CN105654109B - Classification method, inspection method and check device - Google Patents
Classification method, inspection method and check device Download PDFInfo
- Publication number
- CN105654109B CN105654109B CN201510847745.6A CN201510847745A CN105654109B CN 105654109 B CN105654109 B CN 105654109B CN 201510847745 A CN201510847745 A CN 201510847745A CN 105654109 B CN105654109 B CN 105654109B
- Authority
- CN
- China
- Prior art keywords
- group
- evaluation
- image
- estimate
- sample
- 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.)
- Active
Links
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
Abstract
The present invention relates to classification method, inspection method and check devices.Image of the classification method provided by the invention based on article and by a group in taxonomy of goods to multiple groups, this method comprises: determining the evaluation method of the evaluation of estimate for obtaining image by using at least some of sample image；The evaluation of estimate of sample image is obtained by determining evaluation method；Change group belonging to the sample of at least one sample image with abnormal evaluation of estimate in sample image；Evaluation method is changed by using the sample image after group belonging to the sample for changing at least one sample image；Obtain the evaluation of estimate of the image of article using the evaluation method of change, and the evaluation of estimate based on the image of article and will be in a group in taxonomy of goods to multiple groups.
Description
Technical field
The present invention relates to fill classification method, inspection method and the inspection in a group in taxonomy of goods to multiple groups
It sets.
Background technique
As the device for executing visual examination or internal check to article, provides and pass through for use by image capture
Elements capture article and the image (target image) that obtains execute the check device of inspection.The check device executes following institute
" study " of meaning: the image of multiple samples in a group in multiple groups has been classified by using each of these
(study image) determines the evaluation method of the evaluation of estimate for obtaining image.The figure of article is obtained based on determining evaluation method
The evaluation of estimate of picture, and in a group being classified into multiple groups based on the evaluation of estimate of acquisition of the article.Therefore, check device
It can be by learning to determine evaluation method, to improve the performance of taxonomy of goods.
Patent document 1 (Japanese Patent Publication No.2010-102690) proposition determined by using multiple study images by
By the combined method for the characteristic quantity that evaluation method uses.Also, patent document 2 (Japanese Patent Publication No.2010-157154)
Following method is proposed, wherein the sample in different classes of is categorized by device and user and is designated in this method, and
And prompt user changes evaluation method so that the sample is categorized into same category by device and user.
In the method described in patent document 1, for example, if multiple study images include to have small defect or low comparison
Spend the image of defect etc., then evaluation method may not be able to be determined to sample classification relevant to the image is original to sample
In the group that should be classified into.In this case, patent document 1 does not describe to change evaluation method to improve taxonomy of goods
Performance technology.In the method described in patent document 2, due to being categorized into different classes of via device and user
Sample is designated, and therefore, whenever evaluation method change, needs to make device by all sample classifications.This may make determination will be by
For the process (learning process) of the evaluation method of taxonomy of goods to be complicated.
Summary of the invention
The present invention provides the complexity for for example advantageously reducing learning process and improves the skill of the performance of taxonomy of goods
Art.
According to an aspect of the present invention, a kind of image based on article is provided and by one in taxonomy of goods to multiple groups
Classification method in a group, this method comprises: being determined by using at least some of sample image for obtaining image
The evaluation method of evaluation of estimate, the figure for multiple samples in a group that sample image instruction has been classified into respectively in multiple groups
Picture；The evaluation of estimate of sample image is obtained by determining evaluation method；Changing in sample image has abnormal evaluation of estimate
At least one sample image sample belonging to group；By using belonging to the sample for changing at least one sample image
Sample image after group changes evaluation method；And the evaluation of the image of article is obtained using the evaluation method of change
Value, and the evaluation of estimate based on the image of article and will be in a group in taxonomy of goods to multiple groups.
According to an aspect of the present invention, provide it is a kind of based on the image of article by one in taxonomy of goods to multiple groups
Classification method in group, this method comprises: being determined by using at least some of sample image for obtaining commenting for image
The evaluation method of value, the image for multiple samples in a group that sample image instruction has been classified into respectively in multiple groups；
The evaluation of estimate of sample image is obtained by evaluation method；Change at least one sample with abnormal evaluation of estimate in sample image
Group belonging to the sample of this image；It is obtained and is changed belonging to the sample of at least one sample image by evaluation method
The evaluation of estimate of sample image after group, and determined based on the evaluation of estimate of the sample image of acquisition for by taxonomy of goods
The threshold value of evaluation of estimate；And in-service evaluation method obtains the evaluation of estimate of the image of article, and based on threshold value and by article point
In a group in class to multiple groups.
According to an aspect of the present invention, a kind of inspection method of inspection for executing article is provided, this method comprises: passing through
Article is captured to obtain the image of article；With use the classification method will be in a group in taxonomy of goods to multiple groups, wherein point
Class method is the image based on article by the method in a group in taxonomy of goods to multiple groups, and include: by using
At least some of sample image determines the evaluation method of the evaluation of estimate for obtaining image, sample image instruction respectively by
It is categorized into the image of multiple samples in a group in multiple groups；The evaluation of sample image is obtained by determining evaluation method
Value；Change group belonging to the sample of at least one sample image with abnormal evaluation of estimate in sample image；By making
Sample image belonging to the sample for changing at least one sample image after group changes evaluation method；Change with using
Evaluation method obtain the evaluation of estimate of the image of article, and the image based on article evaluation of estimate and by taxonomy of goods to more
In a group in a group.
