that processing unit 12 appointment is contained in defectiveness product group and overlapping part is at least one sample of ascending order. As an alternative, processing unit 12 specifies the intensity of anomaly among the sample being contained in no defective product group and overlapping part to be at least one sample falling sequence. Such as, the quantity of the sample specified can by user preset. Processing unit 12 can show the study image specifying sample in the region 13g of display unit 13. Although specifying at least one sample with unusual evaluation of estimate in the present embodiment by processing unit 12, but such as can performing this by user and operate. In this case, user, by selecting the study image with unusual evaluation of estimate by input unit 14 in the histogram in the region 13a being shown in display unit 13, specifies at least one sample.
By the feature list created in step S3-12, such as, evaluation method may not be determined to be categorized in defectiveness product group the image with the such as defect of little defect or low contrast defect. In order to tackle this point, in the process of step S3-13��S3-15, it is determined that whether different degree meets permissible value. In the process of step S3-16, specify the sample with unusual evaluation of estimate from the distribution of the intensity of anomaly (evaluation of estimate) of multiple study image.
In step S3-17, processing unit 12 changes evaluation method by adding characteristics of image to feature list, with the evaluation of estimate of the study image of at least one sample specified in step S3-16 by changing, increase different degree (orthogonal rate) (to meet permissible value). Now, processing unit 12 can change evaluation method so that the change of the evaluation of estimate of the study image of at least one sample becomes to be greater than the mean value of the change of the evaluation of estimate of multiple study image.
Such as, processing unit 12 obtains, about the study image of at least one sample specified in step S3-16, the characteristic quantity X of characteristics of image not being contained in feature lists. Note, s represent the characteristics of image not being contained in feature list numbering (s=k+1, k+2 ..., N), N represents in step S3-12 the sum from the characteristics of image of multiple study image zooming-out. Processing unit 12 passes through formula (6) by characteristic quantity XsNormalization method. Normalization characteristic amount YsProvide by following formula:
Here, MsRepresent the characteristic quantity X in multiple study imagesMean value, ��sRepresent the characteristic quantity X in multiple study imagesStandard deviation. The multiple study images loaded by being used in step S3-11, calculating mean value MsAnd standard deviations��
With the use of mean value MsAnd standard deviationsNormalization method, obtain normalization characteristic amount Ys. Accordingly, it may be possible to by the normalization characteristic amount Y of each compared in multiple study imagesWith the normalization characteristic amount of the study image of at least one sample, the study image at least one sample specified in step S3-16 is selected to have the characteristics of image of high contribution rate. That is, as the characteristic quantity Y of the characteristics of image adding feature list tosTime bigger, when evaluation method changes, compared with the mean value of the change of the evaluation of estimate of each study image, it is possible to change the evaluation of estimate of the study image of at least one sample specified in step S3-16 larger. Therefore, processing unit 12 by adding the normalization characteristic amount Y among the characteristics of image not being contained in feature list to feature listsFor falling the characteristics of image of sequence, change evaluation method. Add the normalization characteristic amount Y of the characteristics of image in feature list tosCan be 3 or bigger. By above-mentioned method, processing unit 12 can redefine the weight of the characteristic quantity of each characteristics of image being contained in feature list. In addition, processing unit 12 can get rid of characteristics of image among the multiple characteristics of image being contained in feature list, that the change of the evaluation of estimate of the study image of at least one sample has minimum contribution rate (weight) from feature list so that performing classification process institute's time spent drops in tolerable limit.
In step S3-18, similar with the process of step S3-13, processing unit 12 obtains the evaluation of estimate of each in multiple study image with the use of the evaluation method changed in step S3-17. In step S3-19, similar with the process of step S3-14, processing unit 12 regenerates the distribution of the intensity of anomaly (evaluation of estimate) of multiple study images of each group, and, display unit 13 shows it.In step S3-20, processing unit 12 determines whether different degree (orthogonal rate) meets permissible value. If different degree (orthogonal rate) does not meet permissible value, so process returns step S3-16, with the process of repeating step S3-16��S3-19; Otherwise, the acquisition (study) of classified information terminates.
