CN104199823B - A kind of fabric defects dynamic testing method of view-based access control model data-driven - Google Patents
A kind of fabric defects dynamic testing method of view-based access control model data-driven Download PDFInfo
- Publication number
- CN104199823B CN104199823B CN201410334091.2A CN201410334091A CN104199823B CN 104199823 B CN104199823 B CN 104199823B CN 201410334091 A CN201410334091 A CN 201410334091A CN 104199823 B CN104199823 B CN 104199823B
- Authority
- CN
- China
- Prior art keywords
- subgraph
- notable
- detail
- approximate
- brightness
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
Abstract
The invention discloses a kind of fabric defects dynamic testing method of view-based access control model data-driven, specifically implement according to following steps:Step 1, the textile image of the rgb space of collection is converted into HSV space textile image;Step 2, the saturation degree feature S and brightness V-arrangement of image are extracted into saturation degree characteristic pattern and brightness figure;Step 3, drive to form notable figure using vision data on the saturation degree characteristic pattern and brightness figure obtained in step 2;Step 4, on the basis of step 3, by extreme difference threshold value and fault information is split;Step 5, by the fault information fusion split into complete fault information.A kind of fabric defects dynamic testing method of view-based access control model data-driven of the present invention, solves that Detection accuracy in the prior art is not high, and defect segmentation is inaccurate, the problem of detection universality is not strong.
Description
Technical field
The invention belongs to fabric defects dynamic testing method technical field, and in particular to a kind of view-based access control model data-driven
Fabric defects dynamic testing method.
Background technology
Fabric defects detection is one of most important part of quality of textile products control.At present, traditional fabric defects detection
It is to be completed by artificial offline inspection.However, human attention can be influenceed by time, the detection factor such as environment and mood, very
Easily cause the defects such as flase drop and missing inspection.In order to solve the shortcoming manually detected, Automatic Detection of Fabric Defects turns into state in recent years
One of hot subject of inside and outside scholar's research.It is crucial as fabric defects detection with computer, the development of image processing techniques
The image processing algorithm of technology necessarily turns into the focus of research.
In spatial domain, Gauss Gaussian Markov random field texture model is studied fabric defects, utilizes Ma Erke
Husband's model training obtains the model criteria parameter of normal fabric, passes through the model parameter and canonical parameter of relatively more tested image
The distance between judge whether fault;But its method calculate it is complicated, to noise-sensitive, do not need on-line study, thus not
Fast automatic detecting (the bibliography [1] of fabric defects can be realized:Cohen F S,Fan Z,Attali s.Automated
Inspection of Textile Fabric Using Textile Models[J].IEEE Transactions on
Pattern Analysis and Machine Intelligence,1991,13(8):803-808.).In transform domain, Fourier
Leaf transformation can be used for defect detection (bibliography [2]:V.,Jayashree and Shaila
Subbaraman.Identification of twill grey fabric defects using DC suppress
Fourier power spectrum sum features[J].Textile Research Journal,2012,82(14):
1485-1497. bibliography [3]:Abdel Sanlam Malek,Jean-Yves Drean,Lauren Bigue,
etc.Optimization of outomated online fabric inspection by fast Fourier
transform and cross-correlation[J].Textile Research Journal,2013,83(3):256-
268.), but Fourier transform is a kind of method of overall importance, it is impossible to which the location information of any spatial domain is provided.Wavelet transformation has
There is multiple dimensioned resolution character, and there is the ability for characterizing signal local feature in time domain, frequency domain, be especially suitable for singular signal
Detect (bibliography [4]:Tsai DM and Hsiao B.Automatic surface inspection using
wavelet reconstruction[J].Pattern Recognition;2001,34(6):1285–1305.).Based on heredity
The fabric adaptive orthogonal wavelet base of planning preferably goes out adaptive wavelet packet transform using genetic programming algorithm combination fitness function, with
Solve the problem of adaptive wavelet packet transform is difficult to preferred (bibliography [5]:The such as Niu Cuncai, Wang Jun, Zhang Xiaonan are based on genetic planning
Fabric adaptive orthogonal wavelet base construction and optimization [J] textile journals, 2012,33 (9):40-45.).But this algorithm
Adaptive orthogonal wavelet storehouse must first be constructed, amount of calculation is very big, while it is preferred that wavelet basis be not necessarily best wavelet.