According to an aspect of the present invention, a kind of for executing the check device of the inspection of article, the check device is provided
It include: the image capturing unit for being configured as obtaining the image of article by capture article；Be configured as based on article
Image and by the processing unit in a group in taxonomy of goods to multiple groups, wherein processing unit completes following procedure: passing through
The evaluation method of the evaluation of estimate for obtaining image, sample image instruction difference are determined using at least some of sample image
It has been classified into the image of multiple samples in a group in multiple groups；Sample image is obtained by determining evaluation method
Evaluation of estimate；Change group belonging to the sample of at least one sample image with abnormal evaluation of estimate in sample image；It is logical
It crosses using the sample image after group belonging to the sample for changing at least one sample image and changes evaluation method；And it is logical
Cross using the evaluation method of change and obtain the evaluation of estimate of the image of article, and the image based on article evaluation of estimate and by object
Product are categorized into a group in multiple groups.
According to the explanation of exemplary embodiment, other feature of the invention be will be apparent referring to the drawings.
Detailed description of the invention
Fig. 1 is the schematic diagram for indicating check device according to first embodiment；
Fig. 2 is the flow chart for showing the classification method in processing unit；
Fig. 3 is the flow chart for showing the method (learning method) for obtaining classification information；
Fig. 4 is to indicate to extract each image spy being contained in feature list about each of multiple study images
The table of the result of the characteristic quantity of sign；
Fig. 5 is the diagram of the show example on the screen for indicate display unit；
Fig. 6 is the diagram for showing the accumulation number distribution shown on the display unit；
Fig. 7 is the diagram for showing the comparative example of accumulation number distribution；
Fig. 8 is the diagram for showing the variation of accumulation number distribution.
Specific embodiment
Description exemplary embodiment of the present invention that hereinafter reference will be made to the drawings.Note that in all the drawings, identical attached drawing
Label indicates identical component, and will not provide its repetitive description.
<first embodiment>
The check device 1 of first embodiment according to the present invention will be described referring to Fig.1.Fig. 1 is to indicate to implement according to first
The schematic diagram of the check device 1 of example.For example, check device 1 is executed for such as metal part or resin portion in industrial products
The visual examination of the article 2 divided.May occur such as scraping on the surface of article 2, uneven (for example, uneven color) or
The defect of convex-concave.Check device 1 detects the defect occurred on the surface of article 2 based on the image of article 2, and by article
2 are categorized into a group in multiple groups.The example that first embodiment is described below, plurality of group includes no defective product
Article 2 is categorized into no defective product group and faulty goods by group (first group) and faulty goods group (second group) and check device 1
In one in group.Although first embodiment checks the appearance (surface of article 2) of article 2 by wherein check device 1 is described
Example, but the present embodiment is suitable for wherein checking the situation of the inside of article 2 using X-ray etc..
Check device 1 may include image capturing unit 11, processing unit 12, display unit 13 and input unit 14.Image
Capturing unit 11 is including, for example, lighting unit and camera, and image capturing unit 11 obtains article 2 by capture article 2
Image (target image).The image of the article 2 obtained by image capturing unit 11 is sent to processing unit 12.Processing unit
12 can be by including CPU 12a (central processing unit), RAM 12b (random access memory) and HDD 12c (hard drive)
Information processing unit is realized.Processing unit 12 executes following processing (classification processing): acquisition is obtained by image capturing unit 11
The evaluation of estimate of target image and evaluation of estimate range (threshold value) in the evaluation of estimate and each group based on acquisition and article 2 is divided
In a group in class to multiple groups.CPU 12a executes the program in the group that article 2 is categorized into multiple groups, and
RAM 12b and HDD 12c store the program and data.Display unit 13 is shown including, for example, monitor by processing unit
The result of 12 classification processings executed.Input unit 14 is transmitted including, for example, keyboard and mouse, and by instruction from the user
To processing unit 12.
[classification processing in processing unit 12]
The classification processing in processing unit 12 is described hereinafter with reference to Fig. 2.Fig. 2 is at the classification shown in processing unit 12
The flow chart of the method for reason.In step sl, the condition of image capturing unit 11 when the setting of processing unit 12 captures article 2, with
It obtains wherein by the visual image of defect on the surface of article 2.Condition may include the exposure of such as light angle and camera
Between light time, focus and aperture.In step s 2, processing unit 12 obtains each of these and has been classified into multiple groups
The image (study image, sample image) of multiple samples in one group.Multiple study images can be in advance by image capture
Image unit 11 capture and saved.If there is no the image of preservation, then can be by causing image capturing unit 11 to capture
Multiple samples newly to obtain image.Each of multiple samples are based on its study image via such as user and are classified into
In one in multiple groups (no defective product group and faulty goods group).In the first embodiment, it is (intact that two groups are classified into
Fall into product group and faulty goods group) in one in the image of sample be used as learning image.But the present invention is not limited to
This.For example, as study image, it can be used and more subtly classified according to the type of faulty goods (scrape or uneven etc.)
Sample image.