As mentioned above, it is necessary, in the inspection units 1 according to the first embodiment, processing unit 12 changes evaluation method to increase different degree (to meet permissible value). This allows inspection units 1 to carry out the study of high precision, and by obtaining the evaluation of estimate of target image by the evaluation method changed, during article are accurately categorized into one of multiple groups.
<the 2nd embodiment>
Inspection units according to the second embodiment of the present invention will be described. In the inspection units according to the 2nd embodiment, the method (learning method) of the acquisition classified information performed in the process of step S3 in fig. 2 is different from the method in the inspection units 1 according to the first embodiment. Hereinafter, with reference to Fig. 6, the acquisition (study) according to the classified information in the inspection units of the 2nd embodiment is described.
Step S3-11 in schema shown in step S3-21��S3-29 and Fig. 3��S3-19 is identical. In step S3-30, according to by the condition of user preset, processing unit 12 determines whether the process of repeating step S3-26��S3-29. The process of step S3-26��S3-29 refers to change evaluation method to increase the process of different degree (to meet permissible value). Such as, the process of step S3-26��S3-29 is repeated the number of times preset by processing unit 12. By the process of repeating step S3-26��S3-29, processing unit 12 can obtain multiple candidates of the evaluation method of the evaluation of estimate for obtaining target image. In step S3-31, processing unit 12 selects the candidate with the highest different degree from the multiple candidate obtained among step S3-30, and, it is determined that the candidate of selection is as the evaluation method of the evaluation of estimate for obtaining target image. In the present embodiment, processing unit 12 selects the candidate with the highest different degree from multiple candidate. But, such as, processing unit 12 can be selected that execution classification is processed and taken time the shortest candidate or meet different degree simultaneously and perform the candidate of the condition that classification process is taken time.
<the 3rd embodiment>
In the first embodiment, when specifying at least one sample with unusual intensity of anomaly, having been explained by following example: in this example embodiment, the histogram of the intensity of anomaly in each group is used as representing the information of the intensity of anomaly of each in multiple study image for each group. In a second embodiment, explained by following example: the relation between the quantity of the study image of the defectiveness product group of the intensity of anomaly that each study image of the no defective product group that the use of this example is sorted according to intensity of anomaly (evaluation of estimate) is corresponding with the intensity of anomaly with each study image with no defective product group is as described information. Hereinafter, this relation is called as " cumulative amount distribution ".
Fig. 7 is the diagram illustrating that cumulative amount distributes. Cumulative amount distribution can be produced by processing unit 12 in step S3-14 and show in the region 13a of display unit 13. Transverse axis in Fig. 7 illustrates and is caught to number with the ID of each study image of the no defective product group of the ascending sort of intensity of anomaly, and the longitudinal axis shows the quantity (cumulative amount) of the study image of the defectiveness product group of intensity of anomaly corresponding to the intensity of anomaly with each study image with no defective product group.
Such as, in the figure 7, when ID numbering (transverse axis) of the study image of no defective product group is " 30 ", the cumulative amount (longitudinal axis) of the study image of defectiveness product group is increased to 1. This illustrates, there is a study image of defectiveness group corresponding to the intensity of anomaly of the 30th study image of intensity of anomaly and no defective product group. Specifically, this illustrates between the intensity of anomaly of the 30th the study image that the intensity of anomaly of one in defectiveness product group study image is in no defective product group and the intensity of anomaly of the 31st study image.
Similarly, in the figure 7, when ID numbering (transverse axis) of the study image of no defective product group is " 40 ", the cumulative amount (longitudinal axis) of the study image of defectiveness product group is increased to 2. This illustrates, there is a study image of defectiveness group corresponding to the intensity of anomaly of the 40th study image of intensity of anomaly and no defective product group. Specifically, this illustrates between the intensity of anomaly of the 40th the study image that the intensity of anomaly of one in defectiveness product group study image is in no defective product group and the intensity of anomaly of the 41st study image.
