CN108346141A - Unilateral side incidence type light guide plate defect extracting method - Google Patents
Unilateral side incidence type light guide plate defect extracting method Download PDFInfo
- Publication number
- CN108346141A CN108346141A CN201810025157.8A CN201810025157A CN108346141A CN 108346141 A CN108346141 A CN 108346141A CN 201810025157 A CN201810025157 A CN 201810025157A CN 108346141 A CN108346141 A CN 108346141A
- Authority
- CN
- China
- Prior art keywords
- image
- guide plate
- light guide
- defect
- light
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
- G06T7/0008—Industrial image inspection checking presence/absence
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- 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
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/155—Segmentation; Edge detection involving morphological operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Biochemistry (AREA)
- Signal Processing (AREA)
- Geometry (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Quality & Reliability (AREA)
- General Health & Medical Sciences (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention provides a kind of unilateral side incidence type light guide plate defect extracting method, includes the following steps:Obtain light guide plate image, extraction light conducting plate body image, obtain width M and height N, defect quickly detect, quickly detect whether containing defective, removal noise jamming, greyscale transformation, OTSU Threshold segmentations, auto-partition detection, traversal light guiding points area pixel tonal range, judge that light guide plate whether there is bright dim spot, Morphological scale-space, judge whether to weigh wounded or foreign matter, image segmentation, judge whether to lead scratch defects and extract defective work defect area;The present invention proposes that the exploitation adaptive auto-partition algorithm of light guide plate divides different detection zones, and adjust automatically detection algorithm, realize the extraction of defect automatically according to the dense degree of surface light-conductive hole;The operational efficiency height and accuracy rate of the algorithm are high, and stability and strong robustness can not only identify common defect, also have relatively high detectability for uncommon tiny flaw.
Description
Technical field
The present invention relates to a kind of defects detection algorithm, specially a kind of unilateral side incidence type light guide plate defect extracting method.
Background technology
Light guide plate (Light Guide Plate, LGP) is acrylic/PC plate material using optical grade, then with pole
The hitech materials of high reflectance and not extinction, acryl plates bottom surface laser engraving, V-type cross grid in optical grade
Engraving, UV screen printing technologies stamp light guiding points.LGP have ultra-thin, superbright, it is guide-lighting uniformly, it is energy-saving and environmental protection, no dark space, resistance to
With, be not easy the distinguishing features such as yellow, installation and maintenance are simple and fast, be widely used in the backlight, ultra-thin wide of liquid crystal display
Accuse the fields such as lamp box, therapeutic medical X-ray viewer, plate lamp ornaments lighting, the light efficiency utilization of light engineering, illuminant sign plate
It closes.In the making of the silk-screen of light guide plate, chemical etching, during laser machining and hit the manufacturings such as processing, due to raw material at
Point, the influences of the factors such as equipment service condition, processing technology and operation, surface inevitably bright spot,
The manufacturing deficiencies such as leak source, stain, wire side ink, line scratch, mirror point wound, shadow.If liquid crystal display backlight,
Therapeutic medical X-ray viewer etc. is upper to use defective light guide plate, can influence its uniformly light-emitting, reduce the service efficiency of light, more
For the serious is the eyesights to people to damage.Moreover, with equipment high-performance, high-precision development, to the material of light guide plate
Expect that the requirement of characteristic, surface quality, planform, reliability etc. is also higher and higher.It therefore, must before light guide plate manufacture
Quality testing must be carried out to it, defective light guide plate will be contained and eliminated.
Due to light guide plate defect have type is various, the form of expression is different, edge blurry or without limbus, comparison
Spend the features such as low, domestic light guide plate manufacturing enterprise mainly with the testing staff by professional training various angles beat strong light come
Artificial detection.Artificial detection there are problems that many limitations and:(1) employee's operating environment is poor, and strong light operation is easy that employee is made to regard
Power is deteriorated, and employee may be made to suffer from occupational disease for a long time;(2) it is likely to find defect under the strong light due to wanting multi-angle, to member
The skills and experience requirement of work is very high, and employee is not easy skillfully to grasp workmanship;(3) it is fluctuated by personnel experience, job morale
Influences, the product defects such as variation, focus reduction are not easy whole interceptions, product quality are caused to fluctuate;(4) factory is to quality pipe
The degree of dependence of reason personnel is high, and operating personnel's mobility is big, there is training degree difference between different personnel and subjective mind is poor
It is different, cause detection accuracy and consistency that can not ensure;(5) due to the use of the judgement of eye recognition, the metering outfit of auxiliary is very
Hardly possible forms quantifiable quality standard.Digital image processing techniques are big with information content, the form of expression is intuitive, transmission storage
The advantages that facilitating, with the development of electronics, computer and the communication technology, the surface defects detection based on machine vision becomes can
Can, domestic and foreign scholars have carried out this extensive research, some achievements in research have been successfully applied to the products such as steel ball, rail
Surface defect inspection.
Detection algorithm will finally be applied to the on-line checking of light guide plate surface defect.Due to the making required precision of light guide plate
Relatively high, the defect of light guide plate is generally very small, needs to use high-resolution line-scan digital camera to detect guide-lighting board defect
Imaging, a width light guide plate image have nearly 500MB, this proposes the efficiency of defect on-line checking very high requirement.And light guide plate
Manufacturing enterprise generally requires the detection speed of every piece of light guide plate within 5 seconds, so detection algorithm is relatively high in addition to needing
Surface defect correct recognition rata, should also have very high operational efficiency.Using Curvelet transformation, non-lower sampling
Contourlet transformation, shearlet transformation, wavelet transformation stool multiscale analysis technology, algorithm operational efficiency, which cannot meet, to be wanted
It asks, and is difficult to realize in embedded systems;Some surface defect correct recognition ratas are relatively low, cannot meet required precision.
In addition, light guide plate defects detection algorithm can only detect the evenly distributed light guide plate of light guiding points at present, and unilateral side cannot be detected and entered
The non-uniform light guide plate of light, light guiding points.
Therefore, it is necessary to be improved to the prior art.
Invention content
The technical problem to be solved in the present invention is to provide a kind of, and the unilateral side incident type based on region adaptivity technology is led
Tabula rasa defect extracting method.