By being analyzed above, the either fault dynamic testing method of spatial domain, or the Fourier in transform domain
Conversion and small wave converting method, are all, from algorithm, not make full use of textile image data-driven information itself, no
The contrast between fabric defects information and background information can be effectively increased;Therefore, these fabric defects detection accuracys rate are not high,
Easily influenceed by external environment, do not possess pervasive answer, it is impossible to meet actual dynamic detection needs.
The content of the invention
It is an object of the invention to provide a kind of fabric defects dynamic testing method of view-based access control model data-driven, solve existing
The problem of having in technology not high Detection accuracy, inaccurate defect segmentation and not strong detection universality.
The technical solution adopted in the present invention is, a kind of fabric defects dynamic testing method of view-based access control model data-driven,
Specifically implement according to following steps:
Step 1, the textile image of the rgb space of collection is converted into HSV space textile image;
Step 2, the saturation degree feature S and brightness V-arrangement of image are extracted into saturation degree characteristic pattern and brightness figure;
Step 3, drive to form aobvious using vision data on the saturation degree characteristic pattern and brightness figure obtained in step 2
Write figure;
Step 4, on the basis of step 3, by extreme difference threshold value and fault information is split;
Step 5, by the fault information fusion split into complete fault information.
The features of the present invention is also resided in,
Step 1 specific embodiment is:
The textile image that rgb space textile image is converted to HSV space, transfer process such as formula are gathered by imaging sensor
(1) shown in;
Wherein, R, G, B span are [0,255];H span is [0,360];S span be [0,
1];V span is [0,255].
Step 2 specific embodiment is:
Image color feature, shape are described using the saturation degree S and brightness V that meet in the HSV space of human vision characteristicses
Into the saturation degree characteristic pattern and brightness figure of textile image.
Step 3 is that saturation degree characteristic pattern and brightness figure are carried out into small echo multilayer decomposition, the approximate spy of different resolution
Levy between subgraph, central peripheral operation, central peripheral are carried out between minutia subgraph enter between operating the difference subgraph to be formed
Row is added fusion and obtains approximate overall notable figure, the overall notable figure of level detail and the overall notable figure of vertical detail;
Specifically implement according to following steps:
Saturation degree and brightness figure are carried out small echo multilayer decomposition by step 3.1 respectively;
Step 3.1.1, Selection of Wavelet Basis:
The sym2 small echos with orthogonality, near symmetry, compact sup-port are chosen to filter for Wavelet Multiresolution Decomposition;
Step 3.1.2, Decomposition order is determined:
J layers of decomposition of small echo are carried out to the characteristic pattern of normal fabric using sym2 small echos, jth layer approximation characteristic after decomposition
Figure energyLevel detail feature subgraph energyAnd vertical detail feature subgraph energyUsing formula (2), (3)
(4) shown in;
If the ratio between jth layer details and approximation characteristic subgraph energy are βj, represented using formula (5);
Work as βj-1<βjAnd βj>βj+1, then the wavelet decomposition number of plies is j+1;
Wherein, M × N is sub-graph size, fj LL(x, y), fj LH(x, y), fj HL(x, y) respectively represent jth yardstick it is approximate,
The wavelet coefficient of level detail, vertical detail feature subgraph at (x, y) place.
Step 3.1.3, small echo multilayer is decomposed:
If sym2 wavelet low-pass filters coefficient is hk, high-pass filter coefficient is gk, many points are realized using multilayer decomposition
Resolution is filtered, shown in specific formula (6):
Wherein:Represent that j layers are decomposed approximate respectively
Feature subgraph, level detail feature subgraph, the wavelet coefficient at vertical detail feature subgraph (x, y) place,Represent respectively j-1 layer decomposition approximation characteristic subgraph,
Level detail feature subgraph, vertical detail feature subgraph (k1,k2) place wavelet coefficient;
So as to obtain j layers and decompose approximate subgraphLevel detail subgraphVertical detail subgraphWith diagonal subgraph
Step 3.2, central peripheral operation is carried out;
To the approximation characteristic subgraph obtained in step 3.1Between, level detail feature subgraphIt
Between and vertical detail feature subgraphBetween carry out difference operation, i.e., central peripheral operate, obtain fabric defects figure
The approximate difference feature subgraph of pictureLevel detail Differential Characteristics subgraphWith vertical detail Differential Characteristics subgraphInstitute
Using central peripheral operation such as following formula (7), (8) and (9);
Wherein, central peripheral is operated between Θ is characterized subgraph, and yardstick centered on c, s=c+ δ represent periphery yardstick, δ tables
Show that central peripheral yardstick is poor;
