CN110021023A - A kind of electronics cloth defect segmentation method - Google Patents
A kind of electronics cloth defect segmentation method Download PDFInfo
- Publication number
- CN110021023A CN110021023A CN201910164685.6A CN201910164685A CN110021023A CN 110021023 A CN110021023 A CN 110021023A CN 201910164685 A CN201910164685 A CN 201910164685A CN 110021023 A CN110021023 A CN 110021023A
- Authority
- CN
- China
- Prior art keywords
- image
- point
- electronics cloth
- cluster centre
- classification
- 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.)
- Pending
Links
- 239000004744 fabric Substances 0.000 title claims abstract description 71
- 230000007547 defect Effects 0.000 title claims abstract description 62
- 238000000034 method Methods 0.000 title claims abstract description 45
- 230000011218 segmentation Effects 0.000 title claims abstract description 33
- 238000001914 filtration Methods 0.000 claims abstract description 10
- 238000006243 chemical reaction Methods 0.000 claims abstract description 8
- 238000012545 processing Methods 0.000 claims abstract description 7
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000009466 transformation Effects 0.000 claims 2
- 241000638935 Senecio crassissimus Species 0.000 claims 1
- 230000002401 inhibitory effect Effects 0.000 abstract description 3
- 230000000717 retained effect Effects 0.000 abstract description 2
- 238000001514 detection method Methods 0.000 description 14
- 230000000694 effects Effects 0.000 description 5
- 238000004519 manufacturing process Methods 0.000 description 3
- 239000004753 textile Substances 0.000 description 3
- 235000013399 edible fruits Nutrition 0.000 description 2
- 239000003365 glass fiber Substances 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 239000012779 reinforcing material Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06T7/0008—Industrial image inspection checking presence/absence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30124—Fabrics; Textile; Paper
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Treatment Of Fiber Materials (AREA)
Abstract
The invention discloses a kind of electronics cloth defect segmentation methods, the specific steps are as follows: electronics cloth image to be detected is changed to gray level image first；Secondly ButterWorth filtering processing is carried out to gray level image；Filtered image is clustered using K-means algorithm again；The pixel of clustered image is marked by the classification of cluster finally, binary conversion treatment is carried out according to label result, fabric defects can be partitioned into.Method of the invention is handled electronics cloth image using ButterWorth low-pass filter, can be retained to greatest extent the detailed information of fault while inhibiting background texture, be enhanced the contrast of background and defect regions.Meanwhile filtered image is clustered using K-means algorithm, electronics cloth image may finally be polymerized to as inhomogeneity, and then accurate positioning, the accurate segmentation of electronics cloth fault are reached as a result, carry out binarization segmentation to image according to label.
Description
Technical field
The invention belongs to fabric defects detection technical fields in textile industry, and in particular to a kind of electronics cloth defect segmentation side
Method.
Background technique
Glass fibre electronics cloth, also known as electronics cloth are a kind of industrial goods made of being weaved as glass fiber yarn, as
Reinforcing material is widely used in the fields such as space flight and aviation, machine components.In industrial processes, fault is to influence its price
The key link that an important factor for evaluating with grade and product quality are checked on.Currently, the production of most of electronics cloths is looked forward to both at home and abroad
Industry relies primarily on manually to carry out electronics cloth defect detection.Artificial detection is influenced bigger, such as prolonged view by subjective factor
Feel fatigue, working environment etc., fault recall rate only has 70% or so, and speed is slow, accuracy rate is low, omission factor is high in the presence of detecting
The problems such as, it is unable to satisfy the actual production demand of enterprise, therefore the automatic detection for studying electronics cloth fault has good answer
Use prospect.