In step s3, processing unit 12 executes following so-called " study ": obtaining and obtains for using in step S2
At least some of multiple study images information (hereinafter referred to as classification information) that target image is classified.Classification letter
Breath may include evaluation method for obtaining the evaluation of estimate of image and for by the threshold value of the evaluation of estimate of taxonomy of goods.Evaluation method
It is for example for obtaining the function of the evaluation of estimate of image, and multiple characteristic quantities in image can be by being updated to by evaluation method
Use multiple characteristic quantities as obtaining the evaluation of estimate of image in the evaluation method of parameter.Characteristic quantity indicates the feature in image
The amplitude of (hereinafter referred to as characteristics of image).In step s3, processing unit 12 passes through automatically from some characteristics of image
Extraction is estimated as being used for by each study image classification into a group in no defective product group and faulty goods group
Multiple images feature, and create feature list.Determine the characteristic quantity using the multiple images feature being contained in feature list
The a part of evaluation method as classification information as parameter.The method for obtaining classification information will be described in detail later.?
In step S4, processing unit 12 causes image capturing unit 11 to capture article 2, and based on the classification letter obtained in step s3
Breath and in a group that the image of the acquisition of article 2 (target image) is categorized into no defective product group and faulty goods group.
For example, processing unit 12 obtains the evaluation of estimate of target image using evaluation method determining in step s3, and compares and comment
Value and threshold value are to determine group that article is classified into.In step s 5, processing unit 12 shows inspection on display unit 13
As a result.Other than determining no defective product or faulty goods for article 2, processing unit 12 can also be shown on display unit 13
Show the image of the defect part occurred in article 2, the characteristic quantity of each characteristics of image that is contained in feature list and comment
Value etc. is used as inspection result.
[acquisition of classification information]
By the acquisition (study) of the classification information in the processing in the step S3 for describing flow chart shown in Fig. 2 referring to Fig. 3.
Fig. 3 is the flow chart for showing the method (learning method) for obtaining classification information.
In step S3-1, processing unit 12 creates feature list using at least some of multiple study images, and
And determination uses evaluation method of the characteristic quantity for the multiple images feature being contained in feature list as parameter.It below will solution
It releases and wherein uses n study image to create feature list and use Mahalanobis distance as the example of evaluation method.Example
Such as, in order to emphasize the defects of each of multiple study images study image, processing unit 12 is directed to each study image
Execute the Haar wavelet transformation as one of the wavelet transformation of transform method for being transformed into frequency domain.Haar wavelet transformation is energy
Enough processing that frequency transformation is executed while the information of holding position.Processing unit 12 uses first to the provided by formula (1)
The filter of four seed types of four filters to execute each of multiple study images inner product calculating.In formula (1), the
One filter is the filter for extracting the high fdrequency component in vertical direction, and second filter is for extracting diagonal direction
High fdrequency component filter, third filter is the filter for extracting the high fdrequency component in horizontal direction, the 4th filtering
Device is the filter for extracting low frequency component.
This allows the image of four seed types of acquisition of processing unit 12: the figure that wherein high fdrequency component in vertical direction is extracted
The image that the image that is extracted as, the high fdrequency component wherein in diagonal direction, the high fdrequency component wherein in horizontal direction are extracted
The image being extracted with its low frequency components.Compared with the image before transformation, in the image of thus obtained four seed type
Each has 1/2 resolution ratio.Processing unit 12 repeats the image being extracted for its low frequency components and executes the change of Haar small echo
The processing of the image of four seed types with 1/2 resolution ratio is changed and obtains, thus to obtain having layering to reduce the multiple of frequency
Image.
Processing unit 12 is from the image in each layer obtained by Haar wavelet transformation and the image before transformation
Each extract all pixels value global image feature (such as maximum value, average value, variance yields, kurtosis (kurtosis), partially
Spend (skewness) and geometrical mean etc.).The extractable difference such as between contrast, maxima and minima of processing unit 12
The statistical value of value and standard deviation is as global image feature.Through this process, processing unit 12 can be mentioned from multiple study images
Take some characteristics of image.In the present embodiment, some characteristics of image are obtained using Haar wavelet transformation.But for example, it can make
With another transform method of such as another wavelet transformation, edge extracting, Fourier transform or Gabor transformation, to obtain some figures
As feature.Other than global image feature, some characteristics of image also may include the local image characteristics calculated by filtering processing.
Processing unit 12 is each in the characteristics of image being extracted to be directed to using the study image in no defective product group
A calculating score, and the characteristics of image of classification processing will be used for from some image feature selections being extracted, thus create
Feature list.The alternatively method of characteristics of image, for example, providing as described in patent document 1 using zero defect
Study image in product group is come the method for evaluating the combined compatibility of characteristics of image.In the present embodiment, using this method
To select that the characteristics of image of classification processing will be used for.But it is, for example, possible to use another party of such as principal component method
Method.Principal component method is to select have the characteristics of image of high intrinsic value to avoid multiple images spy by eigen value decomposition
The method of the tediously long property of sign.It can prevent from selecting interminable characteristics of image using this method.Although having already been explained above using nothing
Study image in faulty goods group selects the example of characteristics of image, however, the present invention is not limited thereto.For example, defect can be used to produce
Study image in product group selects characteristics of image using the study image in two groups.