In order to solve the above technical problems, the present invention provides a kind of unilateral side incidence type light guide plate defect extracting method, it is special
Sign is, includes the following steps:
(1) light guide plate image F is obtained, step 2 is executed;
(2) step 3 is executed according to light guide plate image F, extraction light conducting plate body image H (x, y);
(3) the width M and height N for obtaining light conducting plate body image H (x, y), execute step 4;
(4) it uses sparse representation method to carry out defect to light conducting plate body image H (x, y) quickly to detect, obtains light guide plate
The SR values of ontology image H (x, y) execute step 5;
(5) if the SR values of light conducting plate body image H (x, y) are more than threshold value TH, therefore it is band defect image, thens follow the steps
Execute step 6;Conversely, as zero defect image, then judge light guide plate for certified products;
(6) noise jamming in mean filter removal light conducting plate body image H (x, y) is utilized, new images J (x, y) is obtained;
Execute step 7;
(7) greyscale transformation is carried out to new images J (x, y), obtains enhanced image K (x, y), execute step 8;
(8) OTSU Threshold segmentations are carried out to the image K (x, y) that step 7 obtains, light guiding points and background segment is come, are obtained
To foreground image and background image;Execute step 9;
(9) according to the density degree of light guiding points, image K (x, y) auto-partition detection that step 7 obtains is realized;Execute step
Rapid 10;
(10) tonal range for traversing all light guiding points area pixels of foreground image, utilizes formulaIt calculates every
The average value G of one light guiding points gray valueave, N in formulaiFor the sum of all pixels of i-th of light guiding points, GiIt is all for i-th of light guiding points
The gray scale summation of pixel;Execute step 11;
(11) maximum judge value G is setmaxWith minimum judge value GminIf Gave> Gmax, there are bright spots for light guide plate, directly sentence
Break as defective work;If Gave< Gmin, there are dim spots for light guide plate, are directly judged as defective work;If Gmin≤Gave≤Gmax, then
There is no bright dim spots, execute step 12;
(12) etching operation and expansive working are carried out to the image after step 9 subregion, obtain the image after Morphological scale-space,
Then step 13 is executed;
(13) all connected components of the image after the Morphological scale-space that analytical procedure 12 obtains, calculate each piece here
The area S of connected domain, setting decision content Smax;If there is Si> Smax, then light guide plate exist and weigh wounded or foreign matter, directly sentence
Break as defective work;If all Si≤Smax, then follow the steps 14;
(14) light conducting plate body image H (x, y) is split, the image g (x, y) after being divided;Execute step 15;
(15) all areas length L being partitioned into is traversedi(i=0,1,2,3...N), setting differentiate length standard LmaxIf
Li> Lmax, then there are scratch defects for light guide plate, are determined as defective work;If Li≤Lmax, then it is determined as certified products;
(16) to being determined as the light guide plate of defective work, extract all defect region of light guide plate, finally by minimum outside
Boundary's rectangle, defect area is showed, and calculates defect area mathematical feature.
As the improvement to the unilateral side incidence type light guide plate defect extracting method of the present invention, step (4) includes the following steps:
Input:The image block s that light conducting plate body image H (x, y) is divided into, the dictionary D' constructed, the atom of selection
Number k;
Output:Rarefaction representation coefficient X;
For S1 by identical method by image to be detected piecemeal, s is one of;Initial residual error r0=s, indexed set ∧=
Θ, J=Θ, iterations t=1;
S2 calculates residual error rtWith atom related coefficient u in dictionary D', therefrom k maximum value at searching will be with its coordinate pair
The index answered is added in indexed set J;
S3 calculates local restriction weighting coefficient ω=dist (s, D'), and wherein ω indicates similitude between residual error s and D',
Dist (s, D')=[dist (s, d1),...,dist(s,dm)]TIndicate residual error s and dictionary diIt is European between (i=1 ..., m)
Distance;Then k maximum correlation coefficient u is weighted, and related coefficient is met into condition 2 | u | >=| umax| atom be added to
Indexed set ∧, to update supported collection;
S4 carries out Signal approximation using least square method:Xt=arg min ‖ s-D'ΛxT-1‖2, calculate under new supported collection
Residual error rt=s-D'Λxt;
If S5 ‖ rt‖2< 10-1Or sparse coefficient number num (∧) >=6, then go to step S6;Otherwise weight is returned
Multiple step S2, until meeting stop condition;
S6 exports rarefaction representation coefficient X;
S7 goes out the l of coefficient X using sparse table0The ratio of norm and original image size as light conducting plate body image H (x,
Y) evaluation function of degree of rarefication:
SR=| | X | |0/mn。
Be further improved as to the unilateral side incidence type light guide plate defect extracting method of the present invention, step (14) include with
Lower step:
14.1, according to image H (x, y) size M*N, Fast Fourier Transform is carried out to light conducting plate body image H (x, y);
Execute step 14.2;
14.2, an ideal low-pass filter is generated;Execute step 14.3;
Wherein, D0Indicate the radius of passband, the calculation of D (u, v) i.e. the distance of point-to-point transmission;
14.3, convolution operation is carried out to light conducting plate body image H (x, y) in frequency domain with ideal low-pass filter;If
Arbitrary point f (x, y) in light conducting plate body image H (x, y), the point after convolution are g (x, y);Execute step 14.4;
14.4, fast fourier inverse transformation is carried out to the light conducting plate body image H (x, y) after step (14.3) convolution, it will
Defect transforms to spatial domain;Execute step 14.5;
14.5, using fixed threshold TH, image segmentation is carried out to the image obtained in step (14.4) using following formula,
Execute step 15;
G (x, y) is the image after segmentation in formula, and TH is segmentation threshold.
Be further improved as to the unilateral side incidence type light guide plate defect extracting method of the present invention, step (9) include with
Lower step:
According to light guiding points by dredging close direction, image K (x, y) is averagely divided into a region N (i=1,2 ..., N),
Each region area is S;According to the threshold value in step (8) as a result, obtaining NiRegion light guiding points quantity is Mi, by formulaObtain the density P in the regions Nii;Setting differentiates density value Qj(j=1,2,3...9), it is all for image K (x, y)
Region, if (Qj< Pi< Qj+1)U(Qj< Pi+1< Qj+1), then Ni+Ni+1, NiRegion and Ni+1Region merging technique.
It is further improved as to the unilateral side incidence type light guide plate defect extracting method of the present invention, step (16) includes:
Using the pixel of corresponding light conducting plate body image H (x, y) as random variable values f (x, y), the p of area-of-interest T
+ q rank squares areThen the center-of-mass coordinate of area-of-interest T is (x1,y1):
Barycenter is moved to the origin position of reference frame, obtains centre-to-centre spacing:
Similar, find out u00、u20、u02Deng then the length of the minimum enclosed rectangle of area-of-interest T and width:
Obtain the image containing area-of-interest T.
The technical advantage of the unilateral side incidence type light guide plate defect extracting method of the present invention is:
Patent of the present invention proposes the exploitation adaptive auto-partition algorithm of light guide plate, according to the dense degree of surface light-conductive hole,
Automatically different detection zones, and adjust automatically detection algorithm are divided, realizes the extraction of defect.On-line operation experimental result table
Bright, the operational efficiency height and accuracy rate of the algorithm are high, and stability and strong robustness can not only identify common defect, for
Uncommon tiny flaw also has relatively high detectability;
The present invention is adaptable to illumination variation, light guide plate Change of types;
The present invention need to only adjust several control parameters when producing installation, all automatic later to detect without manually guard;
Inventive algorithm is stablized, and the repair and maintenance of system is convenient for;
The present invention can also scratch defect to line and be detected.