In recent years, defect detection has become the research hotspot of Digital Image Processing and field of machine vision.Currently, both at home and abroad
Scholar is roughly divided into the defect detection method of fabric: Statistics-Based Method, the method based on frequency spectrum and the side based on model
Method.Tsai etc. application DFT and Hough transform detection the apparent fabric defects effect of texture it is preferable, but to background frequency ingredient
Similar defect regions, detection effect are poor.Liu Zhoufeng etc. combines partial statistics characteristic and context entirety significance analysis to obtain
Fabric notable figure, using iteration Optimum threshold segmentation fabric defects, real-time is poor.Liapis etc. is existed using wavelet transform
Gray feature is extracted in the channel L, extracts color characteristic in the channel a, b, preferable to color image defect detection effect, handles the time
It is long.Chen etc. uses multiscale matched filtering algorithm on textile image, can detect various sizes of fault, but computationally intensive.
Based on above the study found that there are still following problems: (1) most of defect detection sides in fabric defects detection at present
Method carries out algorithm design both for a certain particular web and defect type, and versatility is poor；(2) more with textile technology
Newly, the fault occurred in electronics cloth is smaller and smaller, and the color of defect regions and background, texture difference become smaller, and leads to existing inspection
Survey method is to fault position inaccurate.
Summary of the invention
The object of the present invention is to provide a kind of electronics cloth defect segmentation methods, solve detection method pair in the prior art
Fault position inaccurate in electronics cloth, the problem of cannot accurately dividing.
The technical scheme adopted by the invention is that a kind of electronics cloth defect segmentation method, the specific steps are as follows:
Electronics cloth image to be detected is converted to gray level image by step 1；
Step 2 carries out ButterWorth filtering processing to gray level image, obtains filtered image；
Step 3 clusters the pixel in filtered image using K-means algorithm, finally by all pixels
Point is clustered into different classifications；
The all pixels point in each classification obtained in step 3 is marked in step 4, then to label after
Image carries out binary conversion treatment, can be partitioned into electronics cloth defect regions.
The features of the present invention also characterized in that
The detailed process of step 2 are as follows:
Step 2.1 enables gray level image obtained in step 1 be f (x, y), and the size of gray level image is M × N, by grayscale image
Picture f (x, y) obtains image F (u, v) through Fourier transform, and the background of gray level image f (x, y) is made to be located at the low frequency component of frequency domain
In, noise and image detail part in gray level image f (x, y) are located in the high fdrequency component of frequency domain；
Step 2.2 after the image F (u, v) through Fourier transform is carried out process of convolution, then carries out inverse fourier transform and obtains
Obtain filtered image g (x, y).
The formula of Fourier transform in step 2.1 are as follows:
Wherein, the pixel size of M, N representative image, μ, ν represent discrete variable, and μ=0,1,2..., M-1, ν=0,
1,2...,N-1。
The formula of inverse fourier transform in step 2.2 are as follows:
G (x, y)=F-1{H(u,v)F(u,v)}
Wherein, H (u, v) represents filter, is expressed asN is order, takes positive integer, is used
To control the rate of decay, D0For cutoff frequency, D (u, v) is distance of the point (u, v) away from origin.
The detailed process of step 3 are as follows:
K point is taken in step 3.1, random image g (x, y) after the filtering, using the pixel value of this k point as at the beginning of k
Beginning cluster centre point obtains initial cluster center point set C={ c1,c2,...,ck}；
Wherein, C indicates the set of all initial cluster center points, and k indicates the number of cluster centre, c1Indicate that first is gathered
Class central point, c2Indicate second cluster centre point, and so on, ckIndicate k-th of cluster centre point；
Step 3.2, calculate image g (x, y) on each pixel pixel value and C={ c1,c2,...,ckIn it is every
The distance between one cluster centre point carries out all pixels point in filtered image g (x, y) according to Euclidean distance
Classification constitutes k set, each set represents a classification；
Step 3.3 averages respectively to the pixel value of all pixels point in each classification, flat by obtained k
The mean value cluster centre point new as k；
Step 3.4 repeats two steps of step 3.2 and step 3.3 when cluster centre is restrained, i.e., cluster centre is restrained
K cluster centre point and the obtained k cluster centre point offset summation of last iteration less than 0.0001, finally by image
All pixels point in g (x, y) is clustered into k classification.
The calculation formula of Euclidean distance in step 3.2 are as follows:
Wherein, g (x, y) indicates pixel, ckIndicate k-th of cluster centre point.