The weight of the characteristic quantity for each characteristics of image being contained in the feature list of creation will be determined referring to Fig. 4 description
Method.Fig. 4 is to indicate to extract each image spy being contained in feature list about each of multiple study images
The table of the result of the characteristic quantity of sign.Referring to Fig. 4, each characteristic quantity in each of multiple study images is by XijIt indicates, this
In, i indicate study image ID number (i=1,2 ..., n), j indicate be contained in feature list feature ID number (j=1,
2,…,k).Note that n indicates the number of study image, k indicates the number for the characteristics of image being contained in feature list.N and k
Between relationship can meet n >=k.It shall yet further be noted that MjIndicate the characteristic quantity X in multiple study imagesijAverage value, σjIt indicates more
Characteristic quantity X in a study imageijStandard deviation.
Processing unit 12 passes through formula (2) for each characteristic quantity X in each of multiple study imagesijNormalization.?
In formula (2), YijIndicate the characteristic quantity being each normalized.Processing unit 12 obtains correlation coefficient r by formula (3)pq, and obtain
As formula (4) provide by each correlation coefficient r11~rkkThe inverse matrix A of the correlation matrix R of formation.Inverse matrix A with by comprising
The weight of the characteristic quantity of each characteristics of image in feature list is corresponding.This can determine processing unit 12 by formula (5)
The Mahalanobis distance MD of expressioniAs evaluation method, Mahalanobis distance MDiUsing being contained in feature list
In each characteristics of image characteristic quantity (characteristic quantity being normalized) be used as parameter.
Referring again to flow chart shown in Fig. 3, in step S3-2, processing unit 12 is by using true in step S3-1
Fixed evaluation method (Mahalanobis distance MDi) come be directed to each of multiple study images obtain evaluation of estimate.Processing is single
Member 12 extracts multiple characteristic quantities from each of multiple study images according to feature list, and multiple characteristic quantities are updated to
In evaluation method, evaluation of estimate is obtained thus directed towards each study image.In the first embodiment, by description wherein using abnormal journey
Spend the example of the evaluation of estimate as each study image.Although passing through Mahalanobis distance MD in the present embodimentiTo obtain
Intensity of anomaly (evaluation of estimate), but can be by Euclidean distance or as the projector distance of a type of subspace method
To obtain intensity of anomaly (evaluation of estimate).
In step S3-3, processing unit 12 is for the multiple intensity of anomaly (evaluation of estimate) for learning images of each group of generation
Distribution, and it is shown on display unit 13.In step S3-4, processing unit 12 determines the figure of the sample between multiple groups
Whether the different degree in the range (evaluation of estimate range) of the evaluation of estimate of picture meets permissible value.If different degree is unsatisfactory for allowing
Value, then processing proceeds to step S3-5；Otherwise, obtaining (study) classification information terminates.For example, can be allowed by user preset
Value.
Fig. 5 is the diagram of the show example on the screen for indicate display unit 13.In the region 13a of display unit 13,
The distribution (histogram) of intensity of anomaly (evaluation of estimate) about each study image of each group of display.In histogram, informal voucher table
Show the number of the study image in no defective product group, and secret note indicates the number of the study image in faulty goods group.?
In the region 13b of display unit 13, the faulty goods of threshold value of the display for determining the evaluation of estimate for target image to be classified
Verification and measurement ratio and orthogonality as different degree.
It is predetermined that the verification and measurement ratio of faulty goods indicates that the sample in predetermined group should be classified into multiple groups is classified into this
Ratio in group, and indicate that such as faulty goods is classified as the ratio of faulty goods.It can be according to the verification and measurement ratio of faulty goods
Carry out threshold value.The verification and measurement ratio of faulty goods can be arbitrarily set by the user, but it is generally set at 100%, to will not
Faulty goods is classified as no defective product.If the verification and measurement ratio of faulty goods is 100%, processing unit 12 sets threshold value
For the value of the minimum value of the intensity of anomaly less than the study image in faulty goods group.That is, 12 given threshold of processing unit, thus
So that all study images in faulty goods group are disposed in the right side of threshold value in histogram shown in Fig. 5.In Fig. 5 institute
In the example shown, threshold value 13c is set as being less than the abnormal journey of the smallest study image 13k in faulty goods group by processing unit 12
Degree.
Orthogonality instruction as different degree all has the study image of the intensity of anomaly smaller than threshold value and zero defect produces
The ratio of all study images in product group.Generally, orthogonality can be high.When orthogonality is 100%, obtain wherein intact
All study images fallen into product group are disposed in all study images in the left side and faulty goods group of threshold value and are arranged
Perfect condition on the right side of threshold value.
Although executing the processing for determining whether different degree meets permissible value by processing unit 12 in the present embodiment,
The processing can be executed by such as user.In this case, if the user determine that different degree (orthogonality) is unsatisfactory for allowing
Value, then he/her presses " additional to learn " button 13d via input unit 14；Otherwise, he/her presses via input unit 14
" terminating study " button 13e.If user's pressing " additional study " button 13d, processing unit 12 proceed to step S3-5.
If user's pressing " terminating study " button 13e, processing unit 12 terminate to obtain (study) classification information.
In step S3-5, processing unit 12 is based on the intensity of anomaly (evaluation for the multiple study images of each group of expression
Value) information, specify with abnormal (singular) intensity of anomaly in the intensity of anomaly of multiple study images at least
One sample.In the first embodiment, as information, the histogram of the intensity of anomaly in each group is used.Processing unit 12 can
It is subordinated to the evaluation of estimate of the image of the range of the evaluation of estimate of the image of the sample of no defective product group and the sample of faulty goods group
Range overlapping part (referred to below as lap) sample, select at least one sample for the evaluation of estimate for having abnormal
This.The range 13f for the intensity of anomaly in histogram that lap instruction is for example shown in the region 13a of display unit 13.