Step 3.3, Differential Characteristics subgraph fusion is carried out;
Respectively to the approximate difference feature subgraph of the textile image obtained in step 3.2Level detail Differential Characteristics
SubgraphWith vertical detail Differential Characteristics subgraphIt is normalized, respectively obtains, it is approximate poor after normalized
Dtex levies subgraphLevel detail Differential Characteristics subgraphVertical detail Differential Characteristics subgraph
Specific processing is using formula such as shown in (10), (11) and (12);
To the approximate difference feature subgraph after normalizedBetween, level detail Differential Characteristics subgraphBetween,
Vertical detail Differential Characteristics subgraphBetween, it is respectively adopted and is merged and obtained respectively using formula (13), (14) and (15)
Approximation characteristic notable figure N (fLL), level detail characteristic remarkable picture N (fLH) and vertical detail characteristic remarkable picture N (fHL);
Wherein, n is the poor maximum of central peripheral yardstick;
Characteristic remarkable picture is fused to overall notable figure, it is shown using formula such as (16), (17), (18):
Wherein, N1(fLL) and N2(fLL) approximate saturation degree feature and brightness notable figure, N are represented respectively1(fLH) and N2
(fLH) level detail saturation degree and brightness notable figure, N are represented respectively1(fHL) and N2(fHL) respectively table vertical detail show saturation
Degree and brightness notable figure;WithRepresent that approximate overall notable figure, level detail are overall respectively
Notable figure and the overall notable figure of vertical detail.
Step 4 is specifically implemented according to following steps:
Step 4.1, according to formula (19) to the approximate overall notable figure that is acquired in step 3Level detail is whole
Body notable figureWith the overall notable figure of vertical detailMiddle determination segmentation threshold Tz;
Step 4.2, image segmentation is carried out according to formula (20), defect regions is classified as above or equal to the part of threshold valuePart less than threshold value is classified as background areaIt is approximate overall notable so as to obtain respectively
Defect regions in figureDefect regions in level detail entirety notable figureWith it is vertical thin
Defect regions in the overall notable figure of section
Wherein,Overall notable figure is represented, z represents overall significantly graph type, LL, LH, HL are taken respectively;WithRepresent that the vector sum minimum that maximum is constituted in overall notable figure each column is constituted respectively
Vector;WithRepresent respectively by the overall notable figure vector sum pole that maximum is constituted in often capable
The vector of small value composition;Mean { } represents mean operation;min[thz1,thz2] represent thz1, thz2In minimum;Defect regions in different types of overall notable figure are represented, i.e.,Represent approximate overall notable figure
In split defect regions,The defect regions split in the overall notable figure of level detail are represented,Represent the defect regions split in the overall notable figure of vertical detail.
Step 5 is specially:
By the fabric defects approximate information split by formula (19) and (20) in step 4Knit
Thing defect level detailed informationWith fabric defects vertical detail informationIt is added together, shape
Into complete fabric defects [R], using formula such as shown in (21):
Wherein, [R] is complete fabric defects,Represent to be added.
The beneficial effects of the invention are as follows a kind of fabric defects dynamic testing method of view-based access control model data-driven is empty in HSV
Between in, using in textile image saturation degree and brightness as master data, decomposed by small echo multilayer, central peripheral is operated,
The proceduredriven formation notable figure such as Differential Characteristics subgraph fusion, and with row extreme difference average, row extreme difference average in notable figure
Minimum value goes out fault information as Threshold segmentation, forms complete fault information by the fusion of fault information, this method can
Fabric defects significance is improved by data-driven, so as to effectively inhibit fabric background texture, fabric defects inspection is improved
All kinds of defect detections are had stronger universality, a kind of new side are provided for Automatic Detection of Fabric Defects by the accuracy rate of survey
Method.
Brief description of the drawings
Fig. 1 is the flow chart of the fabric defects dynamic testing method of view-based access control model data-driven of the present invention;
Fig. 2 is the flow chart of fabric defects notable figure construction step of the present invention;
Fig. 3 is the flow chart of the fabric defects segmentation step of the invention based on threshold value;
Fig. 4 is the flow chart of fault information fusion step of the present invention;
Fig. 5 is that the accuracy rate contrast that detection method carries out defect detection with the wavelet reconstruction method of bibliography 4 is imitated
Fruit is schemed.
Embodiment
The present invention is described in detail with reference to the accompanying drawings and detailed description.