The calculation formula of offset summation in step 3.4 are as follows:
Wherein, E is the offset summation of all cluster centres, and j indicates the number of iterations, ci,jIndicate i-th of classification in jth
The cluster centre obtained at the end of secondary iteration, ci,j-1Indicate that i-th of classification terminates in the cluster that formula obtains in -1 iteration of jth
The heart.
Method is marked in filtered image category in step 4 are as follows: pixel in will be different classes of is with different
Digital representation, all pixels point in the same classification are designated with like numbers.
The number of k is 2,3,4 or 5.
The invention has the advantages that a kind of electronics cloth defect segmentation method, uses ButterWorth low-pass filter pair
Electronics cloth image is handled, and can retain to greatest extent the detailed information of fault, enhancing back while inhibiting background texture
The contrast of scape and defect regions.Meanwhile filtered image is clustered using K-means algorithm, it may finally will be electric
Sub- cloth image is polymerized to as inhomogeneity, and then reaches the essence of electronics cloth fault as a result, carry out binarization segmentation to image according to label
Determine position, accurate segmentation.
Detailed description of the invention
Fig. 1 is a kind of flow chart of electronics cloth defect segmentation method of the present invention；
Fig. 2 (a) is electronics cloth spot defect map to be detected in a kind of electronics cloth defect segmentation embodiment of the method for the present invention
Picture, Fig. 2 (b) are that Fig. 2 (a) carries out the filtered result figure of ButterWorth；
Fig. 3 (a) is the Fourier transform spectrogram of Fig. 2 (a), and Fig. 3 (b) is the Fourier transform spectrogram of Fig. 2 (b)；
Fig. 4 (a) is the grey level histogram of Fig. 2 (a), and Fig. 4 (b) is the grey level histogram of Fig. 2 (b)；
Fig. 5 is the binary segmentation result figure of Fig. 2 (a)；
Fig. 6 (a) is electronics cloth staplings defect map to be detected in a kind of electronics cloth defect segmentation embodiment of the method for the present invention
Picture, Fig. 6 (b) are the image obtained after Fig. 6 (a) ButterWorth is filtered, and Fig. 6 (c) is the binary segmentation knot of Fig. 6 (a)
Fruit；
Fig. 7 (a) is that electronics cloth to be detected is disconnected through defect map in a kind of electronics cloth defect segmentation embodiment of the method for the present invention
Picture, Fig. 7 (b) are the image obtained after Fig. 7 (a) ButterWorth is filtered, and Fig. 7 (c) is the binary segmentation knot of Fig. 7 (a)
Fruit.
Specific embodiment
The following describes the present invention in detail with reference to the accompanying drawings and specific embodiments.
A kind of electronics cloth defect segmentation method of the present invention, as shown in Figure 1, the specific steps are as follows:
Electronics cloth image to be detected is converted to gray level image by step 1；
Step 2 carries out ButterWorth filtering processing to gray level image, obtains filtered image；
Detailed process are as follows:
Step 2.1 enables gray level image obtained in step 1 be f (x, y), and the size of gray level image is M × N, by grayscale image
Picture f (x, y) obtains image F (u, v) through Fourier transform, and the background of gray level image f (x, y) is made to be located at the low frequency component of frequency domain
In, noise and image detail part in gray level image f (x, y) are located in the high fdrequency component of frequency domain；
The formula of Fourier transform are as follows:
Wherein, the pixel size of M, N representative image, μ, ν represent discrete variable, and μ=0,1,2..., M-1, ν=0,
1,2...,N-1。
Step 2.2 after image F (u, v) is carried out process of convolution in frequency domain, then carries out inverse fourier transform and is filtered
Image g (x, y) afterwards.
The formula of inverse fourier transform are as follows:
G (x, y)=F-1{H(u,v)F(u,v)}
Wherein, H (u, v) represents filter, is expressed asN is order, takes positive integer, is used to
Control the rate of decay, D0For cutoff frequency, D (u, v) is distance of the point (u, v) away from origin.