For example, ascending order of the processing unit 12 in the sample being contained in faulty goods group and lap with intensity of anomaly specify to
A few sample.As an alternative, processing unit 12 in the sample being contained in no defective product group and lap with
The descending of intensity of anomaly specifies at least one sample.For example, can be by user preset by the number of appointed sample.Processing unit
12 can show the study image of appointed sample in the region 13g of display unit 13.Although in the present embodiment by handling
Specified at least one sample with abnormal evaluation of estimate of unit 12, but this can be executed by such as user and operated.In such case
Under, user can by the histogram in the region 13a for being shown in display unit 13 via input unit 14 and select tool
There is the study image of abnormal evaluation of estimate, to specify at least one sample.
In step S3-6, processing unit 12 changes group belonging at least one sample specified in step S3-5.
For example, if at least one sample specified in step S3-5 belongs to faulty goods group, processing unit 12 will at least one
Group belonging to a sample becomes no defective product group from faulty goods group.In the example shown in FIG. 5, if in step S3-5
In specify faulty goods group in sample relevant to the study image 13k with minimum intensity of anomaly, then processing unit 12
Group belonging to sample relevant to study image 13k is become into no defective product group.
Processing unit 12 can be configured to the study figure that appointed article is shown in the region 13g of display unit 13
Picture, and prompt the user to determine whether to change group belonging to study image.In this case, processing unit 12 prompts user
Determined whether by referring to the study image shown in the region 13g of display unit 13 by sample relevant to study image
It is determined as no defective product or faulty goods or deletes study image.For example, when the sample specified in step S3-5 belongs to
When faulty goods group, if user presses " no defective product determines " button 13h, processing unit 12 via input unit 14
Group belonging to appointed sample is become into no defective product group.On the other hand, if user's pressing " faulty goods determines "
Button 13i, then processing unit 12 terminates study.As an alternative, it if user presses " deletion " button 13j, handles
Unit 12 deletes the study image of appointed sample.
In the step S3-1 after the processing in step S3-6,12 use of processing unit changes at least in step S3-6
The study image of multiple samples after group belonging to one sample and newly create feature list, and change evaluation method.
In the step S3-2 after the processing in step S3-6, processing unit 12 is directed to multiple study using the evaluation method of change
Each of image obtains evaluation of estimate.In the step S3-3 after the processing in step S3-6, processing unit 12 is about every
A group newly generate it is each study image intensity of anomaly (evaluation of estimate) distribution.The step of after the processing in step S3-6
In S3-4, processing unit 12 determines whether different degree meets permissible value, and new true also according to the verification and measurement ratio of faulty goods
Determine threshold value.In the first embodiment, processing unit 12 returns to step S3-1 after the processing in step S3-6, and newly creates
It builds feature list and changes evaluation method.But the invention is not limited thereto.For example, processing unit 12 can be in step S3-5
Processing after return to step S3-3, and according to the verification and measurement ratio of faulty goods and only new threshold value does not have step S3-1
Evaluation method change or each study image in step S3-3 intensity of anomaly distribution generation.In such case
Under, processing unit 12 obtains the evaluation of estimate of target image using the evaluation method determined in the processing of step S3-1, and
The group that article is classified into is determined according to newly determining threshold value in step S3-3 and by the taxonomy of goods in multiple groups
A group in.
As described above, 1 in-service evaluation method of check device according to first embodiment is for every in multiple study images
One acquisition evaluation of estimate.Check device 1 changes the abnormal evaluation of estimate in multiple evaluations of estimate with multiple study images extremely
The group of a few sample, and feature list is newly created, thus change evaluation method.This allows check device 1 to execute high-precision
Study, and using the evaluation method changed for target image obtain evaluation of estimate and accurately by taxonomy of goods in multiple groups
A group in.
<second embodiment>
In the first embodiment, it when specified at least one sample with abnormal intensity of anomaly, explains following
Example, wherein the histogram of the intensity of anomaly in each group is used as each of multiple study images of each group of expression
Intensity of anomaly information.In a second embodiment, the example being explained as follows, which, which uses, passes through intensity of anomaly (evaluation of estimate)
Each study image of the no defective product group of arrangement with there is intensity of anomaly with each study image of no defective product group
Relationship between the number of the study image of the faulty goods group of corresponding intensity of anomaly, as information.The relationship is claimed below
For " accumulation number distribution ".
Fig. 6 is the diagram for showing accumulation number distribution.Accumulation number distribution can be produced in step S3-3 by processing unit 12
It is raw and shown in the region 13a of display unit 13.Horizontal axis in Fig. 6, which is shown, to be caused with the ascending order arrangement of intensity of anomaly
The ID number of each study image of no defective product group, and the longitudinal axis is shown with each study image with no defective product group
The corresponding intensity of anomaly of intensity of anomaly faulty goods group study image number (accumulation number).
For example, in Fig. 6, in the case where the ID number (horizontal axis) of the study image of no defective product group is " 30 ", defect
The accumulation number (longitudinal axis) of the study image of product group increases to 1.This shows in the presence of the 30th had with no defective product group
Learn a study image of the defect group of the corresponding intensity of anomaly of intensity of anomaly of image.Specifically, it shows defect production
The intensity of anomaly of one in product group study image be in the 30th study image in no defective product group intensity of anomaly and
Between the intensity of anomaly of 31st study image.