Description of the drawings
The specific implementation mode of the present invention is described in further detail below in conjunction with the accompanying drawings.
Fig. 1 is the flow diagram of the unilateral side incidence type light guide plate defect extracting method of the present invention;
Fig. 2 is the flow diagram that light guide plate image carries out that defect quickly detects;
Fig. 3 is that step 1 of the present invention shoots obtained image;
Fig. 4 is the mean filter image that step 6 of the present invention obtains;
Fig. 5 is the setting contrast image that step 7 of the present invention obtains;
Fig. 6 is the light guide plate auto-partition image that step 9 of the present invention obtains;
Fig. 7 is the traversal gray-value image that step 10 of the present invention obtains;
Fig. 8 is image after the expansion that step 12 of the present invention obtains;
Fig. 9 is image after the corrosion that step 12 of the present invention obtains;
Figure 10 is the connection area image that step 13 of the present invention obtains;
Figure 11 is the image after the Fast Fourier Transform that step 14.1 of the present invention obtains;
Figure 12 is the image after the low-pass filtering containing scratch defects that step 14.3 of the present invention obtains;
Figure 13 is the testing result figure that light guide plate contains scratch defects, is defect area in center;
Figure 14 is the artwork that light guide plate contains fleck defect;
Figure 15 is the testing result figure that light guide plate contains fleck defect, is defect area in center;
Figure 16 is that light guide plate contains the artwork for weighing defect wounded;
Figure 17 is that light guide plate contains the testing result figure for weighing defect wounded, is defect area in center;
Figure 18 is the artwork that light guide plate contains scratch defects;
Figure 19 is the testing result figure that light guide plate contains scratch defects, is defect area in center.
Specific implementation mode
With reference to specific embodiment, the present invention is described further, but protection scope of the present invention is not limited in
This.
Embodiment 1, unilateral side incidence type light guide plate defect extracting method, as shown in figures 1-19,
(1) it uses the 16k lines of Dalsa companies to sweep camera, obtains the light guide plate image F of high accuracy grey scale;Execute step 2;
It is observed that various densitys, light guiding points not of uniform size, can make light guide plate uniformly light-emitting.The making precision of light guide plate
It is required that relatively high, the defect of light guide plate is generally very small, needs to use high-resolution linear array to detect guide-lighting board defect
Camera imaging.
(2) removal detection workbench marginal interference, extraction light conducting plate body part;Execute step 3;
First with Gaussian filter to light guide plate image F (X, Y) noise reduction process, it is smooth after image G (X, Y) it is as follows
G (X, Y)=H (X, Y, σ) × F (X, Y)
Wherein, Gaussian function(e is natural constant, e ≈ 2.7182), σ is Gauss
The standard deviation of filter function, controls smoothness, is compared by many experiments, and σ of the present invention is taken as 0.2.
Calculate the gradient magnitude T and deflection θ of image G (X, Y).
θ [X, Y]=arctan (GX(X, Y)/GY(X, Y))
Wherein, GXAnd GYThe respectively local derviation of X, Y.
GXAnd GYFollowing 2 × 2 first-order difference approximate calculation can be utilized:
Gx=[F (X+1, Y)-F (X, Y)+F (X+1, Y+1)-F (X, Y+1)]/2
Gy=[F (X, Y+1)-F (X, Y)+F (X+1, Y+1)-F (X+1, Y)]/2
Non-maxima suppression is carried out to gradient magnitude T (X, Y) using interpolation method (using spline interpolation here).Choose dual threshold
And image border is connected, high-low threshold value parameter is all artificial selected in traditional Canny algorithms, to enhance the applicability of algorithm
And sensitivity, the threshold value of this patent take adaptive threshold, search maximum pixel value in passing through and are denoted as L, build image border
Histogram of gradients, the accumulation for counting in the image after non-maxima suppression (NMS) sum of all pixels for not being 0 are denoted as Hist.
Assuming that gray value at C × Hist (0 < C < 1) is L, then high-low threshold value THH、THLIt calculates as follows
THH=L+1
THL=0.3*THH
Light-guide edge image Q (X, Y) is obtained by above-mentioned adaptive Canny edge detection algorithms, and is extracted and is led with this
Tabula rasa ontology image H (x, y), it is light conducting plate body figure that light guide plate image F (X, Y), which subtracts light-guide edge image Q (X, Y),
As H (x, y).
(3) the width M and height N of light conducting plate body image H (x, y) are obtained;Execute step 4;
(4) it uses sparse representation method to carry out defect to light conducting plate body image H (x, y) quickly to detect;Execute step 5;
The present invention proposes degree of rarefication evaluation function and goes out coefficient in dictionary D ' lower linear tables for characterizing sample to be detected
Degree of rarefication, the size that coefficient is gone out using sparse table judge that image to be detected is zero defect image or defect image, reach guide-lighting
The purpose of board defect detection.
Width light guide plate image (light conducting plate body image H (x, y)) about 500MB for camera acquisition is swept using 16K lines,
This proposes the efficiency of defect on-line checking very high requirement.And the defect rate of light guide plate is generally very low, if every leaded light
Plate all carries out detailed detection, and identifies defect type, is difficult to meet online detection requirements certainly.The present invention first designs
Only detection light guide plate whether there is the rapid detection method of defect, then be examined in detail to defective light guide plate image again
It surveys.
The purpose of sparse signal representation is exactly to indicate signal with atom as few as possible in given super complete dictionary,
The more succinct representation of signal can be obtained, to make more easily to obtain the information contained in signal, be more convenient into
One step is processed signal.It has been investigated that signal is after rarefaction representation, the precision after more sparse then signal reconstruction
It is higher, and rarefaction representation can be according to the suitable super complete dictionary of the adaptive selection of the own characteristic of signal.To signal
The purpose of rarefaction representation is exactly to find a self-adapting dictionary to make the expression of signal most sparse.The characteristics of according to rarefaction representation,
The present invention whether there is defect using rarefaction representation quickly to detect light guide plate.
Flawless light guide plate image is subjected to piecemeal, and using image block as atom, extracts characteristics of image, constructed complete
Standby redundant dictionary.If tile size is K × K, the alternative atom that construction obtains zero defect light guide plate image is D, alternative atom
Each atom (each row) corresponds to an image block in the D of library, is K2× 1 column vector, is denoted as
The alternative atom is redundancy, there is prodigious correlation between row atom, needs further to be trained it preferably, make
In alternative atom D each row atom can by remaining row atom Linearly Representation in alternative atom D, and table go out be
Number is sparse.The dictionary training problem can be modeled as:
It is above-mentionedThe solution of norm is np hard problem, can will be above-mentioned according to compressive sensing theory
l0Minimization problem is converted to l1Least norm problem:
Solution Z={ z1, z2 ..., zN } ∈ R of above formula are sought using orthogonal matching pursuit algorithm (OMP)N×N, count institute in Z
There is the row coordinate for being not all zero, and choose row atom of the row atom as D' in the corresponding D of these row coordinates, that is, completes word
Allusion quotation it is preferred.