Step 3 clusters the pixel in filtered image using K-means algorithm, finally by the institute in image
Different classifications is polymerized to pixel；
Detailed process are as follows:
Step 3.1 takes k point in electronics cloth at random, using the pixel value of this k point as k initial cluster center point,
Obtain initial cluster center point set C={ c1,c2,...,ck}；
Wherein, C indicates the set of all cluster centres, and k indicates the number of cluster centre, c1Indicate first cluster centre
Point, c2Indicate second cluster centre point, and so on, ckIndicate k-th of cluster centre point, the number of k can for 2,3,4 or
5；
Step 3.2, calculate image g (x, y) each pixel pixel value and C={ c1,c2,...,ckIn it is each
The distance between the pixel value of a cluster centre point, according to Euclidean distance by image g (x, y) apart from each cluster in
Heart point assigns to one kind apart from nearest pixel, constitutes k set, each set represents a classification；
The calculation formula of Euclidean distance are as follows:
Wherein, g (x, y) indicates pixel, ckIndicate k-th of cluster centre point.
Step 3.3 averages respectively to the pixel value of all pixels point of each classification, using k average value as k
A new cluster centre；
Step 3.4 repeats two steps of step 3.2 and step 3.3 when cluster centre is restrained, i.e., cluster centre is restrained
K cluster centre point pixel pixel value and the obtained offset of the pixel value of k cluster centre point of last iteration
Summation is measured less than 0.0001, all pixels point in image g (x, y) is finally clustered into k classification.
The calculation formula of offset summation are as follows:
Wherein, E is the offset summation of all cluster centres, and j indicates the number of iterations, ci,jIndicate i-th of classification in jth
The cluster centre obtained at the end of secondary iteration, ci,j-1Indicate that i-th of classification terminates in the cluster that formula obtains in -1 iteration of jth
The heart.
The pixel category in filtered image is marked in step 4, the cluster result obtained according to step 3,
Then binary conversion treatment is carried out to the image after label, electronics cloth defect regions can be partitioned into.
Labeling method are as follows: the different digital representation of pixel in will be different classes of, all pictures in the same classification
Vegetarian refreshments is designated with like numbers.
Method of the invention is described in detail by taking the spot defect of electronics cloth as an example below:
A kind of electronics cloth defect detection algorithm based on ButterWorth filtering and K-means algorithm, specifically according to following
Step is implemented:
Electronics cloth picture size to be detected is zoomed to 256 pixel *, 256 pixel, and color image is converted by step 1
For gray level image, image is unified for .jpg format.
Step 2, electronics cloth background texture information and defect regions have certain similitude on color characteristic, if directly
Clustering processing, texture information meeting severe jamming defect detection effect, Bu Nengyou are carried out to electronics cloth image using K-means algorithm
Effect distinguishes fault and background area.Therefore it needs to carry out at ButterWorth filtering the electronics cloth image obtained after step 1
Reason；
Specific steps are as follows: enabling gray level image obtained in step 1 is f (x, y), and image size is 256 pixel *, 256 pixel,
Gray level image f (x, y) is obtained into image F (u, v) through Fourier transform, the background of gray level image f (x, y) is made to be located at the low of frequency domain
In frequency component, noise and image detail part in gray level image f (x, y) are located in the high fdrequency component of frequency domain；
Fourier transform formula is formula (1):
Wherein, the pixel size of M, N representative image, μ, ν represent discrete variable, and μ=0,1,2..., M-1, ν=0,
1,2...,N-1。
After image F (u, v) is carried out process of convolution in frequency domain, then obtain such as the inverse fourier transform of formula (2)
Filtered image g (x, y), can achieve the purpose of smoothed image by inverse fourier transform.
G (x, y)=F-1{H(u,v)F(u,v)}(2)
Wherein, H (u, v) represents filter, and in all multi-filters, ButerWorth low-pass filter " ring " phenomenon is micro-
It is weak, image detail information can be enhanced, so using ButerWorth low-pass filter.ButterWorth low-pass filter
(BLPF) transfer function H (u, v) is as shown in formula (3).
Wherein, n is order, takes positive integer, for controlling the rate of decay, D0For cutoff frequency, D (u, v) be point (u, v) away from
The distance of origin；
For electronics cloth spot defect sample, it is 2 that order n is chosen in the present embodiment, and cutoff frequency D0 is 50.