A kind of fabric defects dynamic testing method of view-based access control model data-driven of the present invention, as shown in figure 1, it is specific according to
Lower step is implemented:
Step 1, the textile image of the rgb space of collection is converted into HSV space textile image;Specially:
The textile image that rgb space textile image is converted to HSV space, transfer process such as formula are gathered by imaging sensor
(1) shown in;
Wherein, R, G, B span are [0,255];H span is [0,360];S span be [0,
1];V span is [0,255].
Step 2, the saturation degree feature S and brightness V-arrangement of extraction image have into saturation degree characteristic pattern and brightness figure
Body is:Image color feature is described using the saturation degree S and brightness V that meet in the HSV space of human vision characteristicses, formation is knitted
The saturation degree characteristic pattern and brightness figure of object image.
Step 3, drive to form aobvious using vision data on the saturation degree characteristic pattern and brightness figure obtained in step 2
Write figure;Specific method is:Saturation degree characteristic pattern and brightness figure are subjected to small echo multilayer decomposition, the approximate spy of different resolution
Levy between subgraph, central peripheral operation, central peripheral are carried out between minutia subgraph enter between operating the difference subgraph to be formed
Row is added fusion and obtains approximate overall notable figure, the overall notable figure of level detail and the overall notable figure of vertical detail;
Step 4, on the basis of step 3, by extreme difference threshold value and fault information is split;
Step 5, by the fault information fusion split into complete fault information.
Wherein, as shown in Fig. 2 step 3 is specifically implemented according to following steps:
Saturation degree and brightness figure are carried out small echo multilayer decomposition by step 3.1 respectively;
Step 3.1.1, Selection of Wavelet Basis:
The sym2 small echos with orthogonality, near symmetry, compact sup-port are chosen to filter for Wavelet Multiresolution Decomposition;
Step 3.1.2, Decomposition order is determined:
J layers of decomposition of small echo are carried out to the characteristic pattern of normal fabric using sym2 small echos, jth layer approximation characteristic after decomposition
Figure energyLevel detail feature subgraph energyAnd vertical detail feature subgraph energyUsing formula (2),
(3) and shown in (4);
If the ratio between jth layer details and approximation characteristic subgraph energy are βj, represented using formula (5);
Work as βj-1<βjAnd βj>βj+1, then the wavelet decomposition number of plies is j+1;
Wherein, M × N is sub-graph size, fj LL(x,y)、fj LH(x,y)、fj HL(x, y) respectively represent jth yardstick it is approximate,
The wavelet coefficient of level detail, vertical detail feature subgraph at (x, y) place.
Step 3.1.3, small echo multilayer is decomposed:
If sym2 wavelet low-pass filters coefficient is hk, high-pass filter coefficient is gk, many points are realized using multilayer decomposition
Resolution is filtered, shown in specific formula (6):
Wherein:Represent that j layers are decomposed approximation characteristic respectively
Figure, level detail feature subgraph, the wavelet coefficient at vertical detail feature subgraph (x, y) place,Represent respectively j-1 layer decomposition approximation characteristic subgraph,
Level detail feature subgraph, vertical detail feature subgraph (k1,k2) place wavelet coefficient;
So as to obtain j layers and decompose approximate subgraphLevel detail subgraphVertical detail subgraphWith diagonal subgraph
Step 3.2, central peripheral operation is carried out;
To the approximation characteristic subgraph obtained in step 3.1Between, level detail feature subgraphIt
Between and vertical detail feature subgraphBetween carry out difference operation, i.e., central peripheral operate, obtain fabric defects figure
The approximate difference feature subgraph of pictureLevel detail Differential Characteristics subgraphWith vertical detail Differential Characteristics subgraphInstitute
Using central peripheral operation such as following formula (7), (8) and (9);
Wherein, central peripheral is operated between Θ is characterized subgraph, and yardstick centered on c, s=c+ δ represent periphery yardstick, δ tables
Show that central peripheral yardstick is poor;
Step 3.3, Differential Characteristics subgraph fusion is carried out;
Respectively to the approximate difference feature subgraph of the textile image obtained in step 3.2Level detail Differential Characteristics
FigureWith vertical detail Differential Characteristics subgraphIt is normalized, respectively obtains, the approximate difference after normalized
Feature subgraphLevel detail Differential Characteristics subgraphVertical detail Differential Characteristics subgraph
Specific processing is using formula such as shown in (10), (11) and (12);
To the approximate difference feature subgraph after normalizedBetween, level detail Differential Characteristics subgraphBetween,
Vertical detail Differential Characteristics subgraphBetween, it is respectively adopted and is merged and obtained respectively using formula (13), (14) and (15)
Approximation characteristic notable figure N (fLL), level detail characteristic remarkable picture N (fLH) and vertical detail characteristic remarkable picture N (fHL);
Wherein, n is the poor maximum of central peripheral yardstick;
Characteristic remarkable picture is fused to overall notable figure, it is shown using formula such as (16), (17), (18):
Wherein, N1(fLL) and N2(fLL) approximate saturation degree feature and brightness notable figure, N are represented respectively1(fLH) and N2
(fLH) level detail saturation degree and brightness notable figure, N are represented respectively1(fHL) and N2(fHL) respectively table vertical detail show saturation
Degree and brightness notable figure;WithRepresent that approximate overall notable figure, level detail are overall respectively
Notable figure and the overall notable figure of vertical detail.