One flawless image, which can be used, completes the sparse reconstruct of the preferred D' of dictionary, and for defect image, then cannot
Using the completion preferred D' of dictionary come sparse reconstruct.For image to be detected I, a demand goes out it and completes the preferred D' of dictionary
Whether the coefficient that lower linear table goes out, the degree of rarefication that coefficient is gone out according to table can determine whether the image containing defective.
Research finds that orthogonal matching pursuit algorithm (OMP) differs when for rarefaction representation and surely obtains its most sparse solution.
Therefore, the present invention is as follows using improved ROMP algorithms solution rarefaction representation coefficient X:
Input:Image block s to be reconstructed, dictionary D', the atomicity k of selection.
Output:Rarefaction representation coefficient X.
Step 1 initializes.By light conducting plate body image H (x, y) piecemeal, s is one of.Initial residual error r0=s is (initial
Residual error is image block s), indexed set ∧=Θ, J=Θ, iterations t=1;
The 1st subatom of step 2 selects.Calculate residual error rtWith atom related coefficient u in dictionary D', therefrom k is a most at searching
Index corresponding with its coordinate is added in indexed set J big value;
The 2nd subatom of step 3 selects (regularization).Calculate local restriction weighting coefficient ω=dist (s, D'), wherein ω
Indicate similitude between residual error s and D', dist (s, D')=[dist (s, d1),...,dist(s,dm)]TIndicate residual error s and word
Allusion quotation diEuclidean distance between (i=1 ..., m).Then k maximum correlation coefficient u is weighted, and related coefficient is met
Condition 2u | >=| umax| atom be added to indexed set ∧, to update supported collection;
Step 4 updates residual error.Signal approximation is carried out using least square method:xt=arg min ‖ s-D'ΛxT-1‖2, calculate
Residual error r under new supported collectiont=s-D'Λxt;
Step 5 stop condition.If ‖ rt‖2< 10-1Or sparse coefficient number num (∧) >=6, then it gos to step
(6);Otherwise it returns and repeats step (2), until meeting stop condition;
Step 6 exports rarefaction representation coefficient X.Improved ROMP algorithms by related coefficient carry out local restriction weighting,
The regularization for improving atom selection, to obtain accurate X.
Finally, the l of coefficient X is gone out using sparse table0The ratio (being known as sparse rate) of norm and original image size, which is used as, leads
The evaluation function of tabula rasa ontology image H (x, y) degree of rarefication (SR):
SR=| | X | |0/mn
(5) judge that light guide plate whether there is defect;Execute step 6;
If the SR values of light conducting plate body image H (x, y) are more than some threshold value TH, then it represents that the Linearly Representation of image to be detected
Coefficient X is not sparse enough, and image to be detected cannot be by dictionary D' sparse reconstruct well, therefore is band defect image, then executes step
It is rapid to execute step 6;Conversely, as zero defect image, then judge light guide plate for certified products.
(6) noise jamming in mean filter removal light conducting plate body image H (x, y) is utilized, new images J (x, y) is obtained;
Execute step 7;
Due to interference such as electrical noises, inevitably there is noise in light guide plate image, and the present invention is avoided using mean filter
Picture noise impacts detection.
For each pixel P in given image H (x, y), its neighborhood S is taken.If containing M pixel in S, it is taken to add
Weight average value is as the gray value after processing at gained image pixel P.It is flat with the weighting of each pixel grey scale in a neighborhood of pixels
Mean value is exactly neighborhood weighted average technology come the method for replacing the original gray scale of the pixel.The shape of general neighborhood S is square,
Rectangle or cross etc..If noise n (i, j) is additive noise, and each point is orthogonal, is desired for 0, variance σ2, g
(x, y) is that the image by noise pollution, noise-containing image H (x, y) be not as follows after mean filter:
Therefore, the noise variance of filtered image is as follows:
(7) it since the color and background distinction of light guiding points is not strong, needs to carry out greyscale transformation to new images J (x, y), it will
Contrast expands, and improves the accuracy of detection, obtains enhanced image K (x, y);Execute step 8;
To each point is handled with following formula in new images J (x, y), image K (x, y) is obtained:
K (x, y)=a × j (x, y)+b
A and b is the fixed value obtained through experiment, and a takes 1.3, b to take 10.
(8) OTSU Threshold segmentations are carried out to the image K (x, y) that step 7 obtains, light guiding points and background segment is come;It holds
Row step 9;
OTSU algorithms are also referred to as maximum kind differences method, by the gamma characteristic of image, divide the image into background and foreground two parts.
Inter-class variance between background and foreground is bigger, illustrates that the two-part difference for constituting image is bigger, foreground mistake is divided into when part
Background or part background mistake, which are divided into foreground, all can cause two parts difference to become smaller.Therefore, make the maximum segmentation meaning of inter-class variance
Misclassification probability minimum.If gray level image gray level is L, then tonal range is [0, L-1], and image is calculated using OTSU algorithms
Optimal threshold be:
T=Max [w0(t)×(u0(t)-u)2+w1(t)×(u1(t)-u)2]
Wherein:When the threshold value of segmentation is t, w0For background ratio, u0For background mean value, w1For foreground ratio, u1For foreground
Mean value, u are the mean value of light guide plate image.The above maximum t of transition formula evaluation as divides the optimal threshold of image.
It is divided into background image and foreground image, light guiding points to concentrate on foreground picture image K (x, y) using optimal threshold
Picture.
(9) according to the density degree of light guiding points, image K (x, y) auto-partition detection that step 7 obtains is realized;Execute step
Rapid 10;
According to the structure of unilateral side incident type LGP it is found that by various densitys and light guiding points not of uniform size, make light guide plate
Uniformly light-emitting.The light guiding points of unilateral side incidence type light guide plate are unevenly distributed, remoter apart from light source side, and light guiding points distribution is closeer
Collection.In order to solve the problems, such as flase drop that different densities are brought, multidomain treat-ment is carried out.
According to light guiding points by dredging close direction, image K (x, y) is averagely divided into a region N (i=1,2 ..., N),
Each region area is S.According to the threshold value in step (8) as a result, obtaining NiRegion light guiding points quantity is Mi, by formulaObtain NiThe density P in regioni.Setting differentiates density value Qj(j=1,2,3...9), for all regions, if
(Qj< Pi< Qj+1)U(Qj< Pi+1< Qj+1), then Ni+Ni+1, the regions Ni and Ni+1 region merging techniques.