Similarly, it in Fig. 6, in the case where the ID number (horizontal axis) of the study image of no defective product group is " 40 ", lacks
The accumulation number (longitudinal axis) for falling into the study image of product group increases to 2.This shows in the presence of with the 40th with no defective product group
One study image of the defect group of the corresponding intensity of anomaly of intensity of anomaly of a study image.Specifically, it shows defect
The intensity of anomaly of a study image in product group is in the intensity of anomaly of the 40th study image in no defective product group
Between the intensity of anomaly for learning image with the 41st.
Explained later uses the advantages of accumulation number distribution shown in fig. 6.Three are primarily present using accumulation number distribution
Advantage.First the advantage is that graphics shape is uniquely identified.For example, if histogram is used as indicating for each group
The information of intensity of anomaly, then graphics shape cannot be determined in the case where not setting cabinet (bin).But if shown in Fig. 6
Accumulation number distribution be used as information, then graphics shape can be uniquely determined in the case where not setting cabinet etc..
Second the advantage is that the study image of the smallest defect group of intensity of anomaly can be detected easily.Generally, it is examining
Look into system, can how soon and mostly accurately identify that the faulty goods closest to no defective product is accurately by image classification
Big problem.If, cannot be in two histograms without reference to no defective product group and faulty goods group using histogram
In the case where detect the study image of the minimum faulty goods group of intensity of anomaly.But accumulation number distribution shown in Fig. 6
In, by the way that it is minimum can easily and precisely to detect intensity of anomaly only referring to the data expressed by curve (plot line)
Faulty goods group study image.
Whether third advantage is, can readily determine that effective according to the learning outcome of graphics shape.In other words, exist
It accumulates in number distribution, easily can grasp orthogonality (different degree) from graphics shape (slope of a curve).Fig. 7 shows use
In the comparative example (three examples) of accumulation number distribution.The case where solid line 71 in Fig. 7 is shown below, wherein in faulty goods group
All study images intensity of anomaly be greater than no defective product group in study image maximum intensity of anomaly and zero defect produce
Product group and faulty goods group are kept completely separate.In addition, the case where dotted line 72 in Fig. 7 is shown below, wherein although in faulty goods
There is the study image with the intensity of anomaly smaller than the maximum intensity of anomaly in no defective product group in study image in group,
But no defective product group and faulty goods group are well separated.Also, the case where chain-dotted line 73 in Fig. 7 is shown below, wherein
In the study image in faulty goods group, the maximum intensity of anomaly that is particularly lower than there are intensity of anomaly in no defective product group
Study image, and learn deficiency.The reason of as such as graphics shape as shown in chain-dotted line 73 is formed, for example, can be with
Enumerate following reason: the study image that original should be classified into no defective product group is classified into faulty goods group；By nothing
Characteristics of image needed for the classification of faulty goods group is not extracted.
Here, in accumulation number distribution, orthogonality (different degree) is obtained from graphics shape (slope of a curve)；In step
Determine whether orthogonality obtained meets permissible value in rapid S3-4.In addition, accumulation number distribution is not limited to example shown in fig. 6
Son, also, for example, as shown in figure 8, with Fig. 6 horizontally and vertically compared with, can horizontally and vertically overturn.
<other embodiments>
The embodiment of the present invention can also be realized that the computer of the system or device is read by the computer of system or device
And it executes and is recorded on storage medium (it can also be more completely known as " non-transitory computer-readable storage media ")
Computer executable instructions (for example, one or more program) to execute the function of one or more above-described embodiments, and/
It or include for executing one or more circuits of the function of one or more above-described embodiments (for example, specific integrated circuit
It (ASIC)), and can be executable by the system or the computer of device, such as by the way that computer is read and executed from storage medium
Instruction is to execute the function of one or more above-described embodiments, and/or controls one or more circuits to execute one or more
The function of above-described embodiment and the method executed realize the embodiment of the present invention.The computer may include one or more places
It manages device (such as central processing unit (CPU), microprocessing unit (MPU)), and may include isolated computer or separation
Processor network, to read and execute the computer executable instructions.The computer executable instructions can for example by
Computer is supplied to from network or storage medium.Storage medium may include, for example, hard disk, random access memory
(RAM), read-only memory (ROM), the storage equipment of distributed computing system, CD (such as compact disk (CD), digital versatile disc
(DVD) or Blu-ray disc (BD)TM), one or more of flash memory device and storage card etc..
Other embodiments
The embodiment of the present invention can also be realized by following method, that is, pass through network or various storage mediums
The software (program) for executing the function of above-described embodiment is supplied to system or device, the computer of the system or device or in
The method that Central Processing Unit (CPU), microprocessing unit (MPU) read and execute program.
Although illustrating the present invention with reference to exemplary embodiment, it should be appreciated that the present invention is not limited to disclosed exemplary
Embodiment.Scope of the appended claims should be endowed broadest explanation, to include all this modifications and equivalent structure
And function.