Wherein, as shown in figure 3, step 4 is specifically implemented according to following steps:
Step 4.1, according to formula (19) to the approximate overall notable figure that is acquired in step 3Level detail is whole
Body notable figureWith the overall notable figure of vertical detailMiddle determination segmentation threshold Tz;
Step 4.2, image segmentation is carried out according to formula (20), defect regions is classified as above or equal to the part of threshold valuePart less than threshold value is classified as background areaIt is approximate overall notable so as to obtain respectively
Defect regions in figureDefect regions in level detail entirety notable figureAnd vertical detail
Defect regions in overall notable figure
Wherein,Overall notable figure is represented, z represents overall significantly graph type, LL, LH, HL are taken respectively;WithRepresent that the vector sum minimum that maximum is constituted in overall notable figure each column is constituted respectively
Vector;WithRepresent respectively by the overall notable figure vector sum pole that maximum is constituted in often capable
The vector of small value composition;Mean { } represents mean operation;min[thz1,thz2] represent thz1, thz2In minimum;Defect regions in different types of overall notable figure are represented, i.e.,Represent approximate overall notable figure
In split defect regions,The defect regions split in the overall notable figure of level detail are represented,Represent the defect regions split in the overall notable figure of vertical detail.
Wherein, as shown in figure 4, step 5 is specifically implemented according to following steps:By in step 4 by formula (19) and (20)
The fabric defects approximate information splitFabric defects level detail informationAnd fabric
Fault vertical detail informationIt is added together, complete fabric defects [R] is formed, using formula such as (21) institute
Show:
Wherein, [R] is complete fabric defects,Represent to be added.
On the same hardware platform, using inventive algorithm and wavelet reconstruction algorithm (Tsai DM and Hsiao
B.Automatic surface inspection using wavelet reconstruction[J].Pattern
Recognition,2001,34(6):1285-1305.) the comparison figure of accuracy rate during progress defect detection, as seen from Figure 5,
Using all kinds of defect detection accuracys rate of wavelet reconstruction algorithm in 87% -93% change, excursion is big, shows the algorithm not
Only Detection accuracy is low, and all kinds of defect detections do not possess universality, therefore can not meet the need of industry spot on-line checking
Will;All kinds of defect detection accuracys rate of inventive algorithm change between 95% -98%, and excursion is small, illustrates the algorithm not
Only accuracy rate is high, and all kinds of defect detections have universality, shows the need for being adapted to industry spot on-line checking;Both inspections
Survey the difference of result main reason is that, inventive algorithm is to drive fabric defects information with vision data, is effectively increased each
The significance of class fault, has been improved particularly the significance of small fault information, so as to improve Detection accuracy.
The present invention principle be:When in face of complex scene, the notice of the mankind is always promptly notable by a few
Visual object is attracted, and carries out priority treatment to these visual objects;Then, vision system is by increasing notice to the visual field
In remaining part scan for, the low detection target of such significance can be also noted, here it is the mankind are under data-driven
Vision significance mechanism.It is therefore believed that human visual system improves the standard of detection exactly under vision data driving
True rate;If the vision significance mechanism that human data is driven is applied to fabric defects detection process, it is possible to which satisfaction is knitted
The accuracy of thing defect detection, adapts to miscellaneous all kinds of defect detections, fabric defects detection is had universality.
The advantage of the invention is that:
(1) saturation degree is extracted by textile image and brightness is used as master data, then decomposed using small echo multilayer,
The drivings such as central peripheral operation, the fusion of signature differential subgraph build notable figure, it is to avoid the method detection such as wavelet reconstruction target is not
The problem of Detection accuracy caused by significantly is low.