Step 3 clusters filtered image using K-means algorithm, the basic thought of K-means algorithm be with
Machine initialization gives several each and every one cluster centre points, and all pixels point in image g (x, y) is carried out according to Euclidean distance
Classification；Then the average value of each classification is recalculated by the method for average, so that it is determined that new cluster centre point；Continuous iteration until
The moving distance of cluster centre is less than 0.0001.
Specific step is as follows:
For the electronics cloth spot defect sample chosen in the present embodiment, 2 are taken in electronics cloth at random in the present embodiment
Point obtains initial cluster center point set C={ c using the pixel value of this 2 points as 2 initial cluster center points1,c2}；
Wherein, C indicates the set of all cluster centres, c1Indicate first cluster centre point, c2It indicates in second cluster
Heart point, random starting values are c in the present embodiment1=252, c2=254.
Then, the pixel value and C={ c of each pixel of image g (x, y) are calculated1,c2In each cluster centre
The distance between the pixel value of point, according to Euclidean distance by image g (x, y) apart from each cluster centre point distance most
Close pixel assigns to one kind, constitutes 2 set, each set represents a classification；
If electronics cloth spot defect image is as shown below, number represents the pixel value of the point in figure, with (1,1) coordinate
Pixel value for, shown in the calculation formula of Euclidean distance such as formula (4):
| 120 | 123 | 126 |
| 150 | 153 | 156 |
| 250 | 253 | 256 |
D1=| | 120-252 | |=132
D2=| | 120-254 | |=134
D=min { D1,D2}=min { 132,134 }=132
Then the distance of c1 is less than c2 in the pixel value distance cluster of coordinate (1,1) point, therefore the pixel is grouped into c1's
In classification, and so on, traversal calculates each of image g (x, y) pixel at a distance from two cluster centres, by them
It is grouped into nearest class.
It averages to the pixel value of all pixels point of each classification, using average value as new cluster centre；
Finally, repetition two steps of step 3.2 and step 3.3 are restrained until cluster centre, i.e., when cluster centre is restrained
The offset summation of cluster centre and a upper cluster centre is less than 0.0001, finally by all pixels point in image g (x, y)
It is clustered into 2 classifications.
Assuming that electronics cloth spot defect sample iteration 8 times can be obtained cluster centre convergence, then as shown in formula (5), offset
The calculation formula of summation are as follows:
E=d (c1,8-c1,7)+d(c2,8-c2,7) (5)
Wherein, E is the offset summation of all cluster centres, c1,8Indicate that the 1st classification obtains at the end of the 8th iteration
The cluster centre arrived, c1,7Indicate that the 1st classification terminates the cluster centre that formula obtains in the 7th iteration；c2,8Indicate the 2nd classification
The cluster centre obtained at the end of the 8th iteration indicates that the 2nd classification terminates the cluster centre that formula obtains in the 7th iteration.
Step 4, step 3 treated image, are marked as two classifications, one type is not background, the cluster centre
It is 220；Another category is fault, which is 121；Background area is labeled as 1, defect regions are labeled as 2, according to poly-
The label result of class, which carries out binary conversion treatment, can go out fabric defects as shown in formula (6) with Accurate Segmentation；
In formula, G (x, y) indicates that binaryzation exports image, and (x, y) is pixel coordinate, and k represents the classification of element marking；
Complete the Accurate Segmentation of fabric defects in spot defect.
In Figure of description of the invention, Fig. 2 (a) is the original image of electronics cloth to be detected in embodiment, and Fig. 2 (b) is Fig. 2
(a) image obtained through inverse fourier transform；Fig. 3 (a) is the Fourier transform spectrogram of Fig. 2 (a), and Fig. 3 (b) is Fig. 2's (b)
Fourier transform spectrogram, in an amplitude-frequency spectrogram, corresponding central point is high fdrequency component, and corresponding four angles are low frequencies point
Amount, comparison diagram 3 (a), (b) two width figure can be seen that the image spectrum figure after ButterWorth is filtered, four angles
Low frequency component be eliminated, that is, indicate in the time domain, it is suppressed that the background texture of electronics cloth image.Fig. 4 (a) is the ash of Fig. 2 (a)
Histogram is spent, Fig. 4 (b) is the grey level histogram of Fig. 2 (b), available by two width grey level histograms of comparison: ButterWorth
Bimodality is presented in image after filter process, illustrates background that filtered electronics cloth image is included and defect regions in ash
There is different in angle value, can be clustered by K-means and realize defect segmentation.