Claims (14)
1. a kind of evaluation of estimate of the image based on article is by the classification method in a group in taxonomy of goods to multiple groups, special
Sign is, which comprises
Determine the evaluation method for obtaining evaluation of estimate based on sample image, sample image be classified into respectively it is described more
The image of multiple samples in a group in a group；
The evaluation of estimate of sample image is obtained by determining evaluation method；
In at least one of the multiple group group, changing among the sample image in each group has abnormal evaluation of estimate
Sample image sample group；
Change evaluation side based on the sample image in a group being classified into the multiple group respectively after change group
Method；And
Obtain the evaluation of estimate of the image of article using the evaluation method of change, and the article based on acquisition image evaluation
Value and will be in a group in taxonomy of goods to the multiple group.
2. according to the method described in claim 1, wherein, change group and change evaluation method are repeated, so that described more
Different degree in the evaluation of estimate range of sample image between a group meets permissible value.
3. according to the method described in claim 1, wherein,
The multiple group includes first group and second group, and,
In change group, to change sample of the sample of group selected from the evaluation of estimate range for the sample image for belonging to first group with second group
The sample of the part of the evaluation of estimate range overlapping of this image.
4. according to the method described in claim 1, wherein, when obtaining the evaluation of estimate of sample image, being based on from sample image
Each extract as evaluation method by multiple characteristic quantities of the parameter used, by the way that the multiple characteristic quantity is updated to
The evaluation of estimate of each of sample image is obtained in evaluation method.
5. according to the method described in claim 1, wherein, changing evaluation method includes: the sample image obtained after change group
New Appraisement value, and determined based on the New Appraisement value of the sample image of acquisition for by the threshold of the evaluation of estimate of taxonomy of goods
Value.
6. according to the method described in claim 5, wherein, when changing evaluation method, according to should be classified into multiple groups
Sample in predetermined group is classified into the target rate in this predetermined group, carrys out threshold value.
7. in change group, the sample image with abnormal evaluation of estimate is deleted according to the method described in claim 1, wherein,
Rather than change the group with the sample of sample image of abnormal evaluation of estimate.
8. according to the method described in claim 1, wherein, change group includes display for every in each group of expression sample image
The information of the evaluation of estimate of one sample image.
9. according to the method described in claim 8, wherein, the information includes the histogram of the evaluation of estimate in each group.
10. according to the method described in claim 8, wherein,
The multiple group includes first group and second group, and,
The information includes the sample having with first group in first group of the sample image and second group arranged by evaluation of estimate
Relationship between the number of the sample image of the corresponding evaluation of estimate of the evaluation of estimate of this image.
11. according to the method described in claim 8, wherein, change group includes to specify tool based on the information shown in display
There is the sample image of abnormal evaluation of estimate.
12. a kind of evaluation of estimate of the image based on article is by the classification method in a group in taxonomy of goods to multiple groups,
It is characterized in that, which comprises
The evaluation method for obtaining evaluation of estimate is determined based on sample image, sample image is to be classified into multiple groups respectively
In a group in multiple samples image；
The evaluation of estimate of sample image is obtained by evaluation method；
In at least one of the multiple group group, changing among the sample image in each group has abnormal evaluation of estimate
Sample image sample group；
Obtain the evaluation of estimate of sample image after change group by evaluation method, and the sample image based on acquisition is commented
Value is determined for by the threshold value of the evaluation of estimate of taxonomy of goods；And
In-service evaluation method obtains the evaluation of estimate of the image of article, and based on threshold value by one in taxonomy of goods to multiple groups
In a group.
13. a kind of inspection method for the inspection for executing article, which is characterized in that the described method includes:
The image of article is obtained by shooting article；And
Using classification method by a group in taxonomy of goods to multiple groups,
Wherein, classification method is the evaluation of estimate of the image based on article and by the side in a group in taxonomy of goods to multiple groups
Method, and include:
The evaluation method for obtaining evaluation of estimate is determined based on sample image, sample image is to be classified into multiple groups respectively
In a group in multiple samples image；
The evaluation of estimate of sample image is obtained by determining evaluation method；
In at least one of the multiple group group, changing among the sample image in each group has abnormal evaluation of estimate
Sample image sample group；
Change evaluation method based on the sample image for a group being classified into the multiple group respectively after change group；
And
Obtain the evaluation of estimate of the image of article using the evaluation method of change, and the article based on acquisition image evaluation
Value and will be in a group in taxonomy of goods to multiple groups.
14. a kind of for executing the check device of the inspection of article, which is characterized in that the check device includes:
Image capturing unit is configured as obtaining the image of article by shooting article；With
Processing unit is configured as based on the evaluation of estimate of the image of article come by a group in taxonomy of goods to multiple groups,
Wherein, the processing unit completes following procedure:
The evaluation method for obtaining evaluation of estimate is determined based on sample image, sample image is to be classified into multiple groups respectively
In a group in multiple samples image；
The evaluation of estimate of sample image is obtained by determining evaluation method；
In at least one of the multiple group group, changing among the sample image in each group has abnormal evaluation of estimate
Sample image sample group；
Change evaluation method based on the sample image for a group being classified into the multiple group respectively after change group；
And
Obtain the evaluation of estimate of the image of article using the evaluation method of change, and the article based on acquisition image evaluation
Value and will be in a group in taxonomy of goods to multiple groups.