(10) the light guiding points region extracted according to step 8 traverses the gray scale model of all light guiding points area pixels of foreground image
It encloses, utilizes formulaCalculate the average value G of each light guiding points gray valueave, N in formulaiFor the picture of i-th of light guiding points
Plain sum, GiFor the gray scale summation of i-th of light guiding points all pixels;Execute step 11;
(11) maximum judge value G is setmaxWith minimum judge value GminIf Gave> Gmax, there are bright spots for light guide plate, directly sentence
Break as defective work;If Gave< Gmin, there are dim spots for light guide plate, are directly judged as defective work;If Gmin≤Gave≤Gmax, then
There is no bright dim spots, execute step 12.There are bright spots and dim spot to be all determined as defective work for light guide plate.
(12) since high density area light guiding points mutual distance is close, even some are led in the image after step 9 subregion
Luminous point has had been attached to together, and detection is caused greatly to interfere, so the region needs to carry out etching operation.
Patent of the present invention is corroded using a circular configuration element, if X, which is pending image, (passes through step 9 subregion
Image afterwards), B1For structural element, R1For treated image.Structural element B1The mathematic(al) representation that image X is corroded
It is as follows:
Wherein, Θ represents the operator of corrosion, and x refers to the translational movement of set.
The basic process of erosion operation is:The moving structure element B in pending image X2If being moved to and structural element
B2When identical subgraph, just by the subgraph with structural element B2That corresponding element marking of origin come out,
And so on, make structural element B2It is moved in pending image X, the set of finally all labeled pixels is exactly
By structural element B2It is after corrosion as a result, formed corrosion after image.
Since low density area light guiding points mutual distance is bigger, weigh wounded, foreign matter the defects of be easy missing inspection, therefore step 9 is divided
Image behind area carries out expansive working.
If Y is pending image (image after step 9 subregion), B2For structural element, BvFor B2Reflection, R2For
Treated image, then the mathematic(al) representation of expansive working is as follows:
X is the translational movement of set in formula,Represent expansive working symbol.
Expansive working is first by structural element B2Reflection operation is carried out centered on origin, obtains Bv, then in pending figure
As mobile B in Yv, work as BvIt, then will B at this time when at least one non-zero crossing of pending imagevOrigin be marked, directly
Completely open pending image to processing, the set of label pixel at this time and former pending image be exactly image after expanding (i.e.
For the image after Morphological scale-space).
Then step 13 is executed;
(13) judge that whether there is or not weigh wounded or foreign matter light guide plate;All connections of image after 12 Morphological scale-space of analytical procedure
Characteristic of field calculates the area S of each piece of connected domain herei。
Set decision content Smax.If Si> Smax, then light guide plate exist and weigh wounded or foreign matter, be directly judged as unqualified
Product; Si≤Smax, then follow the steps 14.
(14) scratch defects differentiate.
14.1, Fast Fourier Transform is carried out to light conducting plate body image H (x, y).Execute step 14.2;
It can be obtained by step (3), image H (x, y) size is M*N, then image H (x, y) is the M*N in period discrete
Signal, Fourier transform type are 2-DFT, and expression formula is as follows:
Wherein u=0,1,2 ..., M-1;V=0,1,2 ..., N-1, and u, v are frequency values.X, y are the frequency in spatial domain
Rate value, M and N are the size of digital picture.
14.2, an ideal low-pass filter is generated.Execute step 14.3;
Inside frequency domain, high frequency section represents the details of image, texture information;Low frequency part represents the profile of image
Information.If the image fine to a width uses low-pass filter, only low frequency signal can just pass through, then filtered knot
Fruit just only remaining profile.So being operated through low-frequency filter, tiny light guiding points can be cut in light guide plate image, so as to
Prominent scratch defects.
Wherein, D0Indicate the radius of passband.The calculation of D (u, v) i.e. the distance of point-to-point transmission;
Wherein the size of image is M*N;
14.3, convolution operation is carried out to light conducting plate body image H (x, y) in frequency domain with above-mentioned filter.Execute step
14.4;
If arbitrary point f (x, y) in light conducting plate body image H (x, y), the point after convolution is g (x, y);
Defect is transformed to spatial domain by 14.4, fast fourier inverse transformation.Execute step 14.5;
Wherein u=0,1,2 ..., M-1;V=0,1,2 ..., N-1, and u, v are frequency values.X, y are the frequency in spatial domain
Rate value, M and N are the size of digital picture.
14.5, it is finally operated using image in following formula pair 14.4 using fixed segmentation threshold TH, carries out image
Segmentation executes step 15;
G (x, y) is the image after segmentation in formula, and TH is segmentation threshold.Proof of algorithm repeatedly is carried out to the sample of acquisition,
The optimal threshold TH=38 for being directed to this experiment condition is determined.
(15) all areas length L of the image g (x, y) after traversal segmentationi(i=0,1,2,3...N), setting differentiate length
The quasi- L of scalemaxIf Li> Lmax, then there are scratch defects for light guide plate, are determined as defective work;If Li≤Lmax, then it is judged to closing
Lattice product.
(16) through above step, all defect region of light guide plate is can extract, finally by minimum extraneous rectangle, by defect
Region shows, and calculates defect area mathematical feature.
To being determined as the light guide plate of defective work, become using the pixel of corresponding light conducting plate body image H (x, y) as random
The p+q rank squares of magnitude f (x, y), area-of-interest T (i.e. ROI) areThe then matter of target area
Heart coordinate is (x1,y1):
The barycenter of target is moved to the origin position of reference frame, obtains centre-to-centre spacing:
Similar, find out u00、u20、u02Deng.The so length of the minimum enclosed rectangle of ROI and width:
Obtain the image containing area-of-interest T.