Below, by for using the advantage of the distribution of the cumulative amount shown in Fig. 7 to explain. Cumulative amount distribution is used to mainly contain three advantages. First advantage is, graphics shape is uniquely identified. Such as, if histogram is used as representing the information of the intensity of anomaly of each group, so when not having to can not determine graphics shape (bin) between setting district. But, if the cumulative amount distribution shown in Fig. 7 is used as described information, so can uniquely determine graphics shape when not waiting between setting district.
Second advantage is, it is possible to the study image that in defectiveness group, intensity of anomaly is minimum easily detected. Usually, in check system, it is possible to how soon and how accurately identify that the defectiveness product closest to no defective product is the big problem of when being classified by image accurately. If use histogram, so in two histogrammic situations of not reference no defective product group and defectiveness product group, the study image that in defectiveness product group, intensity of anomaly is minimum cannot be detected. But, in the cumulative amount shown in Fig. 7 distributes, by only with reference to the data with curve performance, the study image that in defectiveness product group, intensity of anomaly is minimum just can easily and accurately be detected.
3rd advantage is, it is possible to easily determine that whether the study result according to graphics shape is effective. In other words, in cumulative amount distributes, it is possible to easily grasp orthogonal rate (different degree) from graphics shape (slope of curve). Fig. 8 illustrates the comparative example (three examples) that cumulative amount distributes. Solid line 71 in Fig. 8 illustrates such a case, that is, the maximum intensity of anomaly of the study image that the intensity of anomaly of all study images in defectiveness product group is greater than in no defective product group and no defective product group and defectiveness product group are separated completely. In addition, the dotted line 72 in Fig. 8 illustrates such a case, although that is, there is the intensity of anomaly study image less than the maximum intensity of anomaly in no defective product group, but no defective product group and defectiveness product group are fully separated. In addition, the long and short dash line 73 in Fig. 8 illustrates such a case, that is, among the study image in defectiveness product group, there is the study image that intensity of anomaly is low especially compared with the maximum intensity of anomaly in no defective product group, and, study is not enough.Such as, about formation such as by the reason of the graphics shape shown in long and short dash line 73, it is possible to enumerate some reasons like this: the former study image should being categorized in no defective product group is classified in defectiveness product group; The characteristics of image of classification needed for no defective product group is not extracted; Etc..
Here, in cumulative amount distributes, obtain orthogonal rate (different degree) from graphics shape (slope of curve), and, in step S3-15, determine whether the orthogonal rate obtained meets permissible value. In addition, cumulative amount distribution is not limited to the example shown in Fig. 7, and such as, as shown in Figure 9, transverse axis and the longitudinal axis can be contrary with Fig. 7.
<other embodiment>
Also by reading and perform to be recorded in the computer executable instructions on storage media (also can being more completely called " non-transitory computer-readable storage media ") (such as, one or more program) with one or more the function that performs in above-described embodiment and/or comprise one or more the function for performing in above-described embodiment one or more circuit (such as, application specific integrated circuit (ASIC)) system or the computer of device, or, by by such as reading by the computer of system or device and being performed the computer executable instructions from storage media and with one or more the function that performs in above-described embodiment and/or control the method that one or more circuit performs with one or more the function performed in above-described embodiment, realize embodiments of the invention. computer can comprise one or more treater (such as, central processing unit (CPU), micro-processing unit (MPU)), and independent computer or the network of independent treater can be comprised, to read and perform computer executable instructions. computer executable instructions such as can be provided to computer from network or storage media. storage media can comprise storer, CD (such as compact disk (CD), digital versatile disc (DVD) or the Blu-ray disc (BD) of such as hard disk, random access memory (RAM), read-only storage (ROM), distributed computing systemTM), one or more in flash memory equipment and storage card etc.
Other embodiment
Embodiments of the invention can also be realized by following method, namely, by network or various storage media, the software (program) performing the function of above-described embodiment being supplied to system or device, the computer of this system or device or central processing unit (CPU), micro-processing unit (MPU) read and the method for steering routine.