Staplings defect image and the disconnected dividing method through defect image and disconnected dividing method and spot through defect image lack
Sunken process is the same:
The 6 (a) of Figure of description are the staplings defect image of electronics cloth, and Fig. 6 (b) is Fig. 6 (a) ButterWorth filtering
The image obtained after processing, Fig. 6 (c) are the binary conversion treatment result figure of Fig. 6 (a)；
Fig. 7 (a) is the disconnected through defect image of electronics cloth, and Fig. 7 (b) is to obtain after Fig. 6 (a) ButterWorth is filtered
Image, Fig. 7 (c) be Fig. 7 (a) binary conversion treatment result figure；
By result above it can be seen that using ButterWorth low-pass filter to electronics cloth in method of the invention
Image is handled, and can be retained to greatest extent the detailed information of fault while inhibiting background texture, be enhanced background and defect
The contrast in point region.Meanwhile filtered image is clustered using K-means algorithm, it may finally be by electronics Butut
As being polymerized to as inhomogeneity, and then according to label as a result, carry out binary conversion treatment to image, reach electronics cloth fault it is accurate calmly
Position, accurate segmentation.
Claims (9)
1. a kind of electronics cloth defect segmentation method, which is characterized in that specific step is as follows:
Electronics cloth image to be detected is converted to gray level image by step 1；
Step 2 carries out ButterWorth filtering processing to gray level image, obtains filtered image；
Step 3 clusters the pixel in filtered image using K-means algorithm, finally gathers all pixels point
Class is at different classifications；
The all pixels point in each classification obtained in step 3 is marked in step 4, then to the image after label
Binary conversion treatment is carried out, electronics cloth defect regions can be partitioned into.
2. a kind of electronics cloth defect segmentation method according to claim 1, which is characterized in that the specific mistake of the step 2
Journey are as follows:
Step 2.1 enables gray level image obtained in step 1 be f (x, y), and the size of gray level image is M × N, by gray level image f
(x, y) obtains image F (u, v) through Fourier transform, is located at the background of gray level image f (x, y) in the low frequency component of frequency domain, ash
Noise and image detail part in degree image f (x, y) are located in the high fdrequency component of frequency domain；
Step 2.2 after the image F (u, v) through Fourier transform is carried out process of convolution, then carries out inverse fourier transform and is filtered
Image g (x, y) after wave.
3. a kind of electronics cloth defect segmentation method according to claim 2, which is characterized in that Fourier in the step 2.1
The formula of leaf transformation are as follows:
Wherein, the pixel size of M, N representative image, μ, ν represent discrete variable, and μ=0,1,2..., M-1, ν=0, and 1,
2...,N-1。
4. a kind of electronics cloth defect segmentation method according to claim 2 or 3, it is characterised in that Fu in the step 2.2
The formula of vertical leaf inverse transformation are as follows:
G (x, y)=F-1{H(u,v)F(u,v)}
Wherein, H (u, v) represents filter, is expressed asN is order, takes positive integer, for controlling
The rate of decay, D0For cutoff frequency, D (u, v) is distance of the point (u, v) away from origin.