Experiment one:
(1) camera is swept using line, obtains gray level image as shown in Figure 3;
(2) 7*7 is carried out to mean filter to gray level image, obtains mean filter image as shown in FIG. 6 (step 1 and step
The step of other steps between rapid 2 use above-described embodiment and formula);
(3) greyscale transformation is carried out according to formula 1 to mean filter image, contrast is further expanded into Pout(x,y)
=a × Pin(x, y)+b, a takes 1.3, b to take 10 (experiment obtains) in formula;
(4) according to the density degree of light guiding points, auto-partition, such as Fig. 7 are realized to light guide plate image;
If N=10, Q=3, image is divided into 4 regions automatically, and (4 regions are obtained by testing, between step 4 and step 5
Other steps use the step of above-described embodiment and formula)
(5) automatic threshold segmentation is carried out to the image after subregion, segmentation threshold t is determined using following formula;
T=Max [w0(t)×(u0(t)-u)2+w1(t)×(u1(t)-u)2] (experiment obtains)
(6) high-density region uses etching operation, structural element B1Using 2 pixel radius circulars, using following formula into
Row corrosion, obtains Fig. 9 (experiment obtains);
(7) density regions use expansive working, structural element B2Using 3 pixel radius circulars, using following formula into
Row expansion, obtains Fig. 8 (experiment obtains);
(8) connected component is analyzed, all areas area S is calculatediIf Smax=90, if Si> Smax, then to weigh wounded or
Foreign matter defect;
(9) according to following formula, Fast Fourier Transform is done to image;
(10) an ideal low-pass filter H (u, v) is generated;
(11) defect is transformed to spatial domain by fast fourier inverse transformation;
(12) fixed threshold TH=38 is partitioned into scratch defects using following formula;
(13) minimum enclosed rectangle M is made to the defect area of extractionpq, defect is shown;
Finally, it should also be noted that it is listed above be only the present invention several specific embodiments.Obviously, this hair
Bright to be not limited to above example, acceptable there are many deformations.Those skilled in the art can be from present disclosure
All deformations for directly exporting or associating, are considered as protection scope of the present invention.
Claims (5)
1. unilateral side incidence type light guide plate defect extracting method, which is characterized in that include the following steps:
(1) light guide plate image F is obtained, step 2 is executed;
(2) step 3 is executed according to light guide plate image F, extraction light conducting plate body image H (x, y);
(3) the width M and height N for obtaining light conducting plate body image H (x, y), execute step 4;
(4) it uses sparse representation method to carry out defect to light conducting plate body image H (x, y) quickly to detect, obtains light conducting plate body
The SR values of image H (x, y) execute step 5;
(5) if the SR values of light conducting plate body image H (x, y) are more than threshold value TH, therefore it is band defect image, thens follow the steps execution
Step 6;Conversely, as zero defect image, then judge light guide plate for certified products;
(6) noise jamming in mean filter removal light conducting plate body image H (x, y) is utilized, new images J (x, y) is obtained;It executes
Step 7;
(7) greyscale transformation is carried out to new images J (x, y), obtains enhanced image K (x, y), execute step 8;
(8) OTSU Threshold segmentations are carried out to the image K (x, y) that step 7 obtains, light guiding points and background segment is come, before obtaining
Scape image and background image;Execute step 9;
(9) according to the density degree of light guiding points, image K (x, y) auto-partition detection that step 7 obtains is realized;Execute step 10;
(10) tonal range for traversing all light guiding points area pixels of foreground image, utilizes formulaEach is calculated to lead
The average value G of luminous point gray valueave, N in formulaiFor the sum of all pixels of i-th of light guiding points, GiFor i-th light guiding points all pixels
Gray scale summation;Execute step 11;
(11) maximum judge value G is setmaxWith minimum judge value GminIf Gave> Gmax, there are bright spots for light guide plate, are directly judged as
Defective work;If Gave< Gmin, there are dim spots for light guide plate, are directly judged as defective work;If Gmin≤Gave≤Gmax, then do not deposit
In bright dim spot, step 12 is executed;
(12) etching operation and expansive working are carried out to the image after step 9 subregion, obtains the image after Morphological scale-space, then
Execute step 13;
(13) all connected components of the image after the Morphological scale-space that analytical procedure 12 obtains calculate each piece of connection here
The area S in domain, setting decision content Smax;If there is Si> Smax, then light guide plate exist and weigh wounded or foreign matter, be directly judged as not
Certified products;If all Si≤Smax, then follow the steps 14;
(14) light conducting plate body image H (x, y) is split, the image g (x, y) after being divided;Execute step 15;
(15) all areas length L being partitioned into is traversedi(i=0,1,2,3...N), setting differentiate length standard LmaxIf Li>
Lmax, then there are scratch defects for light guide plate, are determined as defective work;If Li≤Lmax, then it is determined as certified products;
(16) to being determined as the light guide plate of defective work, all defect region of light guide plate is extracted, finally by minimum extraneous square
Shape shows defect area, and calculates defect area mathematical feature.
2. unilateral side incidence type light guide plate defect extracting method according to claim 1, which is characterized in that step (4) is wrapped
Include following steps:
Input:The image block s that light conducting plate body image H (x, y) is divided into, the dictionary D' constructed, the atomicity k of selection;
Output:Rarefaction representation coefficient X;
For S1 by identical method by image to be detected piecemeal, s is one of;Initial residual error r0=s, indexed set ∧=Θ, J=
Θ, iterations t=1;
S2 calculates residual error rtWith atom related coefficient u in dictionary D', therefrom k maximum value at searching will rope corresponding with its coordinate
Draw and is added in indexed set J;
S3 calculates local restriction weighting coefficient ω=dist (s, D'), and wherein ω indicates similitude between residual error s and D', dist
(s, D')=[dist (s, d1),...,dist(s,dm)]TIndicate residual error s and dictionary diEuclidean distance between (i=1 ..., m);
Then k maximum correlation coefficient u is weighted, and related coefficient is met into condition 2 | u | >=| umax| atom be added to index
Collect ∧, to update supported collection;
S4 carries out Signal approximation using least square method:Xt=arg min ‖ s-D'ΛxT-1‖2, calculate the residual error r under new supported collectiont
=s-D'Λxt;
If S5 ‖ rt‖2< 10-1Or sparse coefficient number num (∧) >=6, then go to step S6;Otherwise it returns and repeats step
S2, until meeting stop condition;
S6 exports rarefaction representation coefficient X;
S7 goes out the l of coefficient X using sparse table0The ratio of norm and original image size is dilute as light conducting plate body image H (x, y)
Dredge the evaluation function of degree:
SR=| | X | |0/mn。
3. unilateral side incidence type light guide plate defect extracting method according to claim 2, which is characterized in that step (14) is wrapped
Include following steps:
14.1, according to image H (x, y) size M*N, Fast Fourier Transform is carried out to light conducting plate body image H (x, y);It executes
Step 14.2;
14.2, an ideal low-pass filter is generated;Execute step 14.3;
Wherein, D0Indicate the radius of passband, the calculation of D (u, v) i.e. the distance of point-to-point transmission;
14.3, convolution operation is carried out to light conducting plate body image H (x, y) in frequency domain with ideal low-pass filter;If guide-lighting
Arbitrary point f (x, y) in plate ontology image H (x, y), the point after convolution are g (x, y);Execute step 14.4;
14.4, fast fourier inverse transformation is carried out to the light conducting plate body image H (x, y) after step (14.3) convolution, by defect
Transform to spatial domain;Execute step 14.5;
14.5, using fixed threshold TH, image segmentation is carried out to the image obtained in step (14.4) using following formula, is executed
Step 15;
G (x, y) is the image after segmentation in formula, and TH is segmentation threshold.