Although describing the present invention with reference to exemplary embodiment, it should be appreciated that the invention is not restricted to disclosed exemplary embodiment. The scope of claims should be endowed the widest explaining to comprise all such change modes and equivalent structure and function.
Claims (14)
1. a sorting technique, based on the image of article by taxonomy of goods in one of multiple groups, it is characterised in that, this sorting technique comprises:
Determine obtaining the evaluation method of evaluation of estimate of image with the use of at least some in sample image, described sample image refers to respectively to control oneself the image of the multiple samples in being classified into one of multiple groups;
The evaluation of estimate of sample image is obtained by the evaluation method determined;
By changing the evaluation of estimate of at least one sample image with unusual evaluation of estimate among sample image, change evaluation method, with the different degree of the evaluation of estimate scope of sample image increased between described multiple groups;And
With the use of the evaluation method changed, obtain the evaluation of estimate of image of article, and the evaluation of estimate of the image based on article, by taxonomy of goods in one of multiple groups.
2. method according to claim 1, wherein, when changing evaluation method, the change that evaluation method is changed to the evaluation of estimate making at least one sample image described is greater than the mean value of the change of the evaluation of estimate of sample image.
3. method according to claim 1, wherein,
Described multiple groups comprise first group and the 2nd group, and,
At least one sample image described is selected from and belongs to the sample image of lower part: in the portion, and the evaluation of estimate scope of the sample image of first group is overlapping with the evaluation of estimate scope of the sample image of the 2nd group.
4. method according to claim 1, wherein,
When changing evaluation method,
Make described degree meet the process of permissible value by repeating to change evaluation method, obtain multiple candidates of evaluation method, and
The candidate that described degree among described multiple candidate is the highest is confirmed as the evaluation method of the evaluation of estimate of the image for obtaining article.
5. method according to claim 4, wherein, when changing evaluation method, the process that change evaluation method makes described degree meet permissible value repeats the number of times preset.
6. method according to claim 1, wherein, when obtaining the evaluation of estimate of sample image, at least one characteristic quantity is extracted as the parameter to be used by evaluation method from each sample image, and, by being updated in evaluation method by least one characteristic quantity described, obtain the evaluation of estimate of each in sample image.
7. method according to claim 1, wherein, when changing evaluation method, the evaluation of estimate of sample image is obtained by the evaluation method changed, further, the evaluation of estimate of the sample image obtained based on the evaluation method by changing, it is determined that for the threshold value by the evaluation of estimate of taxonomy of goods.
8. method according to claim 7, wherein, when changing evaluation method, according to the ratio that the sample should being classified in predetermined group of among described multiple groups is classified in this predetermined group, it is determined that described threshold value.
9. method according to claim 1, wherein, change group comprises the information that show needle represents the evaluation of estimate of each in sample image to each group.
10. method according to claim 9, wherein, described packet is containing the histogram of the evaluation of estimate in each group.
11. methods according to claim 9, wherein,
Multiple groups comprise first group and the 2nd group, and
Described packet is containing according to the relation between the sample image of first group of evaluation of estimate sequence and the quantity of sample image having in the 2nd group of evaluation of estimate corresponding to the evaluation of estimate of the sample image with first group.
12. methods according to claim 9, wherein, change group comprises the information based on the display when showing and specifies at least one sample image described.
13. 1 kinds perform the inspection method of the inspection of article, comprising:
The image of article is obtained by catching article; And
With the use of sorting technique, by taxonomy of goods in one of multiple groups,
Wherein, described sorting technique is the image based on article by taxonomy of goods to the method in one of multiple groups, and comprises:
Determine obtaining the evaluation method of evaluation of estimate of image with the use of at least some in sample image, described sample image refers to respectively to control oneself the image of the multiple samples in being classified into one of multiple groups;
The evaluation of estimate of sample image is obtained by the evaluation method determined;
By changing the evaluation of estimate of at least one sample image with unusual evaluation of estimate among sample image, change evaluation method, with the different degree of the evaluation of estimate scope of sample image increased between multiple groups; And
With the use of the evaluation method changed, obtain the evaluation of estimate of image of article, and the evaluation of estimate of the image based on article, by taxonomy of goods in one of multiple groups.