5. a kind of electronics cloth defect segmentation method according to claim 1, which is characterized in that the specific mistake of the step 3
Journey are as follows:
K point is taken in step 3.1, random image g (x, y) after the filtering, it is initial poly- using the pixel value of this k point as k
Class central point obtains initial cluster center point set C={ c1,c2,...,ck}；
Wherein, C indicates the set of all initial cluster center points, and k indicates the number of cluster centre, c1It indicates in first cluster
Heart point, c2Indicate second cluster centre point, and so on, ckIndicate k-th of cluster centre point；
Step 3.2, calculate image g (x, y) on each pixel pixel value and C={ c1,c2,...,ckIn each
The distance between cluster centre point, according to Euclidean distance to all pixels point minute in filtered image g (x, y)
Class constitutes k set, each set represents a classification；
Step 3.3 averages respectively to the pixel value of all pixels point in each classification, the k average value that will be obtained
The cluster centre point new as k；
Step 3.4 repeats k of two steps of step 3.2 and step 3.3 when cluster centre is restrained, i.e., cluster centre is restrained
The k cluster centre point offset summation that a cluster centre point and last iteration obtain is less than 0.0001, finally by image g
All pixels point in (x, y) is clustered into k classification.
6. a kind of electronics cloth defect segmentation method according to claim 5, which is characterized in that the step 3.2 is Central European several
In distance calculation formula are as follows:
Wherein, g (x, y) indicates pixel, ckIndicate k-th of cluster centre point.
7. a kind of electronics cloth defect segmentation method according to claim 5, which is characterized in that deviated in the step 3.4
Measure the calculation formula of summation are as follows:
Wherein, E is the offset summation of all cluster centres, and j indicates the number of iterations, ci,jIndicate that i-th of classification changes in jth time
The cluster centre obtained at the end of generation, ci,j-1Indicate that i-th of classification terminates the cluster centre that formula obtains in -1 iteration of jth.
8. a kind of electronics cloth defect segmentation method according to claim 1, which is characterized in that after being filtered in the step 4
Image category method is marked are as follows: the different digital representation of pixel in will be different classes of, in the same classification
All pixels point be designated with like numbers.
9. a kind of electronics cloth defect segmentation method according to claim 5 or 6 or 7, which is characterized in that the number of the k
It is 2,3,4 or 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910164685.6A CN110021023A (en) | 2019-03-05 | 2019-03-05 | A kind of electronics cloth defect segmentation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910164685.6A CN110021023A (en) | 2019-03-05 | 2019-03-05 | A kind of electronics cloth defect segmentation method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110021023A true CN110021023A (en) | 2019-07-16 |
Family
ID=67189237
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910164685.6A Pending CN110021023A (en) | 2019-03-05 | 2019-03-05 | A kind of electronics cloth defect segmentation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110021023A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110751661A (en) * | 2019-10-28 | 2020-02-04 | 南京泓图人工智能技术研究院有限公司 | Clustering algorithm-based facial chloasma region automatic segmentation method |
| CN113012103A (en) * | 2021-02-07 | 2021-06-22 | 电子科技大学 | Quantitative detection method for surface defects of large-aperture telescope lens |
| CN113706521A (en) * | 2021-09-08 | 2021-11-26 | 常州市新创智能科技有限公司 | Carbon fiber surface hairiness detection method and device, storage medium and electronic equipment |
| CN113920112A (en) * | 2021-11-18 | 2022-01-11 | 杭州云图智检科技有限公司 | Fabric flaw detection method based on independent classification type feature extraction |
| CN114022415A (en) * | 2021-10-15 | 2022-02-08 | 成都博视广达科技有限责任公司 | Liquid crystal display defect detection method based on single-pixel feature clustering cluster establishment |
| CN115047070A (en) * | 2022-08-11 | 2022-09-13 | 江苏恒力化纤股份有限公司 | Fabric surface defect detection method based on friction vibration signal |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780464A (en) * | 2016-12-15 | 2017-05-31 | 东华大学 | A kind of fabric defect detection method based on improvement Threshold segmentation |