4. unilateral side incidence type light guide plate defect extracting method according to claim 3, which is characterized in that step (9) is wrapped
Include following steps:
According to light guiding points by dredging close direction, image K (x, y) is averagely divided into a region N (i=1,2 ..., N), each
Region area is S;According to the threshold value in step (8) as a result, obtaining NiRegion light guiding points quantity is Mi, by formulaObtain the density P in the regions Nii;Setting differentiates density value Qj(j=1,2,3...9), it is all for image K (x, y)
Region, if (Qj< Pi< Qj+1)U(Qj< Pi+1< Qj+1), then Ni+Ni+1, NiRegion and Ni+1Region merging technique.
5. unilateral side incidence type light guide plate defect extracting method according to claim 4, which is characterized in that step (16) is wrapped
It includes:
Using the pixel of corresponding light conducting plate body image H (x, y) as random variable values f (x, y), the p+q ranks of area-of-interest T
Square isThen the center-of-mass coordinate of area-of-interest T is (x1,y1):
Barycenter is moved to the origin position of reference frame, obtains centre-to-centre spacing:
Similar, find out u00、u20、u02Deng then the length of the minimum enclosed rectangle of area-of-interest T and width:
Obtain the image containing area-of-interest T.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810025157.8A CN108346141B (en) | 2018-01-11 | 2018-01-11 | Method for extracting defects of single-side light-entering type light guide plate |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810025157.8A CN108346141B (en) | 2018-01-11 | 2018-01-11 | Method for extracting defects of single-side light-entering type light guide plate |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108346141A true CN108346141A (en) | 2018-07-31 |
CN108346141B CN108346141B (en) | 2021-10-22 |
Family
ID=62960478
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810025157.8A Active CN108346141B (en) | 2018-01-11 | 2018-01-11 | Method for extracting defects of single-side light-entering type light guide plate |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108346141B (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109272525A (en) * | 2018-09-17 | 2019-01-25 | 浙江理工大学 | A kind of light guide plate white-spot defects detection method |
CN109886914A (en) * | 2018-12-19 | 2019-06-14 | 刘咏晨 | Paper sheet defect detection method based on local luminance invariance priori |
CN110095474A (en) * | 2019-04-12 | 2019-08-06 | 浙江大学台州研究院 | A kind of bandage screened based on machine vision and flaw is around winding apparatus and bandage winding method |
CN110298825A (en) * | 2019-06-17 | 2019-10-01 | 杭州舜浩科技有限公司 | Light guide plate black spot defect detection method |
CN110530894A (en) * | 2019-06-14 | 2019-12-03 | 杭州舜浩科技有限公司 | Light guide plate bright spot defect detection method |
CN113340902A (en) * | 2021-05-19 | 2021-09-03 | 沈阳风驰软件股份有限公司 | Method and device for detecting surface defects of catalyst carrier, and electronic device |
CN113610852A (en) * | 2021-10-10 | 2021-11-05 | 江苏祥顺布业有限公司 | Yarn drafting quality monitoring method based on image processing |
CN113706498A (en) * | 2021-08-25 | 2021-11-26 | 广东奥普特科技股份有限公司 | Visual detection device for lithium battery pole piece |
CN114240926A (en) * | 2021-12-28 | 2022-03-25 | 湖南云箭智能科技有限公司 | Board card defect type identification method, device and equipment and readable storage medium |
CN114882026A (en) * | 2022-07-08 | 2022-08-09 | 山东科尔自动化仪表股份有限公司 | Sensor shell defect detection method based on artificial intelligence |
CN114897906A (en) * | 2022-07-14 | 2022-08-12 | 南通华烨塑料工业有限公司 | Modified plastic appearance defect control method based on computer vision |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TW200844429A (en) * | 2007-05-15 | 2008-11-16 | Chi-Hao Yeh | An automatic optical inspection approach for detecting and classifying the surface defects on coating brightness enhancement film |
CN102129573A (en) * | 2011-03-10 | 2011-07-20 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image segmentation method based on dictionary learning and sparse representation |
CN102937599A (en) * | 2012-10-25 | 2013-02-20 | 中国科学院自动化研究所 | Non-destructive testing systems and method used for detecting a metal-containing object through X-ray detection |
CN103499343A (en) * | 2013-09-25 | 2014-01-08 | 武汉海达数云技术有限公司 | Automatic classifying and identifying method for road line point cloud based on three-dimensional laser reflection strength |
CN104574274A (en) * | 2013-10-18 | 2015-04-29 | 财团法人资讯工业策进会 | Image processing method and system applying same |
CN104732215A (en) * | 2015-03-25 | 2015-06-24 | 广西大学 | Remote-sensing image coastline extracting method based on information vector machine |
CN104931233A (en) * | 2015-07-03 | 2015-09-23 | 东莞市彤光电子科技有限公司 | Light guide plate detection apparatus |
CN105069771A (en) * | 2015-05-22 | 2015-11-18 | 同济大学 | Active-contour-based SAR image after-calamity water body information extraction method |
CN105787955A (en) * | 2016-03-29 | 2016-07-20 | 河南森源电气股份有限公司 | Sparse segmentation method and device of strip steel defect |
CN105913419A (en) * | 2016-04-07 | 2016-08-31 | 南京汇川图像视觉技术有限公司 | TFT-LCD mura defect detection method based on ICA learning and multichannel fusion |
CN106650770A (en) * | 2016-09-29 | 2017-05-10 | 南京大学 | Mura defect detection method based on sample learning and human visual characteristics |
CN106733701A (en) * | 2016-12-23 | 2017-05-31 | 郑州轻工业学院 | Potato multi-stage combination dysnusia detecting system and detection method |
-
2018
- 2018-01-11 CN CN201810025157.8A patent/CN108346141B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TW200844429A (en) * | 2007-05-15 | 2008-11-16 | Chi-Hao Yeh | An automatic optical inspection approach for detecting and classifying the surface defects on coating brightness enhancement film |
CN102129573A (en) * | 2011-03-10 | 2011-07-20 | 西安电子科技大学 | SAR (Synthetic Aperture Radar) image segmentation method based on dictionary learning and sparse representation |
CN102937599A (en) * | 2012-10-25 | 2013-02-20 | 中国科学院自动化研究所 | Non-destructive testing systems and method used for detecting a metal-containing object through X-ray detection |
CN103499343A (en) * | 2013-09-25 | 2014-01-08 | 武汉海达数云技术有限公司 | Automatic classifying and identifying method for road line point cloud based on three-dimensional laser reflection strength |
CN104574274A (en) * | 2013-10-18 | 2015-04-29 | 财团法人资讯工业策进会 | Image processing method and system applying same |
CN104732215A (en) * | 2015-03-25 | 2015-06-24 | 广西大学 | Remote-sensing image coastline extracting method based on information vector machine |