(2) when fabric defects is split, using notable figure ranks extreme difference minimum mean as segmentation threshold, it effectively prevent big
Law etc. splits the inaccurate problem of fault.
Claims (5)
1. a kind of fabric defects dynamic testing method of view-based access control model data-driven, it is characterised in that specifically according to following steps
Implement:
Step 1, the textile image of the rgb space of collection is converted into HSV space textile image;
Step 2, the saturation degree feature S and brightness V-arrangement of image are extracted into saturation degree characteristic pattern and brightness figure;
Step 3, drive to be formed significantly using vision data on the saturation degree characteristic pattern and brightness figure obtained in step 2
Figure;Saturation degree characteristic pattern and brightness figure are carried out between small echo multilayer decomposition, the approximation characteristic subgraph of different resolution, carefully
Carry out central peripheral operation, central peripheral between section feature subgraph and operate to carry out between the difference subgraph to be formed being added fusion to obtain
Approximate entirety notable figure, the overall notable figure of level detail and the overall notable figure of vertical detail;
Step 4, on the basis of step 3, by extreme difference threshold value and fault information is split;
Step 4.1, according to formula (19) to the approximate overall notable figure that is acquired in step 3Level detail integrally shows
Write figureWith the overall notable figure of vertical detailMiddle determination segmentation threshold Tz;
Step 4.2, image segmentation is carried out according to formula (20), defect regions is classified as above or equal to the part of threshold valuePart less than threshold value is classified as background areaIt is approximate overall notable so as to obtain respectively
Defect regions in figureDefect regions in level detail entirety notable figureAnd vertical detail
Defect regions in overall notable figure
Wherein,Overall notable figure is represented, z represents overall significantly graph type, LL, LH, HL are taken respectively;WithThe vector that the vector sum minimum that maximum is constituted in overall notable figure each column is constituted is represented respectively;WithRepresent to be made up of the overall notable figure vector sum minimum that maximum is constituted in often capable respectively
Vector;Mean { } represents mean operation;min[thz1,thz2] represent thz1, thz2In minimum;Represent not
Defect regions in the overall notable figure of same type, i.e.,Represent to split fault area in approximate overall notable figure
Domain,The defect regions split in the overall notable figure of level detail are represented,Represent vertical
The defect regions split in details entirety notable figure,
Step 5, by the fault information fusion split into complete fault information.
2. the fabric defects dynamic testing method of view-based access control model data-driven according to claim 1, it is characterised in that described
Step 1 is specially:
The textile image that rgb space textile image is converted to HSV space, transfer process such as formula (1) are gathered by imaging sensor
It is shown;
Wherein, R, G, B span are [0,255];H span is [0,360];S span is [0,1];V
Span be [0,255].
3. the fabric defects dynamic testing method of view-based access control model data-driven according to claim 1, described step 2 is specific
For:
Image color feature is described using the saturation degree S and brightness V that meet in the HSV space of human vision characteristicses, formation is knitted
The saturation degree characteristic pattern and brightness figure of object image.
4. the fabric defects dynamic testing method of view-based access control model data-driven according to claim 1, it is characterised in that
Specifically implement according to following steps:
Saturation degree and brightness figure are carried out small echo multilayer decomposition by step 3.1 respectively;
Step 3.1.1, Selection of Wavelet Basis:
The sym2 small echos with orthogonality, near symmetry, compact sup-port are chosen to filter for Wavelet Multiresolution Decomposition;
Step 3.1.2, Decomposition order is determined:
J layers of decomposition of small echo are carried out to the characteristic pattern of normal fabric using sym2 small echos, the jth layer approximation characteristic subgraph energy after decomposition
AmountLevel detail feature subgraph energyAnd vertical detail feature subgraph energyUsing formula (2), (3) and (4)
It is shown;
If the ratio between jth layer details and approximation characteristic subgraph energy are βj, represented using formula (5);
Work as βj-1<βjAnd βj>βj+1, then the wavelet decomposition number of plies is j+1;
Wherein, M × N is sub-graph size, fj LL(x,y)、fj LH(x,y)、fj HL(x, y) represents approximate, the level of jth yardstick respectively