| CN109035195A (en) * | 2018-05-08 | 2018-12-18 | 武汉纺织大学 | A kind of fabric defect detection method |
-
2019
- 2019-03-05 CN CN201910164685.6A patent/CN110021023A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780464A (en) * | 2016-12-15 | 2017-05-31 | 东华大学 | A kind of fabric defect detection method based on improvement Threshold segmentation |
| CN109035195A (en) * | 2018-05-08 | 2018-12-18 | 武汉纺织大学 | A kind of fabric defect detection method |
Non-Patent Citations (4)
| Title |
|---|
| ICE星空: ""Butterworth低通滤波器"", 《HTTPS://BLOG.CSDN.NET/LYN_B/ARTICLE/DETAILS/84928805》 * |
| 张缓缓 等: ""k-means聚类算法在织物疵点检测中的应用"", 《毛纺科技》 * |
| 王为农等, "北京：中国商业出版社" * |
| 王延年等: ""一种改进的印刷丝网布疵点快速检测算法"", 《理论与方法》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110751661A (en) * | 2019-10-28 | 2020-02-04 | 南京泓图人工智能技术研究院有限公司 | Clustering algorithm-based facial chloasma region automatic segmentation method |
| CN113012103A (en) * | 2021-02-07 | 2021-06-22 | 电子科技大学 | Quantitative detection method for surface defects of large-aperture telescope lens |
| CN113706521A (en) * | 2021-09-08 | 2021-11-26 | 常州市新创智能科技有限公司 | Carbon fiber surface hairiness detection method and device, storage medium and electronic equipment |
| CN114022415A (en) * | 2021-10-15 | 2022-02-08 | 成都博视广达科技有限责任公司 | Liquid crystal display defect detection method based on single-pixel feature clustering cluster establishment |
| CN113920112A (en) * | 2021-11-18 | 2022-01-11 | 杭州云图智检科技有限公司 | Fabric flaw detection method based on independent classification type feature extraction |
| CN115047070A (en) * | 2022-08-11 | 2022-09-13 | 江苏恒力化纤股份有限公司 | Fabric surface defect detection method based on friction vibration signal |
| CN115047070B (en) * | 2022-08-11 | 2022-12-20 | 江苏恒力化纤股份有限公司 | Fabric surface defect detection method based on friction vibration signal |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110021023A (en) | A kind of electronics cloth defect segmentation method | |
| CN107292310B (en) | Visual positioning and automatic reading method for circular pointer type dial plate | |
| CN104574353B (en) | The surface defect decision method of view-based access control model conspicuousness | |
| CN110837768B (en) | Online detection and identification method for rare animal protection | |
| CN110930357B (en) | In-service steel wire rope surface defect detection method and system based on deep learning | |
| CN114723704B (en) | Textile quality evaluation method based on image processing | |
| CN107389701A (en) | A kind of PCB visual defects automatic checkout system and method based on image | |
| CN109490316A (en) | A kind of surface defects detection algorithm based on machine vision | |
| CN108648168A (en) | IC wafer surface defects detection methods | |
| CN103955940B (en) | A kind of detection method of the human body cache based on X ray backscatter images | |
| CN113554080A (en) | Non-woven fabric defect detection and classification method and system based on machine vision | |
| CN108376403B (en) | Grid colony image segmentation method based on Hough circle transformation | |
| CN107369155A (en) | A kind of cloth surface defect detection method and its system based on machine vision | |
| CN109509171A (en) | A kind of Fabric Defects Inspection detection method based on GMM and image pyramid | |
| CN109886937B (en) | Insulator defect detection method based on super-pixel segmentation image recognition | |
| CN109781737B (en) | Detection method and detection system for surface defects of hose | |
| CN115331119B (en) | Solid waste identification method | |
| CN109191430A (en) | A kind of plain color cloth defect inspection method based on Laws texture in conjunction with single classification SVM | |
| CN110288571B (en) | High-speed rail contact net insulator abnormity detection method based on image processing | |
| CN111091134A (en) | Method for identifying tissue structure of colored woven fabric based on multi-feature fusion | |
| Bhargava et al. | An adaptive method for edge preserving denoising | |
| CN115994870B (en) | Image processing method for enhancing denoising | |
| CN108537787A (en) | A kind of quality judging method of facial image | |
| CN111738931B (en) | Shadow removal algorithm for aerial image of photovoltaic array unmanned aerial vehicle | |
| CN108399614B (en) | Fabric defect detection method based on non-sampling wavelet and Gumbel distribution |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190716 |
|
| RJ01 | Rejection of invention patent application after publication |