CN105069771A (en) * | 2015-05-22 | 2015-11-18 | 同济大学 | Active-contour-based SAR image after-calamity water body information extraction method |
CN104931233A (en) * | 2015-07-03 | 2015-09-23 | 东莞市彤光电子科技有限公司 | Light guide plate detection apparatus |
CN105787955A (en) * | 2016-03-29 | 2016-07-20 | 河南森源电气股份有限公司 | Sparse segmentation method and device of strip steel defect |
CN105913419A (en) * | 2016-04-07 | 2016-08-31 | 南京汇川图像视觉技术有限公司 | TFT-LCD mura defect detection method based on ICA learning and multichannel fusion |
CN106650770A (en) * | 2016-09-29 | 2017-05-10 | 南京大学 | Mura defect detection method based on sample learning and human visual characteristics |
CN106733701A (en) * | 2016-12-23 | 2017-05-31 | 郑州轻工业学院 | Potato multi-stage combination dysnusia detecting system and detection method |
Non-Patent Citations (1)
Title |
---|
李强国: "" TFT-LCD显示屏Mura缺陷自动光学检测算法研究"", 《中国优秀硕士学位论文全文数据库(电子期刊)信息科技辑》 * |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109272525A (en) * | 2018-09-17 | 2019-01-25 | 浙江理工大学 | A kind of light guide plate white-spot defects detection method |
CN109886914A (en) * | 2018-12-19 | 2019-06-14 | 刘咏晨 | Paper sheet defect detection method based on local luminance invariance priori |
CN109886914B (en) * | 2018-12-19 | 2020-01-17 | 浙江企银印务科技有限公司 | Paper defect detection method based on local brightness invariance prior |
CN110095474A (en) * | 2019-04-12 | 2019-08-06 | 浙江大学台州研究院 | A kind of bandage screened based on machine vision and flaw is around winding apparatus and bandage winding method |
CN110095474B (en) * | 2019-04-12 | 2021-11-16 | 浙江大学台州研究院 | Bandage winding device and method based on machine vision and defect screening |
CN110530894B (en) * | 2019-06-14 | 2022-05-13 | 杭州舜浩科技有限公司 | Light guide plate bright spot defect detection method |
CN110530894A (en) * | 2019-06-14 | 2019-12-03 | 杭州舜浩科技有限公司 | Light guide plate bright spot defect detection method |
CN110298825A (en) * | 2019-06-17 | 2019-10-01 | 杭州舜浩科技有限公司 | Light guide plate black spot defect detection method |
CN110298825B (en) * | 2019-06-17 | 2021-07-02 | 杭州舜浩科技有限公司 | Method for detecting black spot defect of light guide plate |
CN113340902A (en) * | 2021-05-19 | 2021-09-03 | 沈阳风驰软件股份有限公司 | Method and device for detecting surface defects of catalyst carrier, and electronic device |
CN113706498A (en) * | 2021-08-25 | 2021-11-26 | 广东奥普特科技股份有限公司 | Visual detection device for lithium battery pole piece |
CN113610852A (en) * | 2021-10-10 | 2021-11-05 | 江苏祥顺布业有限公司 | Yarn drafting quality monitoring method based on image processing |
CN113610852B (en) * | 2021-10-10 | 2021-12-10 | 江苏祥顺布业有限公司 | Yarn drafting quality monitoring method based on image processing |
CN114240926A (en) * | 2021-12-28 | 2022-03-25 | 湖南云箭智能科技有限公司 | Board card defect type identification method, device and equipment and readable storage medium |
CN114240926B (en) * | 2021-12-28 | 2022-12-13 | 湖南云箭智能科技有限公司 | Board card defect type identification method, device and equipment and readable storage medium |
CN114882026A (en) * | 2022-07-08 | 2022-08-09 | 山东科尔自动化仪表股份有限公司 | Sensor shell defect detection method based on artificial intelligence |
CN114882026B (en) * | 2022-07-08 | 2022-09-02 | 山东科尔自动化仪表股份有限公司 | Sensor shell defect detection method based on artificial intelligence |
CN114897906A (en) * | 2022-07-14 | 2022-08-12 | 南通华烨塑料工业有限公司 | Modified plastic appearance defect control method based on computer vision |
Also Published As
Publication number | Publication date |
---|---|
CN108346141B (en) | 2021-10-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108346141A (en) | Unilateral side incidence type light guide plate defect extracting method | |
Molina et al. | On the detection of defects on specular car body surfaces | |
DE112020004391B4 (en) | Systems and methods for surface modeling using polarization features | |
CN109635806B (en) | Ammeter value identification method based on residual error network | |
CN109934811A (en) | A kind of optical element surface defect inspection method based on deep learning | |
CN108416766A (en) | Bilateral incidence type light guide plate defective vision detection method | |
Ming et al. | A comprehensive review of defect detection in 3C glass components | |
CN104112269B (en) | A kind of solar battery laser groove parameter detection method and system based on machine vision | |
CN109087274A (en) | Electronic device defect inspection method and device based on multidimensional fusion and semantic segmentation | |
CN105069790A (en) | Rapid imaging detection method for gear appearance defect | |
CN104764712A (en) | Method for detecting quality of inner wall of via hole of PCB | |
CN105654121A (en) | Complex jacquard fabric defect detection method based on deep learning | |
Medina et al. | Automated visual classification of frequent defects in flat steel coils | |
CN109801286B (en) | Surface defect detection method for LCD light guide plate | |
CN108830851B (en) | LCD rough spot defect detection method | |
Li et al. | A novel digital method for weave pattern recognition based on photometric differential analysis | |
Fu et al. | Medicine glass bottle defect detection based on machine vision | |
Peng et al. | Automated product boundary defect detection based on image moment feature anomaly | |
Dominguez-Nicolas et al. | Indentation image analysis for Vickers hardness testing | |
CN108986088A (en) | A kind of image based on MATLAB automatically extracts optimization method and equipment | |
Jing et al. | Detection of bobbin yarn surface defects by visual saliency analysis | |
CN109711253A (en) | Ammeter technique for partitioning based on convolutional neural networks and Recognition with Recurrent Neural Network | |
Lin et al. | Defect detection system for optical element surface based on machine vision | |
CN110530894B (en) | Light guide plate bright spot defect detection method | |
Liu et al. | Automated rivet detection in the EOL image for aircraft lap joints inspection |
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 | ||
TA01 | Transfer of patent application right | ||
TA01 | Transfer of patent application right |
Effective date of registration: 20210816 Address after: 311215 room 806-6, building 1, No. 371, Xingxing Road, Xiaoshan District, Hangzhou City, Zhejiang Province Applicant after: Hangzhou Hengdian Technology Co.,Ltd. Address before: 310018, No. 2, No. 5, Xiasha Higher Education Park, Hangzhou, Zhejiang Applicant before: Zhejiang University of Technology |
|
GR01 | Patent grant | ||
GR01 | Patent grant |