14. 1 kinds, for performing the inspection units of the inspection of article, comprising:
It is configured to the image capturing unit obtaining the image of article by catching article; And
It is configured to the image based on article by taxonomy of goods to the processing unit in one of multiple groups,
Wherein, described processing unit performs following operation:
Determine obtaining the evaluation method of evaluation of estimate of image with the use of at least some in sample image, described sample image refers to respectively to control oneself the image of the multiple samples in being classified into one of multiple groups;
The evaluation of estimate of sample image is obtained by the evaluation method determined;
By changing the evaluation of estimate of at least one sample image with unusual evaluation of estimate among sample image, change evaluation method, with the different degree of the evaluation of estimate scope of sample image increased between multiple groups; And
With the use of the evaluation method changed, obtain the evaluation of estimate of image of article, and the evaluation of estimate of the image based on article, by taxonomy of goods in one of multiple groups.
Applications Claiming Priority (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2014242306 | 2014-11-28 | ||
| JP2014-242306 | 2014-11-28 | ||
| JP2015199723A JP6616645B2 (en) | 2014-11-28 | 2015-10-07 | Classification method, inspection method, inspection apparatus, and program |
| JP2015-199723 | 2015-10-07 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105654108A true CN105654108A (en) | 2016-06-08 |
| CN105654108B CN105654108B (en) | 2019-04-12 |
Family
ID=56124428
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510823163.4A Expired - Fee Related CN105654108B (en) | 2014-11-28 | 2015-11-24 | Classification method, inspection method and check device |
Country Status (2)
| Country | Link |
|---|---|
| JP (1) | JP6616645B2 (en) |
| CN (1) | CN105654108B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109425622A (en) * | 2017-08-23 | 2019-03-05 | 富士通株式会社 | Image testing device, image checking method and image inspection program |
| CN112287748A (en) * | 2019-07-24 | 2021-01-29 | 开必拓数据股份有限公司 | Mobile article classification system and method based on self-learning technology |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP6811645B2 (en) * | 2017-02-28 | 2021-01-13 | 株式会社日立製作所 | Image search device and image search method |
| CN113792201A (en) * | 2021-03-02 | 2021-12-14 | 北京沃东天骏信息技术有限公司 | Method and device for pushing information |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1828632A (en) * | 2005-02-28 | 2006-09-06 | 株式会社东芝 | Object detection apparatus, learning apparatus, object detection system, object detection method |
| US20080273761A1 (en) * | 2004-06-07 | 2008-11-06 | Kozo Kawata | Image Recognition Device, Image Recognition Method, and Program for Causing Computer to Execute the Method |
| US20090222753A1 (en) * | 2008-03-03 | 2009-09-03 | Kohei Yamaguchi | Defect inspection tool and method of parameter tuning for defect inspection tool |
| US20100086176A1 (en) * | 2008-10-03 | 2010-04-08 | Jun Yokono | Learning Apparatus and Method, Recognition Apparatus and Method, Program, and Recording Medium |
| CN101794392A (en) * | 2009-01-09 | 2010-08-04 | 索尼公司 | Object detecting device, learning device, object detecting method, and program |
| CN102165488A (en) * | 2008-09-24 | 2011-08-24 | 佳能株式会社 | Information processing apparatus for selecting characteristic feature used for classifying input data |
| CN103793714A (en) * | 2012-10-26 | 2014-05-14 | 卡西欧计算机株式会社 | Multi-class discriminating device, data discrimination device, multi-class discriminating method and data discriminating method |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH10254903A (en) * | 1997-03-14 | 1998-09-25 | Omron Corp | Image retrieval method and device therefor |
| JP2009080556A (en) * | 2007-09-25 | 2009-04-16 | Seiko Epson Corp | Setting method, identification method and program |
-
2015
- 2015-10-07 JP JP2015199723A patent/JP6616645B2/en not_active Expired - Fee Related
- 2015-11-24 CN CN201510823163.4A patent/CN105654108B/en not_active Expired - Fee Related