Applications Claiming Priority (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2014-242305 | 2014-11-28 | ||
| JP2014242305 | 2014-11-28 | ||
| JP2015199722A JP6643856B2 (en) | 2014-11-28 | 2015-10-07 | Classification method, inspection method, inspection device, and program |
| JP2015-199722 | 2015-10-07 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105654109A CN105654109A (en) | 2016-06-08 |
| CN105654109B true CN105654109B (en) | 2019-03-05 |
Family
ID=56122196
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510847745.6A Active CN105654109B (en) | 2014-11-28 | 2015-11-27 | Classification method, inspection method and check device |
Country Status (2)
| Country | Link |
|---|---|
| JP (1) | JP6643856B2 (en) |
| CN (1) | CN105654109B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP6381144B2 (en) * | 2016-10-06 | 2018-08-29 | ソリッドビジョン株式会社 | Imaging processing device |
| WO2018179559A1 (en) * | 2017-03-27 | 2018-10-04 | 三菱重工業株式会社 | Defect detection system for aircraft component and defect detection method for aircraft component |
| JP7130984B2 (en) * | 2018-03-01 | 2022-09-06 | 日本電気株式会社 | Image judgment system, model update method and model update program |
| WO2019194064A1 (en) * | 2018-04-02 | 2019-10-10 | 日本電産株式会社 | Image processing device, image processing method, appearance inspection system, and appearance inspection method |
| JP2018164272A (en) * | 2018-05-28 | 2018-10-18 | ソリッドビジョン株式会社 | Imaging processing apparatus |
Citations (3)
| 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 |
| 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 |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2001156135A (en) * | 1999-11-29 | 2001-06-08 | Hitachi Ltd | Method and device for sorting defective image and manufacturing method of semiconductor device using them |
| JP4606779B2 (en) * | 2004-06-07 | 2011-01-05 | グローリー株式会社 | Image recognition apparatus, image recognition method, and program causing computer to execute the method |
| JP2009080557A (en) * | 2007-09-25 | 2009-04-16 | Seiko Epson Corp | Identification method and program |
| JP4629118B2 (en) * | 2008-03-03 | 2011-02-09 | 株式会社日立ハイテクノロジーズ | Defect inspection apparatus and parameter adjustment method used for this defect inspection apparatus. |
| JP2009282699A (en) * | 2008-05-21 | 2009-12-03 | Seiko Epson Corp | Detection of organ area corresponding to image of organ of face in image |
| JP4670976B2 (en) * | 2008-10-03 | 2011-04-13 | ソニー株式会社 | Learning device and method, recognition device and method, program, and recording medium |
| EP2678665A4 (en) * | 2011-02-24 | 2017-07-05 | 3M Innovative Properties Company | System for detection of non-uniformities in web-based materials |
| JP5214762B2 (en) * | 2011-03-25 | 2013-06-19 | 株式会社東芝 | Recognition device, method and program |
| JP2013117861A (en) * | 2011-12-02 | 2013-06-13 | Canon Inc | Learning device, learning method and program |
-
2015
- 2015-10-07 JP JP2015199722A patent/JP6643856B2/en active Active
- 2015-11-27 CN CN201510847745.6A patent/CN105654109B/en active Active
Patent Citations (3)
| 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 |
| 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 |
| CN101794392A (en) * | 2009-01-09 | 2010-08-04 | 索尼公司 | Object detecting device, learning device, object detecting method, and program |
Also Published As
| Publication number | Publication date |
|---|---|
| JP2016110625A (en) | 2016-06-20 |
| JP6643856B2 (en) | 2020-02-12 |
| CN105654109A (en) | 2016-06-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105654109B (en) | Classification method, inspection method and check device | |
| JP5414416B2 (en) | Information processing apparatus and method | |
| CN107111872B (en) | Information processing apparatus, information processing method, and storage medium | |
| US10891508B2 (en) | Image processing apparatus, method, and non-transitory computer-readable storage medium | |
| US20110158535A1 (en) | Image processing apparatus and image processing method | |
| US20170330315A1 (en) | Information processing apparatus, method for processing information, discriminator generating apparatus, method for generating discriminator, and program | |
| JP2010507139A (en) | Face-based image clustering | |
| JP2015041164A (en) | Image processor, image processing method and program | |
| US10248888B2 (en) | Classifying method, storage medium, inspection method, and inspection apparatus | |
| WO2013005815A1 (en) | Object detection device, object detection method, and program | |
| JP5290227B2 (en) | Object detection device and learning device thereof | |
| JP2017102865A (en) | Information processing device, information processing method and program | |
| KR100868884B1 (en) | Flat glass defect information system and classification method | |
| WO2020129041A1 (en) | Classifying defects in a semiconductor specimen | |
| CN105654108B (en) | Classification method, inspection method and check device | |
| US9959482B2 (en) | Classifying method, storage medium, inspection method, and inspection apparatus | |
| Bharadwaj et al. | Can holistic representations be used for face biometric quality assessment? | |
| WO2016092783A1 (en) | Information processing apparatus, method for processing information, discriminator generating apparatus, method for generating discriminator, and program | |
| JP5644655B2 (en) | Edible fruit inspection method, inspection apparatus and program | |
| CN114298137A (en) | Tiny target detection system based on countermeasure generation network | |
| WO2016092779A1 (en) | Information processing apparatus, information processing method, and program | |
| JP5592414B2 (en) | Template evaluation apparatus, microscope apparatus, and program | |
| JP2019149028A (en) | Information processing device, control method and program therefor | |
| JP2014142213A (en) | Photographing parameter determination device and control method of the same | |
| CN116012607B (en) | Image weak texture feature extraction method and device, equipment and 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 |