The wavelet coefficient of details, vertical detail feature subgraph at (x, y) place;
Step 3.1.3, small echo multilayer is decomposed:
If sym2 wavelet low-pass filters coefficient is hk, high-pass filter coefficient is gk, multiresolution is realized using multilayer decomposition
Filtering, it is specific as shown in formula (6):
Wherein,Represent that j layers are decomposed approximation characteristic respectively
Figure, level detail feature subgraph, the wavelet coefficient at vertical detail feature subgraph (x, y) place,Represent respectively j-1 layer decomposition approximation characteristic subgraph,
Level detail feature subgraph, vertical detail feature subgraph (k1,k2) place wavelet coefficient;
Approximate subgraph is decomposed so as to obtain j layersLevel detail subgraphVertical detail subgraph
With diagonal subgraph
Step 3.2, central peripheral operation is carried out;
Specifically, to the approximation characteristic subgraph obtained in step 3.1Between, level detail feature subgraphIt
Between and vertical detail feature subgraphBetween carry out difference operation, i.e., central peripheral operate, obtain fabric defects figure
The approximate difference feature subgraph of pictureLevel detail Differential Characteristics subgraphWith vertical detail Differential Characteristics subgraphAdopted
With central peripheral operation such as following formula (7), (8) and (9);
Wherein, central peripheral is operated between Θ is characterized subgraph, yardstick centered on c, and s=c+ δ represent periphery yardstick, during δ is represented
Entreat periphery yardstick poor;
Step 3.3, Differential Characteristics subgraph fusion is carried out;
Specifically, respectively to the approximate difference feature subgraph of the textile image obtained in step 3.2Level detail Differential Characteristics
SubgraphWith vertical detail Differential Characteristics subgraphIt is normalized, respectively obtains, it is approximate poor after normalized
Dtex levies subgraphLevel detail Differential Characteristics subgraphVertical detail Differential Characteristics subgraph
Specific processing is using formula such as shown in (10), (11) and (12);
To the approximate difference feature subgraph after normalizedBetween, level detail Differential Characteristics subgraphBetween, it is vertical thin
Save Differential Characteristics subgraphBetween, it is respectively adopted and is merged using formula (13), (14) and (15) and obtain approximation characteristic respectively
Notable figure N (fLL), level detail characteristic remarkable picture N (fLH) and vertical detail characteristic remarkable picture N (fHL);
Wherein, n is the poor maximum of central peripheral yardstick;
Characteristic remarkable picture is fused to overall notable figure, it is shown using formula such as (16), (17), (18):
Wherein, N1(fLL) and N2(fLL) approximate saturation degree feature and brightness notable figure, N are represented respectively1(fLH) and N2(fLH)
Level detail saturation degree and brightness notable figure, N are represented respectively1(fHL) and N2(fHL) respectively table vertical detail show saturation degree and
Brightness notable figure;WithRepresent that approximate overall notable figure, level detail are integrally notable respectively
Figure and the overall notable figure of vertical detail.
5. the fabric defects dynamic testing method of view-based access control model data-driven according to claim 1, it is characterised in that described
Step 5 is specially:
By the fabric defects approximate information split by formula (19) and (20) in step 4Fabric defects
Level detail informationWith fabric defects vertical detail informationIt is added together, formed completely
Fabric defects [R], using formula such as shown in (21):
Wherein, [R] is complete fabric defects,Represent to be added.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410334091.2A CN104199823B (en) | 2014-07-14 | 2014-07-14 | A kind of fabric defects dynamic testing method of view-based access control model data-driven |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410334091.2A CN104199823B (en) | 2014-07-14 | 2014-07-14 | A kind of fabric defects dynamic testing method of view-based access control model data-driven |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104199823A CN104199823A (en) | 2014-12-10 |
CN104199823B true CN104199823B (en) | 2017-09-01 |
Family
ID=52085116
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410334091.2A Active CN104199823B (en) | 2014-07-14 | 2014-07-14 | A kind of fabric defects dynamic testing method of view-based access control model data-driven |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104199823B (en) |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105931243B (en) * | 2016-04-26 | 2018-07-20 | 江南大学 | It is a kind of based on the fabric defect detection method for singly drilling wavelet analysis |
CN105931246A (en) * | 2016-05-05 | 2016-09-07 | 东华大学 | Fabric flaw detection method based on wavelet transformation and genetic algorithm |
CN106846396B (en) * | 2017-01-04 | 2019-08-20 | 西安工程大学 | The fabric pilling grade evaluation method of view-based access control model attention mechanism |
CN107132235B (en) * | 2017-06-21 | 2019-07-05 | 江南大学 | Online fabric defect detection method |
CN108399614B (en) * | 2018-01-17 | 2020-12-22 | 华南理工大学 | Fabric defect detection method based on non-sampling wavelet and Gumbel distribution |
CN109063781B (en) * | 2018-08-14 | 2021-12-03 | 浙江理工大学 | Design method of fuzzy image fabric imitating natural color function and form |
CN109410192B (en) * | 2018-10-18 | 2020-11-03 | 首都师范大学 | Fabric defect detection method and device based on multi-texture grading fusion |
CN112950594B (en) * | 2021-03-08 | 2023-06-23 | 北京理工大学 | Method, device and storage medium for detecting surface defects of product |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102331425A (en) * | 2011-06-28 | 2012-01-25 | 合肥工业大学 | Textile defect detection method based on defect enhancement |
CN103729842A (en) * | 2013-12-20 | 2014-04-16 | 中原工学院 | Fabric defect detection method based on local statistical characteristics and overall significance analysis |
-
2014
- 2014-07-14 CN CN201410334091.2A patent/CN104199823B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102331425A (en) * | 2011-06-28 | 2012-01-25 | 合肥工业大学 | Textile defect detection method based on defect enhancement |
CN103729842A (en) * | 2013-12-20 | 2014-04-16 | 中原工学院 | Fabric defect detection method based on local statistical characteristics and overall significance analysis |
Non-Patent Citations (6)
Title |
---|
Fabric Defect Detection Based on Fusion Technology of Multiple Algorithm;Shengqi Guan;《2010 2nd International Conference on Signal Processing Systems》;20101231;553-557 * |
从RGB到HSV色彩空间转换公式的修正;石美红等;《纺织高校基础科学学报》;20080930;第352页第1段,公式(1) * |
基于小波分解的织物疵点检测;管声启等;《昆明理工大学学报(理工版)》;20090228;48-51,103 * |
基于小波静态分解的离散小疵点检测;管声启等;《天津工业大学学报》;20101031;73-76 * |
基于视觉显著性的平纹织物疵点检测;管声启等;《纺织学报》;20140430;第57页第二栏倒数第1段、图1,第58页,第59页,第60页第一栏第1-2段 * |
新的基于图像显著性区域特征的织物疵点检测算法;赵波等;《计算机应用》;20120601;第1575页第1栏第2段、第二栏倒数第2段,图1 * |
Also Published As
Publication number | Publication date |
---|---|
CN104199823A (en) | 2014-12-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104199823B (en) | A kind of fabric defects dynamic testing method of view-based access control model data-driven | |
CN106250870B (en) | A kind of pedestrian's recognition methods again of joint part and global similarity measurement study | |
CN108763606A (en) | A kind of floor plan element extraction method and system based on machine vision | |
CN105809693B (en) | SAR image registration method based on deep neural network | |
CN108648168A (en) | IC wafer surface defects detection methods | |
CN109409190A (en) | Pedestrian detection method based on histogram of gradients and Canny edge detector | |
CN107389701A (en) | A kind of PCB visual defects automatic checkout system and method based on image | |
CN106296638A (en) | Significance information acquisition device and significance information acquisition method | |
CN108090888A (en) | The infrared image of view-based access control model attention model and the fusion detection method of visible images | |
CN106447708A (en) | OCT eye fundus image data registration method | |
CN106815819B (en) | More strategy grain worm visible detection methods | |
CN103729842A (en) | Fabric defect detection method based on local statistical characteristics and overall significance analysis | |
CN106373123A (en) | K_tSL central clustering algorithm-based industrial component surface defect detection method | |
CN111724372A (en) | Method, terminal and storage medium for detecting cloth defects based on antagonistic neural network | |
CN106355579A (en) | Defect detecting method of cigarette carton surface wrinkles | |
CN109886944B (en) | A kind of white matter high signal intensity detection and localization method based on multichannel chromatogram | |
CN111460953B (en) | Electrocardiosignal classification method based on self-adaptive learning of countermeasure domain | |
CN104778692A (en) | Fabric defect detection method based on sparse representation coefficient optimization | |
Shi et al. | A faster-rcnn based chemical fiber paper tube defect detection method | |
CN111080574A (en) | Fabric defect detection method based on information entropy and visual attention mechanism | |
CN109559273A (en) | A kind of quick joining method towards vehicle base map picture | |
CN102496032B (en) | Electrical equipment X ray digital image processing algorithm support system | |
CN105930793A (en) | Human body detection method based on SAE characteristic visual learning | |
CN106407975B (en) | Multiple dimensioned layering object detection method based on space-optical spectrum structural constraint | |
Guan et al. | Defect detection and classification for plain woven fabric based on deep learning |
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 |