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080273761A1 (en) * | 2004-06-07 | 2008-11-06 | Kozo Kawata | Image Recognition Device, Image Recognition Method, and Program for Causing Computer to Execute the Method |
| CN1828632A (en) * | 2005-02-28 | 2006-09-06 | 株式会社东芝 | Object detection apparatus, learning apparatus, object detection system, object detection method |
| US20090222753A1 (en) * | 2008-03-03 | 2009-09-03 | Kohei Yamaguchi | Defect inspection tool and method of parameter tuning for defect inspection tool |
| CN102165488A (en) * | 2008-09-24 | 2011-08-24 | 佳能株式会社 | Information processing apparatus for selecting characteristic feature used for classifying input data |
| US20100086176A1 (en) * | 2008-10-03 | 2010-04-08 | Jun Yokono | Learning Apparatus and Method, Recognition Apparatus and Method, Program, and Recording Medium |
| CN101794392A (en) * | 2009-01-09 | 2010-08-04 | 索尼公司 | Object detecting device, learning device, object detecting method, and program |
| CN103793714A (en) * | 2012-10-26 | 2014-05-14 | 卡西欧计算机株式会社 | Multi-class discriminating device, data discrimination device, multi-class discriminating method and data discriminating method |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109425622A (en) * | 2017-08-23 | 2019-03-05 | 富士通株式会社 | Image testing device, image checking method and image inspection program |
| CN112287748A (en) * | 2019-07-24 | 2021-01-29 | 开必拓数据股份有限公司 | Mobile article classification system and method based on self-learning technology |
| CN112287748B (en) * | 2019-07-24 | 2023-09-15 | 开必拓数据股份有限公司 | Mobile article classification system and method based on self-learning technology |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105654108B (en) | 2019-04-12 |
| JP6616645B2 (en) | 2019-12-04 |
| JP2016110626A (en) | 2016-06-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11816579B2 (en) | Method and apparatus for detecting defect pattern on wafer based on unsupervised learning | |
| CN108960266B (en) | Image target detection method and device | |
| US9779354B2 (en) | Learning method and recording medium | |
| US20220269996A1 (en) | Information processing apparatus, information processing method, and storage medium | |
| CN102165488B (en) | Information processing apparatus for selecting characteristic feature used for classifying input data | |
| CN106355188A (en) | Image detection method and device | |
| US20110158535A1 (en) | Image processing apparatus and image processing method | |
| US10275682B2 (en) | Information processing apparatus, information processing method, and storage medium | |
| CN103927510B (en) | Pattern recognition device and image-recognizing method | |
| CN105760488B (en) | Image expression method and device based on multi-level feature fusion | |
| Li et al. | A data-driven metric for comprehensive evaluation of saliency models | |
| CN105654108A (en) | Classifying method, inspection method, and inspection apparatus | |
| CN107004265A (en) | Information processor, the method for processing information, discriminator generating means, the method and program for generating discriminator | |
| CN105654109A (en) | Classifying method, inspection method, and inspection apparatus | |
| CN107103608B (en) | A kind of conspicuousness detection method based on region candidate samples selection | |
| US10248888B2 (en) | Classifying method, storage medium, inspection method, and inspection apparatus | |
| CN105023025B (en) | A kind of opener mark image sorting technique and system | |
| US20140233841A1 (en) | Method of chekcing the appearance of the surface of a tyre | |
| US11972552B2 (en) | Abnormal wafer image classification | |
| CN103198489B (en) | Based on the obvious object automatic testing method of conspicuousness density and skirt response | |
| CN111007871B (en) | Unmanned aerial vehicle dynamic feature identification method, medium, equipment and device | |
| US9959482B2 (en) | Classifying method, storage medium, inspection method, and inspection apparatus | |
| CN103530646A (en) | Complex-object detection using a cascade of classifiers | |
| Zhang et al. | A dual-structure attention-based multi-level feature fusion network for automatic surface defect detection | |
| CN111598116A (en) | Data classification method and device, electronic equipment and readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190412 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |